RAM shortages are prompting some unheard-of moves on the part of some companies, like Transcend telling its customers that all deliveries were delayed, and Phison's CEO warning that smaller consumer electronics players might be headed for closure. Now, Silicon Power (SP) has added a notification to the header of its online U.S. store stating that "a full refund of the original purchase price will be issued if there is a shortage of replacement products." The new measure hints that, in some cases, SP is unable to source replacement parts or that the cost of replacement parts is now so high as to be unfeasible.

The latter is highlighted by the recent adventures of an Australian enthusiast's struggle to return his RAM kit for a warranty replacement, owing to the fact that the kit has quadrupled in price since he bought it. Retailers try their best to protect themselves from scarce availability and constantly fluctuating prices, all while consumers have to resort to paying highly inflated prices for what were once affordable staple items.

SP's statement is conditional and seems to indicate that in the case of an RMA, the SP store will make an attempt at replacing the item, but also prepares buyers for the possibility that they won't get a replacement unit. This is almost certainly due to the fact that flash chips have multiplied in price and become absurdly scarce, particularly for smaller companies.

As for its UK web-store, Silicon Power doesn't have a specific notice, and its warranty policy simply states that "due to ongoing technology or in case of discontinuity of any product, SP will be unable to repair the product, and therefore replace the product with a comparable one."

We took the opportunity to do a cursory browse of major memory and SSD manufacturers' RMA policies, and at the present time, we found that at least on paper, almost every major player says it will honor a replacement. Having said that, as the crises tighten, these policies might well change, and sooner rather than later. Note that our findings are consumer advice and not legal analysis:

• Corsair - Replacement: "If your Corsair product is not operating in accordance with its published technical specifications, the product will be replaced at Corsair’s expense."
  
• G.Skill - Replacement, conditional: "Any replacements provided will be with the same product model. This warranty obligation is conditioned upon the hardware being returned to the original place of purchase, or another place as directed by G.SKILL [...]."
  
• Kingston - Replacement: "Kingston will, at its option, either repair or replace any part of its products that prove defective by reason of improper workmanship or materials."
  
• Western Digital - Replacement, then equivalent item, then refund: "Western Digital will provide a Customer Replacement Unit (CRU) [...] it may instead provide a product with at least equivalent capabilities to the failed Product. [...] If Western Digital cannot provide a CRU replacement or an equivalent product, it may instead provide the user with a credit."
  
• Samsung - Replacement, then refund at current market value: "(1) repair or replace the Product with new or refurbished Product of equal or greater capacity and functionality; or (2) refund the then current market value of the Product at the time the warranty claim is made to SAMSUNG."
  
• Sandisk - Company chooses replacement or refund at current market value. "(1) repair or replace the Product with a new reconditioned or refurbished Product of equal or greater capacity, or another equivalent product; or (2) refund the current market value of the Product at the time the warranty claim is made."

The fact that nearly every company says it will honor a replacement is relevant against the backdrop of regional consumer protection laws. Most buyers' primary point of contact will be the store where they bought their hardware, but while most regional laws provide the buyer with multiple degrees of protection, for primary Western markets, the seller is apparently not obligated to replace the item and can instead issue a refund if said replacement would prove too onerous.

That seems to be the case with the European Union directive 2019/771, article 13.1.2.a, that sates that the consumer does have a choice between repair or replacement, "unless the remedy chosen would be impossible or, compared to the other remedy, would impose costs on the seller that would be disproportionate". The RAM price rise of 4-5x since last year would likely qualify, though individual EU member-states may offer additional protection, however.

The situation doesn't look too different in the United Kingdom, where the Consumer Rights Act of 2015 says, in section 23, item (3)(a), that although the customer has a choice between replacement or refund, they also "cannot require the trader to repair or replace the goods if that remedy (the repair or the replacement) [...] (b) is disproportionate compared to the other of those remedies."

As for the United States, most major PC hardware makers offer limited warranties, as opposed to full warranties, generally meaning that the manufacturer's policy dictates available options, save when specific state or regional laws apply. Broadly speaking, the stores are merely intermediaries in the process, and any protections they offer are a bonus.

The land down under has what appear to be the most consumer-oriented laws, as recently highlighted by the aforementioned Umart case. A browse of the Australian government's page dedicated to returns, refunds, and replacements clearly states that in case of a major failure of the item, "the consumer can choose," and should the seller be unable or unwilling to replace the item, the customer can "ask for a replacement [or] ask for a refund".

What this all means is that, should a person's RAM or SSD break under the manufacturer's warranty period in most regions, it might be wiser to try and go directly to the manufacturer instead of the place of sale, particularly considering that RAM almost always carries a limited lifetime warranty.

Just don't be too surprised if the AI component crisis continues to impact returns and refund policies, as other well-known manufacturers might be forced to follow SP, due to the shortage of DRAM or NAND flash chips.

New reports are now coming in from Japan, courtesy of IT Media, highlighting that customers are starting to pivot to high-capacity microSD cards to meet their storage needs. As a result, high-capacity MicroSD cards are now also encountering shortages.

We're currently traversing through one of the worst memory shortages in recent times, a situation that's expected to only get worse next year. Alongside DRAM, storage is also affected on an unprecedented scale, as data centers need every bit of flash they can get, leaving scraps for consumers to sift through.

"The larger versions of the popular series keep running out of stock," commented Technohouse Toei, a popular computer vendor in Tokyo. "This might drag on for a while." The models affected are 512 GB, 1TB, and 2TB microSD cards that offer a real substitute to SSDs and HDDs. Now that microSD Express is a thing, leveraging the PCIe interface and NVMe protocol with proper HBM support, it's actually feasible in these dark times that compact flash storage is replacing M.2 drives, at least when archival duties are not in the picture.

Speaking of drive shortages, Japanese stores also mention how large-capacity HDDs are being swept up and anything that remains on shelves is increasingly overpriced. One shopkeeper said, "It looks like large-capacity models are getting sucked up for AI use, so we aren't getting any stock at all. Everyone knows what's going on, so even if the price goes up, the drives sell out immediately… and the cycle repeats." In comparison, SSDs are "not as bad as HDDs," according to the retailer PC SHOP Ark in Chiyoda.

Memory is still the worst outlier, with DDR5 prices on the rise exponentially, racing to beat all records. The same 64GB Corsair Vengeance 5600 MT/s kit that cost ¥40,000 (~260) in early November, now costs ¥70,000 (~460) just three weeks in. That's an almost doubling on top of the already inflated $260 price tag it was sitting at. Still, these numbers are not as harrowing as the situation in the West, at least for now. Graphics cards, on the other hand, are relatively stable, but the rumor mill suggests they'll also be hit in 2026.

"Even when we finally get a shipment, the wholesale price has gone up by tens of thousands of yen. It’s happening all the time," said Tsukumo eX when referring to the ongoing DDR5 memory shortage. The number of physical price tags or stickers is visibly shrinking in stores because there's just not enough stock to sell. The AI boom is singlehandedly responsible for creating this landscape, and it's showing few signs of stopping anytime soon.

The Small Capacity Memory Card Championship (Japan), in which participants submitted their smallest memory cards, has been a tightly run race, but now the results are in. The smallest memory card found by the retro-computer-loving followers of X68PRO-HD on X was a 0.5MB SmartMedia card. While entrants such as the 1.44MB Flash FDD stick and Casio 2KB battery-backed RAM card didn’t make the cut (they didn’t quite fit the category), we were intrigued to see them.

Without further ado, the full list of tiny-capacity winners was as follows (machine translation):

• SD card: 8MB (SD, miniSD, microSD)
• MMC card: 4MB
• CF card: 2MB
• Memory Stick: 4MB
• SmartMedia card: 512KB (5V) / 2MB (3.3V)
• xD-Picture Card: 16MB
• PCMCIA card: 10MB / 2KB (SRAM)

If I were to join in this low-capacity competition, the best entrant from my drawers of detritus would be the 1GB xD-Picture card I found (see top picture). I also remember spending far too much on a 64MB CF card (and PCMCIA adaptor) back when digital cameras were cutting-edge tech objects of desire. Even that, though, seems massive compared to the 0.5MB SmartMedia card found by X68PRO-HD and their social media acquaintances.

The competition poster tipped a hat to other interesting small-capacity flash storage card entrants, but several were disqualified from the final list. Formats like PS1 memory cards, ‘bubble memory,’ USB thumb drives, and battery-backed flash RAM cards didn’t make the grade. Some of these were out of bounds as they weren’t ‘general-purpose’ flash memory cards.

Nevertheless, the rejects pile offered up some very interesting artifacts from earlier computing eras. Probably the most remarkable small-capacity entry that didn’t make it was the Casio 2KB battery-backed RAM card. This was a type of memory card exclusively for the Casio ‘Pokecon’ (ポケコン) range of pocket computers from the 1980s and 90s.

Another entry that didn’t quite make the grade was a 1MB PCMCIA card that was produced before it became a standard – that was a near miss.

Last but not least, it was interesting to see pictures and information about a 1.44MB ‘Flash FDD’ stick. This was a flash drive with the same capacity as a standard HD floppy. Moreover, it was designed with firmware to emulate a real floppy disk and could be booted from, like a floppy, by Windows and Linux users.

Kingston Technology has announced its first cable-free SSD. It will hope it has doubled the chances of its debut product getting a warm reception by fitting dual captive USB-A and –C connectors, one at either end of this high-performance memory stick. Kingston's Dual Portable SSD Storage Solution products are available right now. The firm’s handy USB 3.2 Gen 2 dual-interface SSD thumb drives are priced between $97 for a 512GB, and $239 for a whopping 2TB model. We might as well also mention, the Goldilocks 1TB model is $144. But, whichever you choose, they all boast “Up to 1,050MB/s read, 950MB/s write” speeds.

• Buy the 512GB Kingston Dual Portable SSD on Amazon
• Buy the 1TB Kingston Dual Portable SSD on Amazon
• Buy the 2TB Kingston Dual Portable SSD on Amazon

The recommended use case for these compact, flexibly interfaced drives is as an affordable, reliable, portable solution for data back-ups and transfers. Specifically, Kingston reckons the Dual Portable SSDs are a good choice “for large files, high-res photos and 4K videos.” It looks like a humble flash drive, albeit with a compact, durable metal casing, but this Kingston offers much better performance than a budget alternative might deliver.

“More and more consumers are looking to take their data into their own hands,” said Liny Cheliyan, Business Manager - Prosumer Flash and SSD Kingston EMEA. “Now with the convenience of Kingston’s Dual Portable SSD, users can do just that and easily transfer, share, or back up their important files across a variety of USB-A and USB-C devices.”

In addition to the dual-interface flexibility, the broad OS support of the Kingston Dual Portable SSDs will be welcomed by those carting their precious data between systems, places, and platforms. Checking the official specs table data, above, you can see users can easily transfer data between diverse platforms - as long as they sport either a USB-A or USB-C connector (or have an adaptor).

These devices look pretty durable, too, which is desirable for easily portable objects that may find themselves rattling around in bags, pockets, or worse. We'd have liked to see some kind of water/dust resistance added to the specs, but the cold/heat durability is welcome.

Kingston is giving buyers of its class as a premium portable flash solution a five-year warranty with free tech support.

Recovery teams working on the Titan submersible have found the vessel's specialist stills and video camera intact. Fascinatingly, while there was some damage to the camera’s housing and internal components, tech and science enthusiast Scott Manley reveals that the internal SD card was “undamaged.” Contents of the memory card have since been investigated, and 12 stills and nine videos have been recovered.

Click 'See more' for images.

In the images, you can see a SubC-branded Rayfin Mk2 Benthic Camera, recovered from the wreckage of the ill-fated submersible operated by OceanGate. This still and video camera is rated to withstand depths up to 6,000m (19,685 feet, 3,281 fathoms). The titanium and synthetic sapphire crystal constructed device features both onboard and expansion memory (the titular SD card).

Casing intact, lens shattered but remained in place

Inside the camera's tube-like form, it is observed that the PCBs had suffered some slight damage. For example, connectors between two boards were sheared off, and some surface mount components were similarly damaged.

In some images of the PCB that are shared, you will notice details are blurred at the request of the Canada-based underwater imaging specialist and Rayfin Mk2 Benthic Camera manufacturer. However, SubC’s requested trade secret obfuscation hasn’t stopped internet sleuths asserting that they know exactly what components have been redacted.

Picking through Manley's Tweet thread replies, the key PCBs in the camera were likely an Inforce 6601 System on Module (SoM) based on the Qualcomm SD820 processor, which comes with 4GB of RAM on board and 64GB of UFS storage. Another component is thought to be the Teensy 4.0 or 3.2, which acted as an MCU. Last but not least, the undamaged SD card is almost certainly a SanDisk Extreme Pro 512GB ($62.99 at the time of writing), though it was de-branded in the photos.

Data recovery process and results

With an undamaged SD card, of course, investigators (and others) were interested in what details of the tragedy may have been captured and stored by this camera system. The first step was to make “an exact binary image of the SD card” so the original could be left untampered.

Check out Manley’s Tweet thread, containing investigator report screengrabs, to take in the full gamut of back and forth between the data forensics investigator, Canada’s Transportation Safety Board of Canada (TSB), and SubC. To cut a long technical story short, though, the parties eventually met up at a lab/office in Newfoundland, where a recovered NVRAM chip and SD card image were interfaced with a “surrogate SoM board.” This did the trick, and 12 still and nine videos were recovered!

Recovered images were at a resolution of 4,056 x 3,040 pixels, implying a pretty common 12.3MP sensor was used by the SubC Rayfin Mk2 Benthic Camera. Videos were at a more standard 3,840 x 2,160 pixels – commonly referred to as 4K Ultra HD (UHD) video.

Somewhat disappointingly, the images and videos shared in the report were taken in the vicinity of the ROV shop at the Marine Institute, also in Newfoundland. The location was the logistical base for Titanic dive missions. No deep-sea shenanigans around the Titanic wreck were revealed. Manley explains in his Twitter thread that “the camera had been configured to dump data onto an external storage device, so nothing was found from the accident dive.” Nothing particularly pertinent to the tragic accident, that is.

Decades of evolution have made NAND flash memory both cheap and incredibly sophisticated. No matter what modern device you pick up, the likelihood is that it will use some kind of NAND, which has become ubiquitous in both consumer products and data centers. If you cast your mind back several decades, NAND was expensive, despite its capability to be faster than spinning disks. As the technology has matured, its use has grown exponentially. But what's next?

In the coming years, we're going to see cheaper, more advanced non-volatile flash memory inside our devices. Those 3D NAND devices will be built from over 500 layers and boast interface speeds of up to 4800 MT/s. This blistering speed may pave the way to all-new kinds of devices with significantly increased performance compared to today's best SSDs.

The next step

Like other technologies, 3D NAND develops in multiple directions. End users demand higher capacity, higher performance, lower power, compact dimensions, and lower cost. Manufacturers themselves want to earn money, so while meeting demand, they try to cut costs by making smaller memory devices. Given the current state of industry development, the best way to keep the size of 3D NAND dies in check is to increase the number of layers and decrease the size of the actual data storage location — the memory cell itself. This practice is commonplace among memory makers today.

Leading makers of 3D NAND memory, including Kioxia, Micron, Samsung, SK Hynix, and YMTC, have all developed multiple 3D NAND generations, consistently increasing layer counts while reducing cell sizes. For now, the majority of 3D NAND-based devices, including SSDs and smartphones, rely on memory with 2xx layers, with each layer holding an incredible number of memory cells for data storage.

This might be 218-layer 3D NAND from Kioxia, 232-layer 3D NAND from Micron, or 286-layer 3D NAND from Samsung.

The pace at which manufacturers adopt new layer counts and boost transfer rates varies across the industry, making things difficult to track. So let's examine where we were in 2023, how it's going, and where we might be in 2027. Just as a reminder: Kioxia also has a close production partnership with Western Digital's flash division, which also owns SanDisk as a subsidiary. Within this article, we'll refer to the duo as Kioxia / Western Digital.

In recent years, the industry faced significant challenges in building an SSD with a PCIe 5.0 x4 interface that would fully saturate that interface. ICs with a high-performance interface that could fully saturate the lane using an eight-channel controller (e.g., 2400 MT/s) are rare. Instead, ICs that supported lower speeds are much more common.

By early 2023, all three major NAND vendors — Kioxia / Western Digital, Micron, and Samsung — had established 2400 MT/s as the baseline interface speed across TLC products. Kioxia's BiCS 6T and 6Q nodes, Micron’s B58, and Samsung’s V8T are representative of these tiers as a mainstream standard, and perhaps this reflects a consistent performance scaling for both consumer and enterprise storage segments.

In 2024, Samsung took the next incremental step by introducing its V9T TLC at 3000 MT/s. This intermediate speed tier, not shared by other vendors in the roadmap, highlights Samsung’s focus on being a vertically integrated company that focuses mainly on its own-brand products, like smartphones and SSDs.

Kioxia / Western Digital reached 3200 MT/s in the second half of 2024, with its 8th Gen BiCS 3D TLC, followed by QLC (BiCS 8Q) products in 2025 H2. However, Kioxia / Western Digital has not mentioned using 7th Gen products at all, skipping it entirely.

This reflects a general push to align both TLC and QLC offerings on higher-performance platforms. Notably, Kioxia / Western Digital is the only vendor explicitly listed in the roadmap as deploying 3200 MT/s in both cell types, showing a strong commitment to QLC performance uplift.

In 2024, Micron came to leadership with its B68S IC, sporting a 3600 MT/s data transfer rate. By the second half of 2025, the company’s QLC roadmap plans to catch up with Samsung via the N69 generation. This highlights Micron’s early adoption of faster interfaces for high-end products, but since it's yet to be released, its impact and uptake remain to be seen. Kioxia/Western Digital also lists its BiCS 8Q QLC reaching 3600 MT/s in 2025 H2, indicating a broader QLC performance push across the industry.

All vendors eventually converge at 4800 MT/s by 2027, which becomes the high-performance standard. Kioxia/WD reaches this milestone with BiCS 9T TLC in 2027 H1, Micron with B78 TLC and N79 QLC in 2027 H1–H2, and Samsung with its 11th generation of V-NAND.

Performance headwinds

Manufacturing high-speed NAND flash (3600 MT/s and higher) requires a more expensive logic process technology, which drives up production costs and inflates flash memory prices.

An increase in 3D NAND data transfer rates will have a very practical implication on how devices like SSDs or smartphones are built.

To fully saturate a PCIe 5.0 x4 interface — which provides up to 15.75 GB/s of effective bandwidth — a modern NAND flash setup requires an SSD controller with eight channels at 2400 MT/s, six channels at 3600 MT/s, or just four channels at 4800 MT/s. In a year or two, PCIe 5.0 SSD controllers won't require eight channels to fully saturate the hungry interface.

For PCIe 6.0 x4, which doubles throughput to around 31.5 GB/s effective (after protocol overhead), the number of required channels increases 16 at 2400 MT/s, 12 at 3600 MT/s, and eight at 4800 MT/s. As NAND I/O performance approaches these thresholds, high-end SSDs can exploit PCIe 6.0 capabilities with eight high-speed NAND devices. However, it remains to be seen whether PCIe 6.0 x4 SSDs will be a reality for consumer storage devices within the next few years.

Die capacities

Based on the roadmap data that we have collated, leading 3D NAND makers will converge, landing with 1 Tb and 2 Tb die capacities by around 2026 - 2027, particularly for 3D QLC products. While there are some 2Tb QLC NAND devices announced as of May 2025, none of them are actually on the market yet.

Micron, Kioxia/Western Digital, Samsung, SK Hynix, and YMTC all plan to offer 3D QLC devices reaching 2 Tb capacities, typically paired with interface speeds of 3200MT/s – 4800 MT/s over the next few years. Despite the significantly increased layer count, 3D NAND manufacturers are not jumping to capacities higher than 2Tbs per device for the time being, perhaps due to cost concerns.

Rising layer counts: How does each company stack up?

Kicking off with Kioxia and Western Digital, this duo's 3D NAND roadmap featured BiCS 6 with 162 layers across both TLC and QLC variants in the first half of 2023. BiCS 6 represents a mature and commercially proven node, used in mainstream SSDs and enterprise storage. This generation aligns with the industry’s transition toward high-density vertical stacking, while supporting a 2400 MT/s interface speed.

Kioxia and Western Digital didn't stop there, and in the second half of 2024, these companies introduced BiCS 8 in both TLC (BiCS 8T) and QLC (BiCS 8Q) forms, now featuring between 218 and 232 layers. This generation continues to rely on 2400 MT/s – 3600 MT/s interfaces, marking both vertical and performance scaling. Also, it demonstrates Kioxia’s intent to advance QLC’s competitiveness in both storage density and performance.

BiCS 8 will be a fairly long-standing node for Kioxia and Western Digital, as it will be replaced in the first half of 2027. BiCS 9T is set to replace it, boasting a layer count of greater than 300, representing quite an aggressive step forward. This milestone puts Kioxia in alignment with other vendors’ 3D NAND scaling targets, while also introducing 4800 MT/s interface speeds.

Micron’s lineup includes the now legendary B58 IC with 232 layers and a 2400 MT/s data transfer rate. But its 1st Gen ICs featured a lower rate; its full speed was only achieved sometime in the first half of 2023. Nevertheless, the B58 gave Micron an early lead in vertical scaling versus competitors still shipping 162–192 layer products.

In the second half of 2024, Micron advanced to the B68S IC, increasing the layer count to 276. This generation pairs higher density with elevated performance (3600 MT/s), making it a major step in Micron’s 3D NAND roadmap, and making the company one of the leading makers of enthusiast-grade SSDs with the model 4600.

In the first half of 2027, Micron is expected to introduce its B78 3D TLC NAND ICs, with a 4800 MT/s interface. However, the exact layer count is not known. Given the trend from B58 (232L) to B68S (276L), it is reasonable to expect B78 to reach well beyond 300 layers, likely on par with Kioxia’s and Samsung’s leading nodes. However, there are no official materials on exact layer counts, so this remains speculation.

When it comes to 3D NAND makers, Samsung is a rare animal. All producers of flash memory prefer to sell their chips as parts of their own devices, such as SSDs, and to that end, Micron works closely with Silicon Motion and Phison to optimize controllers for its memory and drives. In a similar vein, Kioxia works with Phison. Unlike its rivals, Samsung has its own 3D NAND and controllers to build its SSDs and other products using NAND flash memory. The company also sells its 3D NAND ICs on the open market as an additional boon.

Samsung offered its V8T IC in 2023 H2 with 236 layers, placing it alongside Micron’s B58 and slightly ahead of Kioxia’s BiCS 6. This node continues to be Samsung’s workhorse today.

By the second half of 2025, Samsung is set to mass produce V9T, scaling to 286 layers — a 50-layer jump over V8T. This milestone indicates V9T’s interface speed remains at 3600 MT/s, indicating a controlled ramp before moving to more aggressive performance targets.

Looking ahead to 2026 H1, Samsung’s roadmap lists V10T IC with an expected layer count of over 400, representing the most ambitious vertical target. While this node’s interface speed is also pegged at 4800 MT/s, its standout feature is the sheer layer height, suggesting multi-tier stacking and next-gen integration techniques, such as string stacking and hybrid bonding. However, Samsung has not officially announced that it is using any of those specific techniques quite yet.

SK hynix’s current product lineup includes the V8T IC with 238 layers, a 2400 MT/s interface, and has been available since the second half of 2023, marking their entry into late-Gen 3D NAND scaling. The next major leap occurs with V9T, reaching 321 layers in the first half of 2025, and then increasing interface performance to 3600 MT/s by the second half of 2025.

This shows a rapid vertical scaling effort while improving I/O bandwidth. The company's most ambitious node, V10T, is set to debut in 2027 H1, with 500 layers and a 4800 MT/s interface — matching top-tier bandwidth targets from other vendors, and surpassing them in vertical stack height. The roadmap extends further to V11T (4800 MT/s), though its layer count is not explicitly known.

On the QLC side, SK hynix’s roadmap begins with V7Q in the first half of 2024 at 176 layers, then it extends to V9Q, which delivers 3200 MT/s performance and a 2 Tb die capacity — implying a 300L+ stack. By the first half of 2027, QLC products are also expected to adopt the 500-layer structure (V10Q) at 4800 MT/s, mirroring TLC’s configuration.

When it comes to layer counts, China-based YMTC is not a leader, but its 3D NAND development shows a clear trajectory from mid-tier to high-tier stacking. YMTC’s 3D TLC product (X4-9060) features 128 layers and a 3600 MT/s interface, and was followed quickly by X3-9070 in the second half of 2024. X3-9070 jumped to 232 layers while maintaining the same data transfer rate. This indicates a rapid scaling of vertical density — despite sanctions by the U.S. government. This is likely enabled by the company’s proprietary Xtacking architecture, which separates logic and memory layers to simplify integration. However, no QLC-specific layer counts are known for now.

By 2026 – 2027, YMTC’s roadmap aims to enter the high-performance class of its competitors. The X5-9080 and X5-9070 3D TLC NAND products operate at 4800 MT/s, and the latter is explicitly listed as a 1 Tb die, suggesting a likely move to 300+ layers, although exact counts are not known. Meanwhile, the company’s 3D QLC line (e.g., X4-6080, X5-6080) appears on the roadmap in parallel, also reaching 4800 MT/s by the second half of 2026, hinting at comparable architectural advances.

YMTC’s roadmap outlines a fast-paced rise from 128L to 232L and potentially 300L+ stacking within three years—driven by a balance of aggressive layer scaling and interface speed improvements across both TLC and QLC products.

What's next for 3D NAND?

The future of 3D NAND is defined by fairly aggressive vertical scaling, faster interfaces, and a push toward higher per-die capacities, which will enable products that will enable faster, denser, and more cost-efficient storage across consumer and enterprise segments. To make that happen, all major vendors are set to converge at a 4800 MT/s interface speed and 300–500-layer stacks by 2027.

However, this progress is not without complexity. Vertical scaling introduces new challenges in terms of yield, while higher speeds now also require logic process nodes, which raises manufacturing costs. Eventually, when everyone in 3D NAND manufacturing converges at hybrid bonding (which YMTC already uses, while sporting one of the industry’s widest patent portfolios). These techniques will add another layer of complexity, and companies will require some time to get used to these new processes.

In any case, the future of 3D NAND is bright, as the industry races towards faster speeds, higher stacks, and more complex manufacturing technology.

Raja Koduri continues to push his high bandwidth vision for computing, but this time it's flash, namely HBF (High-Bandwidth Flash) from Sandisk. On Wednesday, the former graphics architect at S3, Apple, AMD (twice), and Intel, announced that he had joined Sandisk’s HBF Technical Advisory Board to guide development and strategy for High-Bandwidth Flash memory technology. Sandisk also shared a press release confirming Koduri’s new role, as well as the appointment of Professor David Patterson to the board.

In case you missed it, Sandisk outlined its newest flash memory innovation back in February this year. Dubbed HBF, the key feature of this flash would be the enabling of GPU packing up to 4TB of VRAM – with more to come in future revisions.

Of course, this isn’t for enthusiasts (yet), but the proponents of HBF want to unleash this huge memory pool attached to a GPU for AI inference workloads. When it was announced, we learned that HBF would weave its magic using multiple high-capacity, high-performance flash core dies, interconnected using through-silicon vias (TSVs), on top of a logic die that can access flash arrays (or rather flash sub-arrays) in parallel. Thus, it would “deliver 8-16X [VRAM] capacity at similar cost.” It relies on BICS 3D NAND using the CMOS directly bonded to Array (CBA) design that bonds a 3D NAND memory array on top of an I/O die made using logic process technology.

Ideas man David Delahunty is behind a new flash storage device which looks like a classic MacOS folder icon dragged into the real world. This USB-C attached 128GB capacity storage product is called the FolderDrive, and it will be available soon for about $40.

We don’t have a lot of technical details for the FolderDrive, something which may disappoint the most hard-nosed segment of the Tom’s Hardware readership. However, this product is admittedly one that is first and foremost “whimsical, minimal,” yet functional. So, if you are simply after the best performance and capacity, you better check our guide to the best flash drives.

Dalhunty’s “literal folder for your digital folders” comes in just a single 128GB capacity option, connects via USB-C, and measures 3.25 wide x 2.5 tall x 0.5-inches thick. That’s the full spec sheet, but it is vaguely promised to be fast, reliable and lightweight - without any stats to back up these assertions.

We might have also thought it came in one color – the old MacOS default kind of light blue. Except, Delahunty’s Twitter/X feed shares a glance at some FolderDrives in a few other MacOS appropriate shades.

Small-scale producer Super Fantastic is Delahunty’s partner for getting this FolderDrive into consumer hands (linked top). It indicates the product will be “3D printed with care." Such attention to detail is important, as this product leans heavy on its appearance for being a “great conversation starter for designers, developers, and tech lovers.”

In the Twitter/X thread, we note a little controversy over another designer who originated a folder-icon-like flash drive a few months earlier. But that is quite a different design, and not in production, as far as we are aware.

Reverse skeuomorphism

Apple might have been a pioneer in popularizing skeuomorphism for computer UIs, and I remember some very striking iOS features and apps using this design direction several years ago. Reversing this idea thus seems attractively retro-whimsical.

Similar products based on Floppy icons, CD icons, and other memorable icons would also be quite fun to bring into the real world. A quick search surfaced some floppy disk USB drives for sale, but not in a cutesy computer UI style design. Along similar lines, there is also the WinRAR bag, which we mentioned recently.

SanDisk on Wednesday introduced an interesting new memory that could wed the capacity of 3D NAND and the extreme bandwidth enabled by high bandwidth memory (HBM). SanDisk's high-bandwidth flash (HBF) memory enables access to multiple high-capacity 3D NAND arrays in parallel, thus providing plenty of bandwidth and capacity. The company positions HBF as a solution for AI inference applications that require high bandwidth and capacity coupled with low power requirements. The first-generation HBF can enable up to 4TB of VRAM capacity on a GPU, and more capacity in future revisions. SanDisk also foresees this tech making its way to cellphones and other types of devices. The company hasn't announced a release date yet.

"We are calling it the HBF technology to augment HBM memory for AI inference workloads," said Alper Ilkbahar, memory technology chief at SanDisk. "We are going to match the bandwidth of HBM memory while delivering 8 to 16 times capacity at a similar cost point."

Conceptually, HBF is similar to HBM. It stacks multiple high-capacity, high-performance flash core dies interconnected using through-silicon vias (TSVs) on top of a logic die that can access flash arrays (or rather flash sub-arrays) in parallel. The underlying architecture of HBF is SanDisk's BICS 3D NAND using the CMOS directly bonded to Array (CBA) design that bonds a 3D NAND memory array on top of an I/O die made using logic process technology. That logic may be a key to enabling HBF.

"We challenged our engineers and said, what else could you do with this power of scaling," said Alper Ilkbahar. "The answer they came up with […] was moving to an architecture where we divide up this massive array into many, many arrays and access each of these arrays in parallel. When you do that, you get massive amounts of bandwidth.  Now, what can we build with this? We are going to build high bandwidth flash."

Traditional NAND die designs often treat the core NAND flash memory array as planes, pages, and blocks. A block is the smallest erasable area, and a page is the smallest writable area. HBF seems to break the die into 'many, many arrays' so they can be accessed concurrently. Each sub-array (with its own pages and blocks) presumably has its own dedicated read/write path. While this resembles how multi-plane NAND devices work, the HBF concept seems to go far beyond them.

For now, SanDisk says that its 1st-Gen HBF will use 16 HBF core dies. To enable such devices, SanDisk says that it has invented a proprietary stacking technology that features minimal warpage to enable stacking 16 HBF core dies, and a logic die that can simultaneously access data from multiple HBF core dies. The complexity of logic that can handle hundreds or thousands of concurrent data streams should be higher than that of a typical SSD controller.

Unfortunately, SanDisk does not disclose the actual performance numbers of its HBF products, so we can only wonder whether HBF matches the per-stack performance of the original HBM (~ 128 GB/s) or the shiny new HBM3E, which provides 1 TB/s per stack in the case of Nvidia's B200.

The only thing we know from a SanDisk-provided example is that eight HBF stacks feature 4 TB of NAND memory, so each stack can store 512 GB (21x more than one 8-Hi HBM3E stack that has a capacity of 24 GB). A 16-Hi 512 GB HBF stack means that each HBF core die is a 256 Gb 3D NAND device with some complex logic enabling die-level parallelism. Funneling hundreds of gigabytes of data per second from 16 3D NAND ICs is still quite a big deal, and we can only wonder how SanDisk can achieve that.

What we are sure about is that HBF will never match DRAM in per-bit latency, which is why SanDisk stresses that HBF products are aimed at read-intensive, high-throughput applications, such as big AI inference datasets. For many AI inference tasks, the critical factor is high throughput at a feasible cost rather than the ultra-low latency that HBM (or other types of DRAM) provides. So, while HBF may not replace HBM any time soon, it might occupy a spot on the market that requires high-capacity, high-bandwidth, NAND-like cost but not ultra-low latency. To simplify the transition from HBM, HBF has the same electrical interface with some protocol changes, though HBF is not drop-in compatible with HBM.

" We have tried to make it as close as possible mechanically and electrically to the HBM, but there are going to be minor protocol changes required that need to be enabled at the host devices," said Ilkbahar.

SanDisk didn't touch on write endurance. NAND has a finite lifespan that can only tolerate a certain number of writes. While SLC and pSLC technologies offer higher endurance than the TLC and QLC NAND used in consumer SSDs, this comes at the expense of capacity and adds cost. NAND is also typically written to at block granularity, whereas memory is addressable at the cache line level (i.e. typically 128KB for NAND blocks versus 32 bytes for a cache line). That's another key challenge.

SanDisk has a vision of how its HBF will evolve over three generations. Nonetheless, for now, SanDisk's HBF is largely a work in progress. SanDisk wants HBF to become an open standard with an open ecosystem, so it is forming a technical advisory board consisting of 'industry luminaries and partners.

Despite the rise of cloud-based storage, USB flash drives have been the go-to option for those who need local, pocketable access to important files for decades. These tiny drives are still sold nearly everywhere (including at chain drug stores, often at highly inflated prices), despite companies like Microsoft, Google, and Dropbox making online storage a convenient and often free (depending on capacity) alternative.

But because flash drives often work just fine for a decade or more, you could be hanging onto one that’s pitifully slow by today’s standards, and too cramped to hold all the data you need here in 2026. Capacities of up to 2 TB are an option with several recent flash drive models (and 1 TB is common), and the performance of the fastest flash drives approaches that of some of the best external SSDs.

We’ve tested dozens of recent flash drives, from leading brands and lesser-known alternatives, and listed the best USB flash drives below, to help narrow down your pocket-friendly storage search. That said, if fast performance is key to your workflow and you can deal with a device that's just a little bit bigger and requires a cable, one of the best external SSDs is usually a better buy.

Also, if you have a spare M.2 SSD from upgrading a laptop or desktop, you can save some money building your own external SSD by mounting an old drive in an enclosure.

 Best Flash Drives You Can Buy Today

Best Fast, Affordable Flash Drive

Just like its roomier 1TB variant, the 256GB Kingston DataTraveler Max is a speed demon in the flash drive world. At $30-$35, Kingston’s 256GB drive is a great balance of speed, affordability, and capacity for those who don’t want to spend much more on a roomier, somewhat faster alternative.

Available with either a USB-C or USB-A connector, the 256GB DataTraveler Max performed nearly as well as its more spacious 1TB sibling in our PCMark, DiskBench, and CrystalDiskMark tests, making it one of the fastest flash drives we’ve tested. And it’s also fairly slim, making it easy to slip into a pocket.

Just note that, if you’re a desktop user and you opt for the model with the USB-C port, you will likely have problems plugging the drive into rear USB-C ports on the motherboard. Thanks to the two-piece plastic slide mechanism that protects the drive’s port when not in use, the USB-C connector is slightly too short to plug into most motherboards successfully.

I tried this with three boards I have on hand and had the same issue with each. But, if your PC case has a front USB-C port or you’re using a laptop, this shouldn’t be an issue. I only had this problem with rear motherboard ports and this drive.

Those looking for similar performance and price without the USB-C connector issues should instead consider the 256GB TeamGroup X1 Max. It has a USB-C port on one end (that we didn't have problems connecting to a motherboard) and a USB-A port on the other. Its performance is also excellent, but the caps on either in don't fit that snugly and will be easy to lose.

Fastest Flash Drive

The line between flash drives and external SSDs is increasingly indistinct, and SK hynix's Tube T31 blurs it even more, by putting an actual M.2 drive on a small PCB and shoving it into a somewhat bulky but surprisingly speedy flash drive form factor.

Capacity is limited to just 512GB and 1TB models, but this is the fastest "SSD stick" we've tested yet, surpassing Transcend's ESD310C and Kingston's DataTraveler Max drives on most of our tests. Priced at $85-$90 when we wrote this, it's also slightly more affordable than those drives at the 1TB capacity.

If you're after a simple portable drive and don't need 20 Gbps speeds or a capacity higher than 1TB, it's a great choice. Its bulk might get in the way of nearby ports on a desktop or hub. But on a laptop, the USB-A port (if you still have one) is likely set apart from other USB ports. And on a desktop, you probably have several USB-A ports to choose from, unlike USB-C.

Read: SK hynix Tube T31 Review

Fastest USB-C Flash Drive

Much like the SK hynix Tube T31 drive (which is generally slightly faster in testing but also larger) Adata's SC750 blurs the line between flash drive and SSD, by essentially putting a retractable USB port on a small SSD inside a plastic housing, making for a compact flash drive that beats nearly every competing drive that doesn't require an external cable.

In our testing, the 10Gbps-rated Adata SC750 beat the Tube T31 in PCMark 10, came in second in reads in our 10GB DiskBench read test, and was faster than the hynix drive in writes. Adata's drive didn't look quite as good in our CrystalDiskMark results, but overall it's generally the second-fastest drive we've tested in this category.

And unlike the Tube T31, it uses USB-C rather than the older USB-A port. So if you want to plug your drive into a phone or you have a laptop with only USB-C ports, it should be at the top of your list for convenient and speedy external storage.

In terms of price, the ADATA SC750 may cost slihgtly more or slightly less than SK hynix's drive, depending on whether the latter is on sale. But given the similar performance and what generally seems to be about a $10 price difference between the two drives, if you're considering both drives, you should probably make your choice based on which port (USB-C for Adata, USB-C for SK hynix) is the most convenient for your use cases.

Read: Adata SC750 review

Best Cheap and Tiny Flash Drive

If you're after a drive that's small enough to leave plugged into your laptop most of the time, or you just want something you can clip to your keychain and forget about until you need it, Samsung's FIT Plus stands out. The 128GB model we tested sells for $18, and when we wrote this the top-capacity 256GB model was selling for just $25 direct from Samsung.

At less than an inch long, including its USB-A connector, it will stick out of your port less than an inch when plugged in. And it has a hole for a lanyard, it feels solid – in fact, this sub-$20 drive feels better than the most expensive drives on our list. And Samsung says it's rated to survive in up to 1 meter of water for 72 hours. In case it doesn't, the company covers the drive with a generous five-year warranty.

That being said, our testing shows this tiny drive is in a much lower performance class than the bigger, pricier, roomier models. On our real-world DiskBench 10GB test, the Samsung Fit Plus read our files at a decent 319.7 MB/s, but could only write at 58.4 MB/s.

That's just over half the read speed and less than 15% of the write speed of Kingston's 1TB DataTraveler Max. That said, the Samsung drive's performance was still significantly better than most of the drives we tested with capacities less than 1TB. SanDisk's Extreme Pro 128GB wrote our test files about twice as fast, but it costs nearly 2.5 times as much and is more than six times longer.

Best High Capacity Flash Drive

There are loads of insanely cheap no-name 2TB flash drives available on Amazon and sites like Aliexpress if you like throwing your money at scammy garbage tech – but the 2TB Buffalo SSD-PUT is real and (at $160 when we wrote this) a surprisingly good deal. It's cheaper than most larger 2TB external SSDs (although some of those will definitely be faster), and significantly smaller (although still bulky for a flash drive).

Overall, this drive did slightly better on our benchmarks than its 1TB sibling, meaning it doesn't beat or often quite match the Kingston DataTraveler Max. But for the price -- and especially considering it's one of a very few 2TB drives available from a known brand, this is a very impressive performer.

On the minus side, the drive has the same creaky, cheap-feeling plastic shell as the 1TB model, and it's nearly an inch wide, making it a tight fit on some laptops where the ports are placed close together. But you get a roomy 2TB of pocket-friendly storage at fast speeds (for a flash drive), and a bundled USB-C adapter should your device not have a USB-A port handy.

If Buffalo would just release this drive in solid-feeling metal housing (or even a solid-feeling plastic one), this would arguably be the perfect flash drive for those who need lots of storage in their pocket.

Flash Drive Shopping Considerations

How We Test Flash Drives

Our current testbed consists of an ASRock Z790 Taichi Lite motherboard and a Core i5-12600K CPU. This system gives us native Thunderbolt 4 ports and a USB 3.2 Gen 2 2x2 port via the front header (connected through Corsair's 5000D Airflow case).

All of our benchmark testing was done with drives plugged into the system's rear USB 3.1 Gen 2/ USB 3.2 (10 Gbps) ports. The Aorus board we used has one Type-C and one Type-A port rated for these speeds, so we were able to accommodate drives with both types of connectors.

To get a sense of how these flash drives compare to a larger (though still usually pocketable) external SSD, we also ran our flash drive tests on the Mushkin CarbonX, a 1TB External SSD that's rated to similar speeds as the fastest flash drives, or "Up to 1,000 MBps."

This drive is no longer widely available, but you can expect similar performance from some of the more affordable options on our Best External SSDs list.

Given external SSDs' general ability to perform better on longer-duration workloads (and particularly small file writes, which we'll see shortly in testing), an external SSD is often a better option if you are going to frequently be moving large amounts of files on and off your drive, and especially if you plan to run programs from your portable storage device.

The larger surface area and improved controllers and components of external SSDs tend to make them better at those kinds of tasks, though there are of course noticeable performance differences in that product category as well.

Trace Testing - PCMark 10 Storage Benchmark

PCMark 10 is a trace-based benchmark that uses a wide-ranging set of real-world traces from popular applications and everyday tasks to measure the performance of storage devices.

Transfer Rates – DiskBench

We use the DiskBench storage benchmarking tool to test real-world file transfer performance with a custom 10GB dataset. We copy 1,204 files (images, videos, and software ISO files) to a folder on the test drive (write). Then, after leaving the system idle for five minutes, we run the same test in reverse, moving the test folder to a different location on our PCIe 4.0 testing drive.

Synthetic Testing CrystalDiskMark

CrystalDiskMark (CDM) is a free and easy-to-run storage benchmarking tool that SSD companies commonly use to assign product performance specifications. It gives us insight into how each device handles different file sizes. We run this test at its default settings.

A Note on Temperatures

If you've used a flash drive in the past and written lots of data on it, there's a good chance you've grabbed it to yank it out when you're done and winced at the hot – or at the very least surprisingly warm – exterior. We've certainly dealt with uncomfortably toasty drives before, but perhaps that's becoming an issue of the past.

We initially checked 10 of the drives we tested, and none felt hot to the touch after writing large amounts of data. We used an IR thermometer to check the temperature of several during a long 100GB write test.

After several minutes of sustained writing, the Transcend and PNY drives got the warmest, but were still under 94 degrees Fahrenheit. The solid-feeling metal-clad Orico drive got up to just 83.4 degrees, and the similarly solid OWC Envoy Pro Mini remained the coolest, at just 78.5 degrees Fahrenheit. The tiny Samsung Fit Plus, which barely has any surface area compared to the other drives, topped out at 82.1 degrees in our testing.

For future drives, we will only measure surface temperatures when a drive feels unusually warm during testing.

Other Flash Drives We Tested

Kingston Dual Portable SSD (1TB): This dual-port 10 Gbps drive from Kingston delivers convenience, versatility, and chart-topping synthetic and sequential read performance, outpacing our favorite USB-A-only SK hynix Tube T31 in some of our tests.

But it isn't quite the fastest when it comes to writes, and its small file performance was mixed, marking it as not the best choice for running programs or an operating system directly off the drive.

Still, the Kingston Dual drive makes a strong case as a fast and compact file mover, and is available in capacities up to 2TB, despite its slim metal shell. It's a worthy alternative to the Tube T31 if you're primarily using the drive to back up and carry files (and you need USB-C connectivity).

But it's also currently expensive, selling for north of $150 for the 1TB model we tested, while the SK hynix drive was available at Micro Center for about $50 less.

The Tube T31 is increasingly scarce, however, while Kingston's drive just recently launched. If the Dual Portable SSD goes on sale, it's a great option if, again, you don't care much about small file performance.

SanDisk Extreme Fit (1TB): This tiny USB-C drive is designed to be a semi-permanent attachment to the side of your laptop (or the back of your desktop), as an easy way to add storage space. Available in capacities ranging from 128GB to 1TB (the capacity we tested), it's not aiming to be a speed demon, with advertised read speeds of "up to 400 MB/s."

In our testing, it achieves that in our sequential CrystalDiskMark test (419.34 MB/s, to be exact), while delivering a decent real-world read speed of 318.3 MB/s in our 10GB file transfer test.

But write speeds are much lower (82.52 MB/s in the same 10GB test), and the small SanDisk drive's score of 274 on our PCMark 10 Data Drive test was less than a quarter that of the chart-topping Kingston drive. For its general purpose of semi-permanent storage, it's fast enough for general use without feeling sluggish (particularly when retrieving files).

But as small as it is, it doesn't sit flush with your USB port, making snags or bumps an obvious worry. And it's wide enough that it comes very close to blocking the adjacent USB-C port on my Asus laptop. So unless you're dealing with sensitive files or have spotty Wi-Fi, you may want to opt for cloud storage instead.

Seagate Ultra-Compact SSD (2TB): Despite being one of the best-looking and premium-feeling of the recent spate of so-called "stick SSDs," Seagate's flash drive-style drive just didn't stand out in benchmark testing, often lagging behind the SK hynix Tube T31 and TeamGroup's X1 Max, both of which cost less.

A price drop would make this drive more appealing, but it just wasn't as speedy in our tests as other high-end stick storage options.

PNY Pro Elite V3 (1TB): This third iteration of PNY's high-end flash drive has a solid metal sliding shell that protects its USB-A port when not in use. Its 1,000 MB/s read and 800 MB/s write speed rating, in particular, is a bit more modest than the claims of its high-end competition.

And while it exceeded those speeds in our sequential test, it disappointed elsewhere, like in PCMark, where it scored only half of some of the best alternatives we've tested.

It also couldn't crack the top five in our real-world read test (though its write speed there was nearly the best). So while the Pro Elite V3 is head and shoulders above the V1 and V2 that came before it, it doesn't stand out in terms of overall performance.

And at $105 for the 1TB model we tested, it's also one of the priciest 1TB drives we've tested. When we wrote this, the faster SK hynix Tube T31 was $86 at the same capacity, making it easily the better drive.

Silicon Power MS70 (1TB) and DS72 (1TB): These two drives from Silicon Power perform roughly the same and have similarly solid-feeling shells and flip-up caps. The DS72 has a USB-C connector on one and and USB-A on the other, while the MS70 makes do with just USB-A.

The Silicon Power drives also perform fairly well overall, but were inconsistent on our tests. In Crystal Diskmark's sequential tests, they were the best flash drives we've tested yet, and they were also among the best on our 10GB DiskBench test.

But on the PCMark 10 Data Drive benchmark, they consistently delivered scores that were roughly half of what the Transcend and Kingston drives we tested and they weren't particularly impressive in our 4K test, either.

Pricing for the 1TB models that we tested was also in an awkward middle area, between our favorite mainstream drive from Buffalo and the better-performing options from Transcend and Kingston.

Still, if the price drops by $10-$20, these would be good options, particularly if you mostly want a drive to move files from one place to another rather than to run programs directly. Their physical design and build quality feel better than some of the more expensive options out there.

Orico USSD-X (512GB): Orico's latest flash drive is wrapped completely in metal, and feels surprisingly solid in your hand. And supposedly it will be offered in several colors. Its performance in our testing hovered between middle of the pack and near the top, excelling in our PCMark and Diskbench tests, especially.

That said, its performance didn't stand above the rest in any single test, and at the time of testing, we couldn't find it for sale in the US. Hopefully, the company improves its supply issues, because this USB-A drive feels better to hold and use than arguably any other flash drive we've tested. And its performance is quite solid. We just need to know how much it will cost.

OWC Envoy Pro Mini (1TB): This drive was the second-fastest overall and sports a solid metal shell. But its design is overly complicated, as is its setup process (which forces you to agree to a EULA which, when we wrote this, was still 404ing). Really though, the primary problem with OWC's drive is price. At $149 for the most spacious 1TB model, it's nearly as expensive as the 2TB Buffalo drive. And often the Buffalo drive goes on sale for less.

Patriot Supersonic Rage Prime (1TB): With a name this obnoxiously aggressive (and $90 price to match), we were expecting impressive things from Patriot's top-end flash drive. And it did quite well in some tests, but struggled on our 10GB file transfer test, managing to write at just 91.7 MB/s–behind all but the lower-capacity, much lower-priced drives on this list.

PNY Pro Elite (1TB): We like the solid metal shell of this PNY drive, but its $130 asking price is higher than any other 1TB drive we've tested. And it struggled on our 10GB write test, managing just 96.9 MB/s, which was slower than even Samsung's 256GB Duo Plus drive. There isn't much that's pro or elite about that.

Orico UFSD-C (1TB): This metal-clad USB-C drive looks and feels great and performed quite well overall, though its benchmark results didn't stand out in any real way. Its main issue is availability. The company sent us the 1TB model we asked for, but it doesn't seem to be for sale anywhere in the US.

You can order it on Aliexpress, but even there it's only available at up to 512GB, and its $82 current price is close to what we'd expect the 1TB model to sell for.

Samsung USB-Type-C (‎128GB): One of Samsung's newer drives, this model sports solely a USB-C connector and isn't much bigger than the USB-A Fit Plus drive. But it's slightly more expensive than the Fit drive and had a tendency to land near the bottom of our tested Samsung drives in terms of performance.

If you need USB-C and don't need a lot of speed (and particularly write speed) and capacity (it tops out at 256GB), this isn't a bad drive. It just doesn't stand out in any substantive way. And we wish its write speed were at least twice as fast as the 59-67 MB/s we saw in our sequential tests. Read speeds were, at least, much faster at more than 300 MB/s.

SanDisk Extreme Pro (128GB): While SanDisk's high-end drive performed well compared to other 128GB models, its $43 price at that capacity is about 2.5x that of Samsung's Fit Plus. And while it is available in up to 1TB capacities, the most spacious model sells for between $130 and $200, making it much more expensive than competing drives.

SanDisk Ultra (256GB): SanDisk's plastic-clad Ultra drive is exactly the kind of thing you'll see drastically overpriced at your local pharmacy or big-box store. Online, it seems fairly reasonable at about $20. But Samsung's Fit Plus is much smaller, performed better in most of our tests, and can be found for $5-10 more in the 256GB capacity.

MORE: Best SSDs

MORE: Best External SSDs and Hard Drives

MORE: How We Test HDDs And SSDs

MORE: All SSD Content

The Nintendo 64 has no shortage of flash carts, considering it was released in 1996. However, the recently released SummerCart64 breaks from the rest by being both fully open-source and having hardware that adds 64DD game compatibility to the cartridge. SummerCart64 is the lovechild of Mateusz Faderewski, aka Polprzewodnikowy on GitHub, and it has a dedicated web page with official documentation and purchase links.

Before that, though, let's talk a bit about what's actually on offer here. For example, why should Nintendo 64DD compatibility matter to anybody in 2024? Well, it was truthfully a limited-life add-on with many canceled projects... but what it did get in its Japan-only release was quite ambitious, and foreshadowed console gaming advancements to come.

So, the Nintendo 64DD got 10 exclusive games or expansions to existing games during its brief run in Japan. Thus, Nintendo 64DD hardware is very much the purview of enthusiasts and game preservationists these days—  but the open-source SummerCart64 helps make those games more accessible than ever. Now, your regular old Nintendo 64 should be able to play those games on authentic hardware.

That's the other side of flash cartridges and console modding, in general. Very often, this hardware has legitimate reasons to exist beyond simply facilitating piracy. For example, a Wii with a broken disc drive (easy if dropped in use) would be unable to play games at all without modding it to run games stored on USB storage.

The easiest way to get your hands on the SummerCart64 is likely through the Phenom Mod store page. If you already have access to a nice 3D printer,  you can print your own shell to save $16, at the time of writing.

So, besides 64DD compatibility (which again, only applies to 10 games), what else makes SummerCart64 so cool? SummerCart64 supports all regions of N64 games with automatic NTSC/PAL switching, and it has seamless integration with PC for homebrew N64 game/app development. It's likely an ideal choice to pair with the likes of Smash Remix and other cutting-edge N64 mods and homebrew. For a 1996 console, the N64 hardware is sure looking lively.

A portable SSD or hard drive is a device that can handle all kinds of storage tasks. It can house huge libraries of files (depending on capacity) and share them between PCs, Macs, tablets, and phones. The best external SSD can also contain full system backup files to restore your system's OS and software in the event of a crash, or it can let you run apps directly from the drive itself. But if you plan on doing the latter, an SSD is a much better choice than a comparatively slow and delicate external hard drive with a read / write head and spinning platters. For related reasons, external SSDs are also much faster than lower-priced external hard drives.

To help you choose the best external storage device for your needs, we test and review dozens of drives as they become available and publish our list of specific recommendations for the best portable SSDs and hard drives below. If you're looking for an internal hard drive for your PC or NAS, head to our list of best hard drives. We also have an expansive list of internal SSDs on our list of the best SSDs. Below, you'll find our list of the best external SSDs and HDDs we've tested.

Best Portable SSD

The SanDisk Pro-G40 is an excellent all-around portable SSD. It has both Thunderbolt 3 and USB modes, so it will work on a wide range of devices, although you may need a Type-C to Type-A adapter or cable. All-around performance is good. Its write performance is exceptional, and your experience in general should be consistently fast. The drive is built on somewhat dated hardware, but this is a mature platform that works excellently for a portable solution. The metal and rubberized casing is another bonusm as it keeps the drive cool while also protecting it against most environmental hazards.

The primary downside to the Pro-G40 is its price. If you don’t need the Thunderbolt functionality you have better options like the Samsung T7 Shield. You can also put together your own portable solution by buying an appropriate enclosure and drive. If you want high-end portability and performance in a rugged package, though, the Pro-G40 is the drive for you.

Read: SanDisk Pro-G40 SSD Review

Best External Hard Drive

If you’re on the hunt for a new external hard drive, WD’s My Passport is an excellent choice. With a solid track record, password protection, and capacities that recently got bumped up to 6TB, it’s prepared to store a lot -- if not all -- of your data and keep it safe.

As street prices have started to fall, it's become a better value than ever. It looks good and comes backed by a plentiful 3-year warranty. To top things off, it boasts top-notch AES 256-bit hardware encryption password protection to keep your content secure from prying eyes.

Read: WD My Passport 5TB Review

We also recently reviewed the 6TB version of the WD My Passport drive. It's a bit thicker and heavier than the 5TB version but otherwise seems to perform about the same. If you need more capacity than lesser models, it's a fine option. But note that there's a $50 price increase for the extra terabyte of capacity, so you'll probably want to ask yourself if you really need that extra space before buying.

Read: WD My Passport 6TB Review

Best USB4 SSD

In the USB4 realm, which is much more widespread than faster Thunderbolt 5, Corsair’s EX400U strikes a great balance between performance and price. Our sequential testing shows it's capable of read speeds above 4,000 MBps, and read speeds over 3,700 MBps. And our sustained Iometer testing showed the drive maintaing a write speed typically between 1,600 and 1,700 MBps for 15 minutes. So it doesn't drastically slow down after its cache has been depleted.

The EX400U certainly isn't cheap at around $189 for the 2TB model we tested, or $349 for the 4TB model. But it's downright affordable compared to the LaCie's Rugged SSD Pro5 Thunderbolt 5 drive, which was selling for $329 for the 2TB model and $529 for the 4TB model when we wrote this. The Corsair SSD also has a magnet on the back that lets it snap onto modern iPhones and some Android phones, making it handy for recording high-res video.

The Corsair drive also ships in a 1TB flavor (around $119), which the LaCie drive lacks. Given Corsair’s drive is much faster and more future-proof than top 20Gbps drives like the Crucial X10 Pro, it makes those older drives hard to argue for, since they still often sell for around the same price as the faster EX400U.

Read: Corsair EX400U (2TB) review

Best SSD for Travel

Sharge Disk Pro may be the most interesting, innovative, and genuinely useful 10 Gbps drives I've ever tested. It's built-in USB / HDMI hub makes it great for travel or on-the-go work (just make sure you carry a USB-C port with a slim connector). And its performance is close enough to the best in its class that you wouldn't notice the difference outside of benchmarks.

For professional tasks, Scharge’s drive also delivered excellent sustained writes, possibly aided by its active cooling, sticking above the 900 MB/s mark for the full 30 minutes of our Iometer benchmark testing. If you need a drive that won’t slow down during massive file transfers or write-heavy workloads, this is one of the best 10 Gbps options I’ve tested. Then there's the magnetic back, which can make using it with compatible phones much more convenient.

Of all the drives I've tested in recent years, if I didn't need anything faster than 10 Gbps, this is the drive I would reach for when traveling. Because my laptop doesn't have enough ports, I don’t have to worry about it slowing down when dumping terabytes of photos, video, or other files onto the drive.

Read: Sharge Disk Pro 2TB review

Best Fast Cross-Platform SSD

LaCie’s Rugged SSD4 occupies an interesting niche. For Windows (and presumably Android) users with USB4, Corsair’s EX400U is nearly as fast in many of our tests, while costing less. And for Mac purists who have the latest hardware, LaCie’s Rugged SSD Pro5 is significantly faster if you have machines with a Thunderbolt 5 port.

But for creators and enthusiasts with production workflows or portable file needs that span multiple platforms, the LaCie Rugged SSD4 delivers the fastest performance we’ve seen across Windows and Macs. We saw sequential speeds around 4,000 MB/s reads and 3,700 MB/s writes in Windows 11, and somewhat slower but still fast reads on a current-gen Mac Min of 3,440 MB/s reads and 3,093 MB/s writes. Add in the three years of included data recovery, and the Rugged SSD4 is easy to recommend for those who need speed and peace of mind while working in Windows, macOS, and maybe Android and iOS as well (though we didn’t test the drive on the latter two platforms).

Read: LaCie Rugged SSD4 4TB Review

Best Rugged Portable Hard Drive

While it is on the pricey side, LaCie’s Rugged RAID Pro isn’t too overpriced considering its market placement and the peace of mind of data redundancy. LaCie includes one month of all Adobe apps for free, a $79.49 (£61) value. More importantly, the drive comes with three years of free data recovery protection. That service can (at times) cost thousands of dollars.

If you are a creative professional in the market for an external HDD, be sure to check this drive out. There aren't many competitors: Most other HDD solutions are much larger, and flash-based SSDs don’t yet offer similarly-priced capacity, nor the same value-adds. The LaCie Rugged RAID Pro 4TB has a unique blend of features and accessories that make it easy to use and quite the versatile travel companion.

Read: Lacie Rugged RAID Pro Review

Most Conveniently Secure Portable SSD

Samsung’s T7 Touch is an innovative portable SSD that blends USB 3.2 Gen 2 performance with convenient AES 256-bit hardware security that’s unlocked by the touch of your fingertip. The built-in fingerprint scanner is the most convenient way to unlock your data that we’ve seen yet. The design is elegant and to a higher standard than your ordinary run-of-the-mill portable drive. The aluminum construction is solid, and various color options are available to suit your unique taste.

That said, G-Technology's recent ArmorLock drive gives Samsung a run for its secure storage money, by using an app and key that's stored on your Android or iOS phone to unlock your drive. It may not be as convenient as swiping your finger across a sensor on your external SSD, but it might just be more secure.

Read: Samsung T7 Touch Portable SSD Review

Best Thunderbolt 5 Portable SSD

The LaCie Rugged SSD Pro 5 combines a tried-and-true rugged design with by far the fastest single-drive speeds we’ve ever seen on an external SSD, when tested on a Thunderbolt 5-equipped Mac.It also ships with five years of Seagate's data recovery service, making it easy to recommend for Mac users – particularly those who only or primarily use current-gen Mac hardware and need the fastest possible performance for media creation or other write-heavy purposes.

Just note that its support on older hardware is so complex that it requires its own compatibility page, and in our testing in Windows 11 over a USB4 / Thunderbolt 4 port, it was slower than recent native USB4 drives. So at least until Thunerbolt 5 becomes more widespread on Macs and PCs, this isn't the best drive for workflows that include Windows PCs.

Read: LaCieRugged SSD Pro5 review

Best Thunderbolt 3 Portable SSD

Driven by an OEM variant of a Samsung 970 EVO and an Alpine Ridge Thunderbolt 3-to-PCIe bridge, Samsung’s X5 is the fastest Thunderbolt 3 portable SSD we’ve tested. Not only will it help speed up your workflow, but it also comes with an additional layer of AES 256-bit hardware-based encryption and password protection for those who need to meet compliance requirements. The three-year warranty is lacking for the professional crowd, and we wish the company offered more color options like those found with the company’s T5 and T7 portable SSDs.

Read: Samsung X5 Portable SSD Review

Drives that didn't make the list

Of course, not every drive we test is the best. But prices change and not everyone has the same storage needs. So if one of these drives goes on a good sale and you like how it performs, it might be worth considering.

We've classified the SK hynix Tube T31 as a flash drive, or SSD stick, since it has a familiar flash drive form factor. But inside, it's actually an m.2 SSD on a small PCB. It's fast by flash drive standards, sports a USB-A port that tops out at 10Gbps, and is only available at 512GB and 1TB capacities. But if those limitations sound good to you, it's well worth considering. It's also on our list of the best flash drives.

Adata's SD810 is a far more traditional external SSD, and one that's surprisingly affordable for a 20Gbps drive. It's great for bursty tasks like moving a few GB of files or perhaps running some light programs. But our testing showed that under sustained speedy writes, its cache can quickly run out, and the drive then writes at hard-drive speeds until the cache can recover. That won't be a problem for everyone. But if you're going to buy a 20Gbps drive, you should probably consider one that can better maintain that speed.

How We Test External SSDs 

Our current testbed consists of an ASRock Z790 Taichi Lite motherboard and a Core i5-12600K CPU. This system gives us native Thunderbolt 4 ports and a USB 3.2 Gen 2 2x2 port via the front header (connected through Corsair's 5000D Airflow case).

We run several tests on the external drives we test, with a mixture of real-world and synthetic benchmarks, and both sequential and small file (4K) reads and writes.

▶️ Trace Testing: PCMark 10 Storage Benchmark

PCMark 10 is a trace-based benchmark that uses a wide-ranging set of real-world traces from popular applications and everyday tasks to measure the performance of storage devices. We report both the proprietary overall score, as well as the bandwidth measured during the test.

▶️ Transfer Rates: DiskBench

We use the DiskBench storage benchmarking tool to test real-world file transfer performance with a custom 50GB dataset. We copy 4,617 files (images, videos, and software ISO files) to a folder on the test drive (write). Then, after leaving the system idle for five minutes, we run the same test in reverse, moving the test folder to a different location on our PCIe 4.0 testing drive. This is a separate drive from our boot drive, to avoid slowdowns due to the operating system background tasks that can be difficult to control for.

▶️ Synthetic Testing: CrystalDiskMark

CrystalDiskMark (CDM) is a free and easy-to-run storage benchmarking tool that SSD companies commonly use to assign product performance specifications. It gives us insight into how each device handles different file sizes. We run this test at its default settings.

The sequential read and write test effectively amounts to a best-case scenario, which often closely aligns with the read /write specs that drive makers put in their specs and on the product box.

The 4K read / write test, meanwhile, is more of a measure of how well the drive handles a plethora of small files, the kind of workload you might encounter when running a program, an OS, or a game directly from the drive.

▶️ Sustained Write Performance: Iometer

A drive's rated write specifications are only a piece of the performance picture. Most external SSDs (just like their internal counterparts) implement a write cache, or a fast area of flash, programmed to perform like faster SLC, that absorbs incoming data.

Sustained write speeds often suffer tremendously when the workload saturates the cache and slips into the "native" TLC or QLC flash. We use Iometer to hammer the SSD with sequential writes for 15 minutes to measure the size of the write cache and performance after the cache is saturated.

This test in particular tends to separate so-called "professional" drives from the rest. Those who often dump massive file libraries onto their external drives, often filling entire drives up in one go, need fast sustained write speeds. More casual storage users, on the other hand, are more likely to write smaller files or libraries of 100GB or less. These kinds of users can still benefit from fast sustained writes, but it may not be as necessary if you rarely or never write enough data to your drive at once from a fast enough source to burn through an SSD's fast SLC cache.

Also tested

Not every drive we test lands on our best external SSD list, but that doesn't mean these alternative drives aren't worth consideration. We recently tested Team Group's PD20, a 20Gbps drive that's available in capacities up to 4TB.

It's an attractive drive with a silver (plastic) and black rubber shell, but its performance in our testing didn't stand out against the competition, particularly when it comes to sustained writes. The drive still performs well enough in most mainstream tasks that many will find its performance just fine. If you like the way it looks, it's worth considering, especially if you can find it on sale.

You can also make your own external SSD

Note that, if you have a spare drive, you can easily make your own portable drive. Dozens of 2.5-inch drive enclosures can be found online for between $10-$25 (£15-25) that will let you drop in an old drive easily, and turn it into an external hard drive or SSD.

And if you have an M.2 drive that you've swapped out of a gaming laptop, ultrabook or upgraded away from in your gaming PC, we've recently looked at NVMe enclosures from MyDigitalSSD and Pluggable. If you have a SATA-based M.2 drive that you'd like to turn into a portable drive, Silverstone's MS09 enclosure lets you do just that. And if you're keen on building your own speedy external SSD but don't have a drive handy to use, the recent WD Blue SN550 is a good candidate for that task. It's only available in capacities up to 1TB, but it's plenty speedy for external storage, and the more spacious model is already selling for as little as $115 at various online outlets.

Just make sure you get an enclosure that matches your drive, be that SATA or NVMe. And also keep in mind that DIY external drives usually aren't sealed, so they're not as likely to stand up to dust and dampness as well as external SSDs and portable hard drives that are designed to do so.

Best External SSD FAQs

When shopping for an external drive or SSD, consider the following:

Finding Discounts on the Best External Storage Drives

Whether you're shopping for one of the best external storage drives or one that didn't quite make our list, you may find savings by checking out the latest Crucial promo codes, Newegg promo codes, Amazon promo codes, Corsair coupon codes, Samsung promo codes or Micro Center coupons.

MORE: Best SSDs

MORE: Best Hard Drive

MORE: How We Test HDDs And SSDs

MORE: All SSD Content

A recently formed company spun off from Lancaster University’s Physics Department, QuInAs Technology has one goal: develop and commercialize UltraRAM, a revolutionary new chip tech. And for the first time, Tom's Hardware was exclusively allowed to see the actual UltraRAM memory chips in a test vehicle at the company's lab at Lancaster University in Lancashire, England.

This potentially disruptive tech is designed to blend the non-volatility of flash storage with faster-than-DRAM speeds. The memory retains data even after power is removed, and the company claims it has at least 4,000X more endurance than NAND and can store data for 1,000+ years. It is also designed to have 1/10th the latency of DRAM and be more energy efficient (by a factor of 100X) than DRAM fabricated on a similar node, drawing the interest of industry heavyweights like Meta. 

The last time we reported on UltraRAM was back in January 2022. Since that time, QuInAs technology has continued to develop, refine, and test the new memory. But over recent months, the biggest changes have come on the business side of things – which is vital for any startup wanting to bring their tech to market.

Lancaster Labs Tour

We were given a tour of the UltraRAM lab to get a first-hand look at the Physics Department’s semiconductor equipment, discuss the various fabrication techniques available at the lab, and discuss the equipment's capabilities and limits. We were also shown a working prototype of the memory riding on a test vehicle.

The best technology available at the lab will help the researchers create UltraRAM devices with features as small as 20nm. Progress to this point will take place over the next few months, and some expensive new testing and verification equipment will arrive shortly to assess UltraRAM scaling efforts.

Another important part of the lab tour was looking at a technology central to UltraRAM’s unique proposition. We looked at the MBE (Molecular Beam Epitaxy) equipment, which precisely deposits the semiconductor layers (GaSb, InAs, and AlSb) required by UltraRAM technology.

The Technology Behind UltraRAM

A triple-barrier resonant tunneling (TBRT) structure is key to UltraRAM’s advances. This highly resistive structure plays a similar role to flash NAND’s oxide layer but, according to the Lancaster University team, can facilitate data storage for 1,000+ years.

UltraRAM is a charge-based memory that uses a floating gate, like flash NAND. Also like flash, the charge state of the floating gate is read non-destructively by measuring the conductance of an underlying ‘channel.’ However, unlike flash, UltraRAM doesn't wear during program and erase cycles because of its TBRT structure. 

This is a major qualifier for the durability claims of 10 million write/erase cycles, and the researchers behind UltraRAM are confident enough to claim future testing is expected to see this durability estimate revised upwards. Meanwhile, mainstream TLC 3D NAND might see gate degradation after a few thousand writes.

Another key claim about UltraRAM is that the floating gate can be switched “extremely quickly and with very little energy,” again these attractive qualities are due to the quantum mechanical phenomenon of resonant tunneling. The benefits are claimed to include UltraRAM’s 100x lower switching energy than DRAM on the same node (1,000x lower than NAND). Moreover, the UltraRAM researchers asserted that the new memory tech is expected to be capable of 1ns write operations, which is about 10x faster than DRAM.

Business Boosters

Two major events have helped QuInAs feel more confident of UltraRAM’s eventual commercial success. Firstly, at the Flash Memory Summit in August, UltraRAM won an award for the “Most Innovative Flash Memory Startup.” We heard that those interested in the future of the technology at the Santa Clara event included Meta (Facebook), who were particularly interested in the UK memory startup's power-saving claims.

Secondly, QuInAs has gained significant financial backing via an ICURe Exploit grant from Innovate UK. The funding win will soon be announced in full by QuInAs, but we know that it was awarded for demonstrating commercial viability and leading-edge science over an intensive six-month program.

With the funding rubber-stamped, the UltraRAM developers are committed to the following:

• To test nanometer scale UltraRAM devices for further proving up of claims regarding performance, efficiency, and durability.
• To cooperate with investors and move towards small volume production.

The wheels of business have certainly been in motion in recent months with the formation of QuInAs. Moreover, earlier this month, a government-sponsored UK Pavilion exhibition at SEMICON Taiwan 2023 was useful for discussions with potential technology and manufacturing partners. We gleaned from our talks with QuInAs that they might find manufacturing partners in Taiwan instead of Europe (IMEC, Belgium).

What Next for UltraRAM and QuInAs Technology?

QuInAs is about to set off on a one-year plan in line with the stated funding goals of both technical and commercial progress.

Better testing machinery is on the way to help with key process scaling and refinement steps. Moreover, IIT Roorkee in India will work as a partner to model UltraRAM performance in wider contexts. This collaboration should help develop the technology and help steer it towards its full potential.

As with all new memory technologies, the challenges of producing the new memory economically in large-scale production will prove to be a key hurdle. On the topic of the first markets for UltraRAM, QuInAs Technology is leaving options open for now, as impending work will shine a clearer light on strengths and any possible weaknesses. It seems that if all goes to plan, the initial small production runs will be targeted at the top of the memory pyramid, as it is the most lucrative segment to address. 

Popular TechTuber Jeff Geerling has delivered an updated take on the old chestnut about the relative merits of pigeon-based vs internet data transfers. With the proliferation of super-fast home connectivity like gigabit fiber, one might expect the carrier pigeon to be blown away in 2023. Spoiler alert: the pigeon with its high-capacity microSD cards won Geerling’s data transfer race by a significant margin. However, as you will learn later, the pigeon gets outpaced at distances over about 600 miles.

Famously, in 2009, a South African company compared the transfer speed of a pigeon carrying a 4 GB memory stick vs local ISP Telkom’s ADSL service. The pigeon delivered its data payload in one hour and eight minutes. An hour was added to physically transfer the 4 GB of data to the computer. Meanwhile, the ADSL service had only managed to send 4% of the same data over a distance of 60 miles.

In Geerling’s 2023 re-race, over a decade later, pigeon evolution hasn’t significantly changed the birds, but data density was significantly higher. The YouTuber managed to kit out the bird with a trio of stripped-down 1TB SanDisk Extreme PRO flash drives (5g each) without overburdening it.

The bird flew for a mile, and it only took about a minute to reach its destination. From this data point and adding the SanDisk flash drive transfer times, Geerling plotted a baseline 3TB transfer graph for the pigeon. It looked like carrier pigeon transfer might be a winning strategy over shorter distances, up to around 600 miles, but the internet connection will triumph at any greater distances.

Geerling wanted to throw another challenger into the mix to spice up the race. So, he donned a pigeon mask and jumped on a plane to carry 3 TB of files from his home in the US to the Canadian data center, which the internet transfer also targeted. In the graphs, you see that PiJeff (jet-assisted Pigeon Jeff) does a little better over longer distances, but will still ultimately be beaten by internet transfers after around the 5,000-mile mark.

To conclude, Geerling says he could have easily done better as PiJeff, stuffing his luggage with very high capacity drives, but wanted to stick to the common 3 TB across all alternatives.

Geering notes in his video that his Gigabit connection doesn’t really live up to its billing, with a best sustained transport speed of about 75 MB/s. Hopefully, another decade later, we will all have broadband measured in petabits, and pigeons won’t have to endure having flash NAND devices strapped to their legs for our amusement (research).

August 25, 2023 Update: We've added new testing for the 2TB PNY XLR8 CS3140 SSD on page 2.

Original Review published June 22, 2022:

PNY has been around for a long time, known in particular for its flash-based products, but also for its memory and GPUs. The company's new XLR8 CS3140 SSD tops out at 7,500/6,850 MBps of sequential read/write throughput for the 2TB and larger models, and it's also flexible in that it has both a version with a heatsink and one without, and you can also use a separate PlayStation 5 heatsink for the latter. 

Last year, PNY reduced the endurance rating on its CS3030 SSD by almost 80% due to the impact of Chia cryptomining, which was known for killing SSDs with heavy write workloads. But PNY has now returned to high endurance ratings with the XLR8 CS3140.

We tested the new 1TB XLR8 CS3140 model, which is a good starting capacity, but the 2TB model offers better peak performance because it is packed with more flash dies. Naturally, the 4TB is even more compelling if you're after more raw capacity. However, a basic primary drive is quite fine at 1TB, especially when coupled with strong hardware and a good SLC caching scheme like we see with the previous-gen CS3140. Let's see how the PNY CS3140 fares in our test suite. 

Specifications

The PNY CS3140 is available in the 1TB, 2TB, and 4TB capacities. These all have a five-year warranty backed up by 700TBW, 1400TBW, and 3000TBW of write endurance, respectively. Performance tops out at 7,500/6,850 MBps of sequential read/write throughput for the 2TB and larger models.

PNY doesn't officially list the IOPS measurements, but we can guess that this drive would be similar to other SSDs that have the same hardware combination. We had to dig up the endurance numbers on PNY's Taiwanese site, and the company seems to operate on the principle of “the less the information, the better.” That does make some sense in an era where SSD manufacturers often change the SSD's hardware after launch, but you'll need to do your homework to find concrete specs for these drives.  

The drive supports 256-bit AES encryption. PNY sells this drive with and without a heatsink, and we are testing the version without. This makes it a good choice for the PlayStation 5, possibly as-is, but PNY sells a separate heatsink for SSDs that is designed especially for the PS5. 

Software and Accessories

The CS3140 doesn't come with anything special. PNY does offer a download for its PCIe SSD toolbox software for firmware updating on their site. This software also gives some basic information about the drive, such as SMART readings, and allows for secure erasing.

A Closer Look

The drive is quite plain, sporting a rear label with basic details about the drive. Under the top label, we see the controller, DRAM, and four NAND packages in a standard configuration.

The flash is labeled IA7BG94AYA, which is 176-layer Micron TLC, a.k.a. B47R. This flash has proven excellent when coupled with the Phison E18 controller. As there are four packages, each should have four 64GB flash dies in a QDP configuration.

Based on data pulled from the drive, this flash is standard B47R running at 1200 MT/s. We’ve seen faster, but this is sufficient with two dies per channel. The endurance and quality of the flash look good, so there are no surprises here.

The Phison E18 controller needs no introduction - it’s been used on a wide variety of drives to good effect. It tends to be optimized for peak performance, specifically in sequential work, and it’s particularly good when paired with newer flash.

The DRAM is SK hynix’s H5AN8G6NCJR-VKC, which is 1GB of DDR4 in the 512M x 16b configuration. This would be C-die DRAM with the normal DRAM-to-NAND ratio, more than enough for any user. We can also spot the Phison power management package (PMIC) near the controller.

MORE: Best SSDs

MORE: Best SSD for the Steam Deck

MORE: Best Hard Drives

MORE: Best External SSDs

Comparison Products

There are some new drives out on the market that weren’t here when we reviewed the 1TB CS3140, including PCIe 5.0 SSDs like the Crucial T700 and Corsair MP700. We also have reviewed some great 4.0 SSDs like the Samsung 990 Pro, the Solidigm P44 Pro/SK hynix Platinum P41, and the WD Black SN850X. There’s also the Sabrent Rocket 4 Plus-G and Adata Legend 960 Max. Lastly, we have the budget-oriented Solidigm P41 Plus. Keep in mind that we got the PS5 heatsink with the 2TB CS3140 which makes it great for the PS5 even if the drive is dated in standard benchmarks.

A Closer Look

Trace Testing - 3DMark Storage Benchmark

Built for gamers, 3DMark’s Storage Benchmark focuses on real-world gaming performance. Each round in this benchmark stresses storage based on gaming activities including loading games, saving progress, installing game files, and recording gameplay video streams.

The CS3140 is below average in 3DMark. It scored slightly worse than the 1TB version.

Trace Testing – PCMark 10 Storage Benchmark

PCMark 10 is a trace-based benchmark that uses a wide-ranging set of real-world traces from popular applications and everyday tasks to measure the performance of storage devices.

The CS3140 is near the bottom in PCMark 10, too, faring worse than the 1TB version.

Transfer Rates – DiskBench

We use the DiskBench storage benchmarking tool to test file transfer performance with a custom, 50GB dataset. We copy 31,227 files of various types, such as pictures, PDFs, and videos to a new folder and then follow-up with a reading test of a newly-written 6.5GB zip file.

Things aren’t quite as bad with DiskBench. The CS3140 is close enough to the other high-end PCIe 4.0 SSDs. The 5.0 SSDs - the T700 and MP700 - pull away due to their faster interface, while the slower P41 Plus is at the bottom.

Synthetic Testing - ATTO / CrystalDiskMark

ATTO and CrystalDiskMark (CDM) are free and easy-to-use storage benchmarking tools that SSD vendors commonly use to assign performance specifications to their products. Both of these tools give us insight into how each device handles different file sizes.

The CS3140’s sequential performance in ATTO and CDM is adequate but not on the level of the better PCIe 4.0 SSDs. This is fine as sequential performance at this level isn’t super important. Random write latency is good — surprisingly good even, but random read latency is fairly bad and significantly worse than the 1TB version of this drive.

The reason for that is obvious; PNY switched to 112-Layer Kioxia TLC (BiCS5) which does not perform as well with this controller as the original 176-Layer Micron TLC (B47R). However, if you’re buying a drive for capacity and especially for the PS5, this is not a big deal.

Sustained Write Performance and Cache Recovery

Official write specifications are only part of the performance picture. Most SSDs implement a write cache, which is a fast area of (usually) pseudo-SLC programmed flash that absorbs incoming data. Sustained write speeds can suffer tremendously once the workload spills outside of the cache and into the "native" TLC or QLC flash.

We use Iometer to hammer the SSD with sequential writes for 15 minutes to measure both the size of the write cache and performance after the cache is saturated. We also monitor cache recovery via multiple idle rounds.

The 2TB CS3140 writes in pSLC mode for over 30 seconds at up to 6.2 GB/s. This makes for a cache over 200GB, which is fairly large. Luckily, the drive does not easily hit a folding state, and instead drops to around 1.6 GB/s in TLC mode. This is not a bad result in any way. The 1TB version with Micron TLC is faster in TLC mode and it would probably be even faster with more dies at 2TB. Still, the BiCS5 2TB CS3140 is fairly consistent. Its recovery response has it lingering in TLC mode, which is luckily not too bad.

Power Consumption and Temperature

We use the Quarch HD Programmable Power Module to gain a deeper understanding of power characteristics. Idle power consumption is an important aspect to consider, especially if you're looking for a laptop upgrade as even the best ultrabooks can have mediocre storage.

Some SSDs can consume watts of power at idle while better-suited ones sip just milliwatts. Average workload power consumption and max consumption are two other aspects of power consumption, but performance-per-watt is more important. A drive might consume more power during any given workload, but accomplishing a task faster allows the drive to drop into an idle state more quickly, ultimately saving energy.

For temperature recording we currently poll the drive’s primary composite sensor during testing with a 24C ambient.

The 2TB CS3140 with BiCS5 flash is not very efficient, especially compared to similar drives using Micron’s TLC instead. This is to be expected as this flash is less efficient in general and does not seem to pair quite as well with Phison’s E18 controller. This result isn’t crtical if this drive is used in a high-end desktop or PS5, but it does demonstrate the downsides of swapping the flash. It’s worth pointing out that other manufacturers can and have changed hardware over the course of an SSD’s lifespan.

We only reached a brief maximum of 67C during testing, which is far away from significant throttling. This drive would be quite happy in a PS5 with the optional heatsink. We would recommend it for that or perhaps some extra fast storage if the drive is priced right. Right now, at the time of the review update, the 2TB SN850X is probably the better option.

Test Bench and Testing Notes

We use an Alder Lake platform with most background applications such as indexing, Windows updates, and anti-virus disabled in the OS to reduce run-to-run variability. Each SSD is prefilled to 50% capacity and tested as a secondary device. Unless noted, we use active cooling for all SSDs.

MORE: Best SSDs

MORE: Best SSD for the Steam Deck

MORE: Best Hard Drives

MORE: Best External SSDs

PNY XLR8 CS3140 1TB SSD

RATING: ★★★★ (4 Stars)

PROS

+ Consistent performance
+ Strong peak and all-around performance
+ Power-efficient
+ Good PlayStation 5 option from PNY

CONS

- Pricing

OUR VERDICT
The 1TB PNY CS3140 is a solid drive that can be used in a PC or in the PlayStation 5, but the latter is best with PNY’s separate PS5 heatsink. Performance is strong, but there’s not much special about this drive versus the competition.

Comparison Products (1TB)

We compared the PNY CS3140 to other PCIe 4.0 drives, including the more budget-oriented WD Black SN770. Contenders using proprietary controllers include the WD Black SN850, the Samsung 980 Pro, and the Crucial P5 Plus. The other three drives use the same Phison E18 controller, but the Gigabyte Aorus Gen4 7000s and Corsair MP600 Pro use the older 96-layer Micron TLC while the last one, the Kingston Fury Renegade, is equipped with the same 176-layer flash as the CS3140.

Trace Testing - 3DMark Storage Benchmark

Built for gamers, 3DMark’s Storage Benchmark focuses on real-world gaming performance. Each round in this benchmark stresses storage based on gaming activities including loading games, saving progress, installing game files, and recording gameplay video streams.

3DMark shows off WD’s excellent optimization, but thanks to their upgraded flash, we also had respectable results for the CS3140 and Fury Renegade.

Trace Testing – PCMark 10 Storage Benchmark

PCMark 10 is a trace-based benchmark that uses a wide-ranging set of real-world traces from popular applications and everyday tasks to measure the performance of storage devices.

Again we see strong optimization from WD and Crucial, with the newer flash helping out the CS3140 and Fury Renegade. PCMark is often considered a “real world” benchmark, and we can look particularly at the latency results to see that the older 96-layer flash on the 7000s and MP600 Pro is appreciably slower.

Transfer Rates – DiskBench

We use the DiskBench storage benchmarking tool to test file transfer performance with a custom, 50GB dataset. We copy 31,227 files of various types, such as pictures, PDFs, and videos to a new folder and then follow-up with a reading test of a newly-written 6.5GB zip file.

This bandwidth-heavy test favors the high-end PCIe 4.0 drives, particularly those based on the Phison E18 controller. The newer 176-layer flash also brings improvements, particularly as we are testing at 1TB rather than 2TB of capacity. The Phison E18 can peak with 32 dies of flash, and considering these use 64GB dies, that means 2TB. The 176-layer flash is also faster per die than the 96-layer flash for copying tasks.

Synthetic Testing - ATTO / CrystalDiskMark

ATTO and CrystalDiskMark (CDM) are free and easy-to-use storage benchmarking tools that SSD vendors commonly use to assign performance specifications to their products. Both of these tools give us insight into how each device handles different file sizes.

The queue depth 1 (QD1) sequential read and write results from ATTO paint a simple picture: drives using the Phison E18 come out on top, with a significant edge given to those with the newer 176-layer flash. To some extent, there is a “wall” due to the limits of the x4 PCIe 4.0 interface, as seen with queue depth 8 sequential reads in CrystalDiskMark. The write result shows a bit more delineation between the tested drives. At this point, we see a theme, with the CS3140 underperforming the Fury Renegade, possibly due to inferior firmware optimization.

Random QD1 performance is good to excellent on the CS3140, where we again see the newer flash pulling ahead. Again, no real surprises, which is not a bad thing — this hardware configuration performs very well on multiple fronts.

Sustained Write Performance and Cache Recovery

Official write specifications are only part of the performance picture. Most SSDs implement a write cache, which is a fast area of (usually) pseudo-SLC programmed flash that absorbs incoming data.  Sustained write speeds can suffer tremendously once the workload spills outside of the cache and into the "native" TLC or QLC flash. We use Iometer to hammer the SSD with sequential writes for 15 minutes to measure both the size of the write cache and performance after the cache is saturated. We also monitor cache recovery via multiple idle rounds.

The CS3140 writes at above 5.8 GBps for almost 20 seconds, revealing an SLC cache of around 116GB. This is not a particularly large cache but should be more than sufficient for a 1TB SSD. The drive then hits its direct-to-TLC phase at up to 2 GBps, which is quite fast for a 1TB drive. This reflects the choice of having a smaller SLC cache while using the faster 176-layer TLC. Performance would be higher at 2TB due to additional interleaving. As the cache is dynamic, it will shrink as the drive is filled. It’s still possible to hit a slower state with sufficient writing, but this is difficult to achieve.

Beyond flash choice, the size of the cache is one thing that separates drives based on the E18 controller. As we can see here with the Fury Renegade’s results, it has a larger, faster cache than the CS3140, but no middle-ground TLC state — this allows some flexibility for the manufacturer and the user. While cache recovery remained slow on both drives, the CS3140's ability to maintain TLC speeds ensures it has better steady state performance for edge cases.

Which is better is arguable, particularly as high-end PCIe 4.0 drives tend to focus on raw bandwidth. However, we believe in the future that sustained performance may become more important. For example, if you’re able to stream assets steadily from a drive via DirectStorage. That would be reads and not writes, of course, but there are nevertheless cases where consistency trumps peak performance alone.

Power Consumption and Temperature

We use the Quarch HD Programmable Power Module to gain a deeper understanding of power characteristics. Idle power consumption is an important aspect to consider, especially if you're looking for a laptop upgrade as even the best ultrabooks can have mediocre storage.

Some SSDs can consume watts of power at idle while better-suited ones sip just milliwatts. Average workload power consumption and max consumption are two other aspects of power consumption, but performance-per-watt is more important. A drive might consume more power during any given workload, but accomplishing a task faster allows the drive to drop into an idle state more quickly, ultimately saving energy.

We also monitor the drive’s temperature via the S.M.A.R.T. data and an IR thermometer to see when (or if) thermal throttling kicks in and how it impacts performance. Remember that results will vary based on the workload and ambient air temperature

Efficiency remains strong on the CS3140, thanks to the flash and the ability to quickly finish workloads with high peak performance. It still trails the Fury Renegade by a small bit.

According to both SMART and the IR gun, the CS3140 idles in the mid-40Cs and peaks in the upper-70Cs. This is hot but not unexpected as the drive lacks a heatsink. The smaller SLC cache helps when the drive hits the TLC flash, but sustained writes still get it pretty close to the throttling mark. In fact, we did see light throttling after about 215GB of writes when the drive hit 78C. Therefore, we recommend a heatsink if you anticipate having heavier workloads and/or a hotter environment.

Test Bench and Testing Notes

We use a Rocket Lake platform with most background applications such as indexing, windows updates, and anti-virus disabled in the OS to reduce run-to-run variability. Each SSD is prefilled to 50% capacity and tested as a secondary device. Unless noted, we use active cooling for all SSDs.

Conclusion

We didn't expect any surprises, and we didn't find any — the PNY XLR8 CS3140 gets close to our performance expectations without any weirdness. As with similar drives, the peak and general performance ranges from good to excellent. The SLC caching scheme is conservative, making for consistent performance in sustained workloads, and the new flash ensures good power efficiency. The PNY CS3140 is also a strong choice for a PlayStation 5 drive due to its option for a tailored heatsink.

On the other hand, there’s not much to set this drive apart from the competition. It is nice to see a functioning SSD toolbox — this is something lacking with, say, Inland drives — but the pricing isn’t particularly competitive. That being said, this is an adequate alternative and would be a good choice if you can find it on sale. PNY isn't pulling any endurance shenanigans here as far as we could tell, even if the company is a bit shy about some of the drive's specifications.

We scored the Kingston Fury Renegade a bit higher. The Fury Renegade edges out the CS3140 in almost every test while coming with cloning software, although currently, it is more expensive. The Kingston SSD is also designed for the PlayStation 5 and is perhaps a more trusted brand, perhaps except for those users who remember the ancient NAND switch-a-roo.

We also previously reviewed the Silicon Power XS70, a similar drive that's also designed for the PlayStation 5. It costs less than the CS3140 and comes with a PS5-capable heatsink out of the box. All of this makes it hard to give the PNY XLR8 CS3140 a higher score, but it is still a strong drive and contender. The biggest challenge for the XLR8 CS3140 is that it's in an increasingly crowded market, but it is a competent alternative.

MORE: Best SSDs

MORE: Best SSD for the Steam Deck

MORE: Best Hard Drives

MORE: Best External SSDs

The Samsung 990 Pro, one of the best SSDs, was launched last year in 1TB and 2TB flavors. Making good on its promise to release a 4TB version, Samsung has confirmed on X (formerly Twitter) that the speedy PCIe 4.0 SSD will hit the market soon.

Samsung's position on PCIe 5.0 SSDs is a bit complex. While the company has some enterprise drives that leverage the PCIe 5.0 interface, it has yet to launch a consumer PCIe 5.0 SSD. The Samsung 990 Pro, an M.2 2280 drive, is still on the PCIe 4.0 interface and is currently the brand's fastest consumer drive. The company has followed up with a 4TB variant, doubling the Samsung 990 Pro's 2TB capacity. Samsung doesn't always offer the 4TB capacity with all its SSDs. For example, the Samsung 980 Pro, the predecessor to the Samsung 990 Pro, maxed out at 2TB.

The Samsung 990 Pro 4TB will arrive in a bare drive (MZ-V9P4T0BW) and heatsink (MZ-V9P4T0CW) versions. The SSD's form factor and design allow it to live inside many devices, including laptops, desktops, and consoles like the PlayStation 5. Therefore, the Samsung 990 Pro 4TB isn't just an SSD for PC consumers. It's a drive that PlayStation 5 owners can look forward to due to its ample capacity and blistering performance.

While the manufacturer hasn't listed the Samsung 990 Pro 4TB on its website, the drive is already mentioned in the product brochure. Similar to the other capacities, the new Samsung 990 Pro 4TB utilizes Samsung's in-house Pascal SSD controller, which features an Arm design. The company produces Pascal on the 8nm process node. As for the NAND, the drive employs the same Samsung 176-layer TLC NAND as the other drives. The difference is that the die sizes are larger to hit 4TB.

Samsung 990 Pro Specifications

The Samsung 990 Pro 4TB SSD's advertised performance mirrors that of the lower capacity 2TB version. The drive delivers sequential read and write speeds, hitting 7,450 MB/s and 6,900 MB/s, respectively. Meanwhile, the random performance is at 1,400,000 IOPS reads and 1,550,000 IOPS writes. The figures may look low compared to the PCIe 5.0 standard, but the Samsung 990 Pro can hang with some PCIe 5.0 drives on the market.

Besides the capacity, the difference between the 4TB and 2TB models is the size of the cache memory and the endurance. The Samsung 990 Pro 4TB has a 4GB LPDDR4 DRAM cache and a 2,400 TBW rating, twice that of the Samsung 990 Pro 2TB. As expected, Samsung backs the Samsung 990 Pro 4TB with a limited five-year warranty.

Pricing on the Samsung 990 Pro has improved significantly over the last few months. The 2TB bare drive had a $289.99 MSRP but retails for as low as $169.99 nowadays. The heatsink versions are typically more expensive than the bare drive models and have identical performance. The MSRP for the Samsung 990 Pro 4TB is unknown, but it shouldn't remain a mystery for long since Samsung has already started teasing the upcoming SSD.

REGISTER NOW and watch on demand. 

The explosion of AI is further heightening demand for storage performance and capacity as organizations feed models and databases with unprecedented amounts of data, meaning the next generation of storage technologies will need to deliver even greater performance, density and capacity than ever before.

Supermicro's fourth annual Open Storage Summit brings together leading storage experts from across the industry including drive manufacturers, compute components manufacturers, software developers and of course Supermicro's industry leading system architects to discuss the latest in storage technologies and how they will solve tomorrow's data challenges from the data center right out to the intelligent edge.

This year’s Summit includes a roundtable keynote session followed by five focus sessions, with guests from the storage industry's leading players including Intel®, AMD, NVIDIA, Micron, Kioxia, Solidigm, and Samsung, as well as Supermicro’s storage software partners. 

New Innovations For Storage Performance

In a time in which pure processing power is game-changing, it’s important to continually reflect on current solutions and look for new ways to keep business players progressing through new levels.  Sometimes, that progress means stopping investment in an old version of that game and crafting a whole new open world instead.

In Session 2 of our 2023 Open Storage Summit, you will hear from NVIDIA on how they are helping organizations build whole new worlds in which to operate. Through the introduction of the third pillar of computing - the Data Processing Unit - DPUs join CPUs and GPUs to create a futuristic blue sky environment in which applications are accelerated well beyond the capabilities of CPUs alone.

This is particularly important in the frenetically growing AI market, in which lightning-fast storage processing time means that critical business initiatives make their way to the leaderboard instead of being relegated to game-over status. 

During this session, players in the audience will: 

• Discover the limitations inherent in traditional storage architectures
• Understand the advantages of GPUDirect storage and RDMA for AI
• Learn how the GPUDirect Storage and RDMA work at the rack-level to combine the resources of multiple systems into one massive compute cluster
• Uplevel their knowledge around how DPUs can effectively offload compute tasks to massively improve storage performance

Watch all of the webinars on demand

Register now for full access to the storage industry's leading online event to get the latest on key storage trends as well as exclusive look into the future of high performance storage from the most influential minds in the industry.

 Join the discussion here! 

Have you ever wondered where defective NAND chips or those that don't pass the requirements go? Not inside the best SSDs, that's for sure. However, you'll likely find them inside knock-off SSD brands that sell for ridiculously low prices on platforms like AliExpress.

NAND yield rates are typically bad until the process node matures. Often, factories come across NAND that doesn't meet the client's requirements or doesn't pass quality control. Instead of throwing the NAND away, which increases electronic waste, the factories recycle them into cheaper products or sell them to local companies, giving rise to obscure SSD brands on the Chinese market.

A Chinese data recovery YouTuber recently took apart a ShineDisk M667 (M667-120G), a 120GB SATA SSD that retails for around $13 in China, and found some alleged SK hynix-manufactured NAND for Apple devices inside. Mind you; this is a knock-off brand and not to be confused with the same SanDisk that Western Digital acquired. 

The SSD controller is the only thing that looks legit on the ShineDisk M667. The drive utilizes the SM2258XT, a DRAM-less SATA III from Silicon Motion. It’s an outdated SSD controller that powers SSDs, such as the Crucial BX500, TeamGroup EX2, and WD Blue 3D.

The NAND chips in the ShineDisk M667 are thicker than the NAND that you typically find on a regular SSD. According to the YouTuber, it’s reportedly SK hynix’s E2NAND, which appears to be a multi-chip package with an MSP controller with embedded ECC support. We couldn’t find public information on E2NAND on SK hynix’s website. However, we did see many mentions of E2NAND associated with Apple devices on various Chinese e-commerce platforms.

The origin of the NAND is unknown. It could have come from a factory making NAND for Apple or some third-party vendor ripping them from defective iPhones. However, the  YouTuber suspects the former since the quality appears to have a level of professionalism.

Reprocessing components for lower-tier products or new products is a regular practice in the hardware world. Companies are always looking to maximize profits. For example, a recent report revealed that some vendors recycle server ICs into consumer memory. The problem is with reliability. 

In the case of the ShineDisk M667 SSD, we have no idea where the NAND came from. It wouldn’t surprise anyone if, one day, the ShineDisk M667 died out of nowhere. That’s the risk of an off-brand SSD that sells for $13 in a market filled with knock-offs. You’re better off looking for one of the best SSD deals on a name-brand drive.

If there’s one thing that can be guaranteed, it’s that Teamgroup will put out a drive to fit any niche, and the MP44S is no exception. It comes as no surprise that Teamgroup wanted to jump on the M.2 2230 bandwagon, given the growing popularity of the Steam Deck and ROG Ally. The portable gaming PC revolution has begun in earnest, and Teamgroup has been eager to put out an SSD to help that along, one that could fulfill even the desire for up to 2TB of tiny - but fast - internal storage. The MP44S may not be the first to market, and it is just one of many. Still, it’s another fair alternative that assists in bringing down prices for this increasingly desirable form factor, especially at the most-popular 1TB capacity point.

It’s an opportunity because the hardware on many of these drives is no different than what’s on popular M.2 2280 drives. Of course, these 2230 drives usually have to be single-sided to fit nicely in portable systems, which adds physical limitations to the drive’s design. You typically won’t find DRAM on an M.2 2230 SSD, except in the rare case that the components are embedded and the drive will be limited to a single NAND package. This puts some constraints on drive capacity, but QLC makes even 2TB a reasonable option, and honestly, the performance for the Steam Deck is pretty good, even with a QLC SSD. This makes drives like the MP44S a pretty solid upgrade, especially if you saved some money going with a 64GB base Steam Deck.

Teamgroup doesn’t compromise the MP44S's controller, using the fast Phison E21T. This controller has been at the heart of many of the best budget PCIe 4.0 SSDs on the market. The Steam Deck is restricted to the PCIe 3.0 interface, but the ROG Ally and other devices can use the additional bandwidth gifted by the 4.0 interface.

Future DRAM-less drives, including those with TLC, might pull ahead of a drive like the MP44S, but it’s a sensible option for now. SD cards are fine and dandy, but this hardware is of a higher class — something to keep in mind as the ROG Ally has had SD card issues. Portable storage also works, but you don’t always want to drag that around with you. So grab an installation guide, take your time, and enjoy a real SSD like the MP44S.

Specifications

The Teamgroup MP44S is available at 1TB and 2TB, currently for $94.99 and $199.99, respectively. This is slightly more expensive than some direct competitors like the Addlink S91. QLC isn’t quite as compelling at 1TB, but it can be workable if it undercuts the TLC alternatives by enough.

The MP44S can hit up to 5,000 / 3,500 MB/s for sequential reads and writes and up to 480K / 500K IOPS for random reads and writes. There’s a five-year warranty with 250TB and 450TB of writes for the two capacities offered. Teamgroup was unable to give us specific performance information for the 1TB, but given what we see with similar drives, it should be close to the 2TB.

Software and Accessories

Teamgroup has its own SSD SMART monitoring software, or SSD SMART Tool, available for download on its website. It comes with a manual that indicates the software can be used to grab basic drive information, such as the SMART readings, and the application also includes a built-in CrystalDiskMark-like benchmark.

A Closer Look

There’s not too much to the MP44S. It’s a single-sided SSD in the M.2 2230 form factor, short and slender enough to fit into the Steam Deck, ASUS ROG Ally, and other portable systems. Teamgroup uses a graphene label to cover the controller and a single NAND package for heat dissipation.

This standard layout is also used on the Sabrent Rocket Q4 2230 and Addlink S91. To reach 2TB, the MP44S has to use Micron’s 176-Layer QLC (N48R), but it’s also used at 1TB. It is technically possible to use TLC flash and achieve 2TB — WD’s SN740 manages the trick — but it’s probably easier and more cost-effective to use QLC. Intel’s QLC is only found in M.2 2230 with the 1TB P41 Plus OEM drive and is, therefore, harder to compare. QLC is more of a compromise at 1TB as there’s not as much interleaving with the dies, but for this level of drive with the E21T, it’s sufficient and is mostly an option to save a few bucks over TLC.

Micron’s flash is also more efficient than WD’s BiCS5 TLC from our testing, and the 2TB SN740 — which can be gauged to some degree by looking at our SN770 review — loses some performance over its 1TB counterpart. Still, the 2TB SN740 is probably the fastest drive at 2TB, but we think that, at least for the Deck, QLC is a perfectly good compromise there. This does open the door to the 2TB OEM Crucial 2400, although that uses the SMI SM2269XT controller instead of the MP44S’s Phison E21T. In our testing of these controllers, the latter performs as well or better with better power efficiency. The competition at 1TB is much stiffer for QLC.

It’s worth looking into what performance compromises are made with QLC at 1TB. The MP44S and S91 offer this option, and the Rocket Q4 2230 may eventually do so, but there are many TLC alternatives that no longer carry a high price premium. On the other hand, requiring only eight dies of QLC for 1TB is more convenient, and if these drives end up being less expensive, they are probably still fast enough to be a worthwhile alternative.

MORE: Best SSDs

MORE: Best External SSDs and Hard Drives

MORE: How We Test HDDs And SSDs

MORE: All SSD Content

Comparison Products

The Teamgroup MP44S goes up against other M.2 2230 SSDs that we have tested, including the Rocket 2230, the Corsair MP600 Mini, the older Inland TN436, the newer Inland TN446, and the similar Sabrent Rocket Q4 2230 and Addlink S91. The WD SN740, not shown, is a shorter OEM version of the WD Black SN770. The Micron 2400, also not shown, is similar to the Crucial P3 Plus but with an SMI controller — it should be somewhat less efficient. These two drives may be referenced where applicable but are not explicitly reviewed.

Steam Deck - General

To gauge general Steam Deck performance, we engage in a variety of typical storage situations with active time measurement. These activities include dealing with the default SteamOS through initial setup, re-imaging, initial booting, and booting in general. For game testing, we see how long it takes to install and boot the popular title Hollow Knight. For a full explanation of our testing procedure, see our Steam Deck storage testing article.

No anomalies were spotted with the MP44S in general Deck testing. You will get good performance from this drive when engaging in basic operations, and it works perfectly well with the Deck. This helps confirm that QLC, even at 1TB, is perfectly capable, although the drives must be less expensive. QLC still has its weaknesses, and keeping more space free to mitigate these means adjusting capacity per dollar must be made when picking a drive.

Steam Deck - KDiskMark and Temperature

One of the most popular, if not the most popular, storage benchmarks is CrysalDiskMark (CDM), which we use in our own Windows-based SSD testing suite. This benchmark relies on Microsoft’s DiskSpd with templated test settings. CDM lets you quickly see how a drive performs against its idealized, rated specifications and the benchmark can also hint at a more “real world” feel with low queue depth I/O testing.

The Linux flavor of this benchmark, which is easy to install on the Steam Deck, is KDiskMark. The “K” refers to the KDE desktop environment used on the Deck. KDiskMark relies on the Flexible I/O tester, or FIO, instead of DiskSpd. For the temperature portion we use data from the SMART sensors on the drive which can be directly polled and from which the maximum temperature can be extracted.

The MP44S is on the level with the other QLC drives, like the Rocket Q4 2230 and S91, which keeps it from being as fast as the newer TLC ones in the all-important low queue depth random 4KB read metric. It is faster than the older TN436 and the Deck’s 64GB stock drive, though.

Other disparities will differ when the drives are utilized in a PCIe 4.0 device like the ROG Ally. This is not hugely relevant for real-world performance and feel, but there is a difference, and certain edge cases — like sustained writes with a fuller drive — will favor the TLC options. The MP44S doesn’t appear fazed, with only 1TB compared to the 2TB QLC drives.

The MP44S does not run particularly hot or cold. Teamgroup does market the graphene label for its cooling properties, which may be of small benefit, but also may require EMI shield adjustment in the Deck.

PC Trace Testing - 3DMark Gaming Storage Benchmark

Built for gamers, 3DMark’s Storage Benchmark focuses on real-world gaming performance. Each round in this benchmark stresses storage based on gaming activities including loading games, saving progress, installing game files, and recording gameplay video streams.

3DMark performance is fair, but the QLC-based solutions like the MP44S are slightly slower. 

PC Trace Testing – PCMark 10 Storage Benchmark

PCMark 10 is a trace-based benchmark that uses a wide-ranging set of real-world traces from popular applications and everyday tasks to measure the performance of storage devices.

PCMark 10 is a surprise with the QLC drives — the MP44S, the Rocket Q4 2230, and the S91 — coming out on top. This is true regardless of interface speed or capacity. One place Micron’s QLC shines is with low queue depth 4KB writes, as demonstrated by our CrystalDiskMark results below.

PC Transfer Rates – DiskBench

We use the DiskBench storage benchmarking tool to test file transfer performance with a custom, 50GB dataset. We copy 31,227 files of various types, such as pictures, PDFs, and videos to a new folder and then follow-up with a reading test of a newly-written 6.5GB zip file.

FIle copy performance is good for a PCIe 3.0 interface, as would be the case on the Steam Deck. When the drives are opened up to the PCIe 4.0 interface, the TLC-based solutions like the Rocket 2230, the MP600 Mini, and the TN446 pull ahead.

PC Synthetic Testing - ATTO / CrystalDiskMark

ATTO and CrystalDiskMark (CDM) are free and easy-to-use storage benchmarking tools that SSD vendors commonly use to assign performance specifications to their products. Both of these tools give us insight into how each device handles different file sizes.

The ATTO results are good both for PCIe 3.0 and 4.0. Write throughput drops a bit with larger block sizes for the latter due to the limitations of QLC flash at the rated bus speed. Likewise, sequential performance in CDM is good at PCIe 3.0 speeds, with a significant gap for sequential writes with PCIe 4.0. Random write latency is good with this flash, but the arguably more important random read latency is worse with QLC.

PC Sustained Write Performance and Cache Recovery

Official write specifications are only part of the performance picture. Most SSDs implement a write cache, which is a fast area of (usually) pseudo-SLC programmed flash that absorbs incoming data.  Sustained write speeds can suffer tremendously once the workload spills outside of the cache and into the "native" TLC or QLC flash. 

We use Iometer to hammer the SSD with sequential writes for 15 minutes to measure both the size of the write cache and performance after the cache is saturated. We also monitor cache recovery via multiple idle rounds.

It’s hard to be satisfied by the slow, steady state write performance of the QLC-based drives, which includes the MP44S. They ultimately write at only about 100 MB/s in the QLC folding mode. This is due to the large cache and the flash’s low raw writing speed. Normally, this performance drop is not noticed as the drives will stay in pSLC mode. However, prolonged writing and/or a very full drive can make this condition arise more readily. The good news is that this QLC is fast enough to saturate with just eight dies for 1TB of capacity, although the cache is correspondingly smaller in absolute terms.

This is certainly one reason to look for a TLC drive, but even that is no guarantee of consistent performance, as demonstrated by the TN436. It’s a good idea to get more capacity than you need and try to leave some space free on the drive, especially with QLC. That means that going with, say, a 512GB TLC Rocket 2230 might be an attractive alternative to 1TB of QLC, but pricing is such that the best capacity per dollar is currently fixed at 1TB.

Power Consumption

We use the Quarch HD Programmable Power Module to gain a deeper understanding of power characteristics. Idle power consumption is an important aspect to consider, especially if you're looking for a laptop upgrade as even the best ultrabooks can have mediocre storage.

Some SSDs can consume watts of power at idle while better-suited ones sip just milliwatts. Average workload power consumption and max consumption are two other aspects of power consumption, but performance-per-watt is more important. A drive might consume more power during any given workload, but accomplishing a task faster allows the drive to drop into an idle state more quickly, ultimately saving energy.

For temperature recording we currently poll the drive’s primary composite sensor during testing with a 24C ambient.

Power efficiency is pretty stellar, although not quite as good as the best TLC drives. All of these drives are more efficient when run at PCIe 3.0 than 4.0, albeit not by a huge amount. They all also easily surpass the older TN436.

The 1TB MP44S has fewer dies than the 2TB S91 and Rocket Q4 2230 but has similar performance, which means it is more efficient  — almost on par with the TLC drives. This indicates that any decision between QLC and TLC at 1TB will not have power efficiency as a significant factor.

The MP44S also ran cooler than our 2TB QLC drives, exhibiting no throttling. Efficiency often correlates with temperature, all else being equal. We wouldn’t put too much faith into the graphene label being able to make a meaningful difference in cooling, but it’s there.

Test Bench and Testing Notes

We use an Alder Lake platform with most background applications such as indexing, Windows updates, and anti-virus disabled in the OS to reduce run-to-run variability. Each SSD is prefilled to 50% capacity and tested as a secondary device. Unless noted, we use active cooling for all SSDs.

Conclusion

There’s not too much to differentiate the MP44S from the Addlink S91 and Sabrent Rocket Q4 2230. It does have a graphene label with similar aspirations as the copper-infused one utilized by Sabrent, although there’s a real limit to how much this will help in a confined space. It also has an SSD toolbox, but that’s pretty standard for the industry, and even if it wasn’t available, there are excellent freeware solutions for most features. Teamgroup’s reputation is not the worst, so it may be a better solution than some shady generic brands, but your decision still comes down to price at the end of the day.

We have nothing bad to say about the MP44S, and it doesn’t have the inconsistencies we saw with the S91. The S91 has a 512GB TLC option, but if we’re talking about high capacity, the Q4 2230 is also a real competitor. Other 2TB options are available on the wider market, including the WD Black SN740 and Crucial 2400, but these were designed for OEM uses. Money can be saved by going to sites like AliExpress to find those types of drives, but there may be more risk involved. The MP44S is, therefore, a relatively safe option if you’re willing to potentially pay more. It’s not half bad at 1TB, but it has to undercut the TLC alternatives to be worthwhile.

The real question is, is it worth going with QLC flash if you're looking for 1TB of storage? Given current pricing, it really seems best to use QLC for 2TB if you want capacity and, if you want maximum performance on a budget, stick with TLC at 512GB with a drive like the TN446. If you have your heart set on 1TB, then the price difference between QLC and TLC is narrow enough to lean in TLC’s favor if performance and full-drive consistency are factors for you.

Still, the 1TB MP44S performed well enough at PCIe 3.0 and in the Steam Deck to suggest you can get by with QLC flash without any large performance gap. It’s just a matter of how much you’re willing to spend or save.

MORE: Best SSDs

MORE: Best External SSDs and Hard Drives

MORE: How We Test HDDs And SSDs

MORE: All SSD Content

Samsung Electronics is poised to employ double-stack architecture for its 9th Generation 3D-NAND memory when it starts its production next year, according to a DigiTimes report that cites Seoul Economic Daily. The sets Samsung apart from SK Hynix, which uses three stacks of NAND to build its 321-layer 3D NAND devices when they enter mass production in the first half of 2025.

Samsung's 9th Generation V-NAND with over 300-layers is set to hinge on the double-stack technique, which Samsung first embraced in 2020 with its 7th generation 176-layer 3D NAND chips. This method involves production of one 3D NAND stack on a 300 mm wafer and then building another stack on top of the first one. Samsung's 300-layer 3D NAND will increase storage density produced on one wafer and will enable makers to build lower-cost SSDs or make the best SSDs cheaper.

Contrarily, rival SK Hynix has revealed its intent to kick off the production of its 321-layer 3D NAND in 2025 using a triple-stack approach. This procedure, distinct from Samsung's, will involve creating three distinct sets of 3D NAND layers, which will increase the number of steps and usage of raw materials, but is meant to maximize yields as it is easier to produce 3D NAND stacks with fewer layers.

Industry speculations that rely on leaked roadmaps suggest that post their 9th Generation 3D NAND, Samsung might adopt a triple-stack methodology for its 10th generation 430-layer 3D NAND. Some experts told Seoul Economic Daily that surpassing 400 layers in 3D NAND would necessitate usage of three separate stacks of 3D NAND, possibly due to yields concerns. Meanwhile, this will naturally increase usage of raw materials and increase costs per 3D NAND wafer. 

Samsung's long-term vision, as presented at the Samsung Tech Day 2022 last October, aspires to reach up to 1,000 layers by 2030. 

Corsair has a habit of bringing out their versions of SSDs a bit later than the competition, as we saw with the MP600 GS, and the same is true with the MP600 Mini following behind Sabrent’s Rocket 2230. There’s nothing wrong with nailing the right hardware in a price-competitive alternative with a targeted capacity and a name brand behind it. The MP600 Mini is designed to hit that popular 1TB capacity point without any frills to get in your way — just an easy upgrade for your Steam Deck and other portable systems. 

The Corsair MP600 Mini is the most efficient M.2 2230 SSD that we have tested to date, and it also delivers excellent all-around performance. It’s not using any special hardware, but it's a solid choice to upgrade your Steam Deck or ROG Ally to 1TB of fast internal storage. It manages to do this at a good price for a TLC option, although the 1TB QLC S91 is currently more than $10 less expensive. If you need to save more money or want additional storage, you have to go elsewhere, though, as the MP600 Mini only comes in a 1TB option.

Corsair is a known name brand, so the MP600 Mini is a safer bet than many of the random M.2 2230 SSDs. It’s equivalent to Inland’s TN446 and Sabrent’s Rocket 2230. Having more options is a good thing as it helps bring down prices. These three drives are the best options at this capacity, so feel free to go with the brand you trust or grab the least expensive one. There’s little not to recommend about the MP600 Mini otherwise.

Specifications

The Corsair MP600 Mini is available only in 1TB, costing $99.99. This undercuts similar drives like the Inland TN446 and Sabrent Rocket 2230 by about $10, but pricing has been shifting over time on all three drives, so keep your eyes open. The MP600 Mini is rated for up to 4,800 / 4,800 MB/s for sequential reads and writes and 850K / 1.1M random read and write IOPS. The IOPS ratings are unusually high for this hardware and should not be taken as meaning the drive is actually faster than its peers. Corsair warranties the drive for 600TB of writes over five years. 

Software and Accessories

Corsair offers a download for its SSD Toolbox software on its site. This application displays drive information and SMART health attributes, plus it allows you to TRIM, clone, overprovision, or securely erase the drive.

A Closer Look

The MP600 Mini is an M.2 2230 form factor SSD with an SSD controller, a single NAND package, a PMIC, and no DRAM. The drive is single-sided for optimal compatibility. There are no embellishments.

The MP600 Mini comes with the familiar Phison E21T SSD controller and Micron’s 176-Layer TLC (B47R). This flash uses 512Gb or 64GB dies, so it requires 16 dies (HDP), which is the effective limit. The E21T has proven to be a reliable performer with good power efficiency, making it a good choice over many other solutions on the market. 

MORE: Best SSDs

MORE: Best External SSDs and Hard Drives

MORE: How We Test HDDs And SSDs

MORE: All SSD Content

Comparison Products

The Corsair MP600 Mini is up against other M.2 2230 SSDs that we have tested, including the similar Rocket 2230, the Addlink S91, the older Inland TN436, the newer Inland TN446, and the Sabrent Rocket Q4 2230. This covers the most current and relevant retail options, with the Team MP44S on the way. The WD SN740, not compared, should be treated as a WD Black SN770. The Micron 2400, for its part, is similar to the Crucial P3 Plus but with an SMI controller. These two drives will be referred to where applicable but are not explicitly reviewed as they are built for OEMs. 

Steam Deck Benchmarks - General

To gauge general Steam Deck performance, we engage in a variety of typical storage situations with active time measurement. These activities include dealing with the default SteamOS through initial setup, re-imaging, initial booting, and booting in general. For game testing, we see how long it takes to install and boot the popular title Hollow Knight. For a full explanation of our testing procedure, see our Steam Deck storage testing article.

The MP600 Mini performs within the expected envelope, although, on the whole, it’s not quite as fast as the Rocket 2230. You would be excused for thinking there’s not much point in getting worked up over a second or two here and there. These tests do show a significant improvement over the stock 64GB drive, which is to be expected. The stock 256GB and 512GB drives would also be somewhat slower than our tested drives, but you won't notice in normal operation if you’re not checking a stopwatch.

However, after some writes and wear, performance over time could be different, especially with a fuller drive. It’s also worth considering efficiency as a newer drive like the MP600 Mini may pull less power over the device's lifetime, which could impact battery wear and lifespan. If you’re the type to reinstall the OS twenty times, multi-boot Windows, and engage in other experiments, there’s value in getting a high-performance SSD with TLC.

Steam Deck Benchmarks - KDiskMark and Temperature

One of the most popular, if not the most popular, storage benchmarks is CrysalDiskMark (CDM), which we use in our Windows-based SSD testing suite. This benchmark relies on Microsoft’s DiskSpd with templated test settings. CDM lets you quickly see how a drive performs against its idealized, rated specifications, and the benchmark can also hint at a more “real world” feel with low queue depth I/O testing.

The Linux flavor of this benchmark, which is easy to install on the Steam Deck, is KDiskMark. The “K” refers to the KDE desktop environment used on the Deck. KDiskMark relies on the Flexible I/O tester, or FIO, instead of DiskSpd. For the temperature portion, we use data from the SMART sensors on the drive, which can be directly polled and from which the maximum temperature can be extracted.

The MP600 Mini’s KDiskMark performance is adequate to good. Read performance at low queue depth remains an important result, whether sequential or random, as it correlates with “real world” feel, which includes game and app loading times to a smaller degree. This is more noticeable in some cases than others, although the difference suggested here is not huge. It’s worth upgrading from the stock drive for both performance and capacity, but it might not be worth upgrading from the TN436 for performance alone if you happen to already have one - depending on how you use your Deck.

The MP600 Mini’s temperature is within reason, coming in six degrees cooler than the TN436. It runs a lot hotter than the stock drive, but that makes sense, given the limited capacity and performance of the 64GB eMMC original.

PC Trace Testing - 3DMark Gaming Storage Benchmark

Built for gamers, 3DMark’s Storage Benchmark focuses on real-world gaming performance. Each round in this benchmark stresses storage based on gaming activities including loading games, saving progress, installing game files, and recording gameplay video streams.

The MP600 Mini comes out near the top, matching the TN446 and Rocket 2230. It performs better than the QLC-based S91 and Rocket Q4 2230 and much better than the older TN436. Latency is somewhat improved in PCIe 4.0 mode.

PC Trace Testing – PCMark 10 Storage Benchmark

PCMark 10 is a trace-based benchmark that uses a wide-ranging set of real-world traces from popular applications and everyday tasks to measure the performance of storage devices.

The QLC drives do better in PCMark 10, but the MP600 Mini’s score is not bad by any means. The QLC drives have massive pSLC caches compared to the restrained ones on the TLC drives, which can have interesting impacts on performance. One advantage of a large cache is that the drive can retain some data in pSLC for later reading, which makes more sense on a QLC drive as the native read latency is higher than pSLC and TLC. 

PC Transfer Rates – DiskBench

We use the DiskBench storage benchmarking tool to test file transfer performance with a custom, 50GB dataset. We copy 31,227 files of various types, such as pictures, PDFs, and videos to a new folder and then follow-up with a reading test of a newly-written 6.5GB zip file.

The DiskBench results are predictable with the TLC drives, including the MP600 Mini, pulling away in PCIe 4.0 mode. In PCIe 3.0 mode, the drives perform very similarly. This does imply that a QLC drive would work well with the Steam Deck for this type of workload, while the ROG Ally can eke out more storage performance.

PC Synthetic Testing - ATTO / CrystalDiskMark

ATTO and CrystalDiskMark (CDM) are free and easy-to-use storage benchmarking tools that SSD vendors commonly use to assign performance specifications to their products. Both of these tools give us insight into how each device handles different file sizes.

ATTO performance is good to excellent with the MP600 Mini, as anticipated. Sequential performance in CrystalDiskMark (CDM) is also good to great compared to the competition. In PCIe 4.0 mode, the TLC drives, like the MP600 Mini, pull away during sequential write workloads. Noticeably, the TLC drives have significantly better 4KB read latency, which is an important result. Real-world impact is probably not large, but the difference is there.

PC Sustained Write Performance and Cache Recovery

Official write specifications are only part of the performance picture. Most SSDs implement a write cache, which is a fast area of (usually) pseudo-SLC programmed flash that absorbs incoming data.  Sustained write speeds can suffer tremendously once the workload spills outside of the cache and into the "native" TLC or QLC flash. 

We use Iometer to hammer the SSD with sequential writes for 15 minutes to measure both the size of the write cache and performance after the cache is saturated. We also monitor cache recovery via multiple idle rounds.

The MP600 Mini writes at its maximum speed for over 20 seconds in both PCIe 3.0 and 4.0 modes. In the former, its pSLC cache is similar in size to the TN446 and Rocket 2230 at around 70GB. In the latter, it appears to be larger, closer to 100GB, as it is writing for the same amount of time but at a higher speed without the restriction of a PCIe 3.0 interface. This does not impact the average write performance, with steady state performance and TLC mode at about 1.8 GB/s. The different cache does have implications for the MP600 Mini’s use in the PCIe 4.0 ROG Ally, but the impact should be small.

The drive can maintain relatively decent performance in its TLC mode and can afford to take some time to recover its pSLC. The QLC drives rely more on pSLC and, with their large caches, must recover more quickly.

Power Consumption

We use the Quarch HD Programmable Power Module to gain a deeper understanding of power characteristics. Idle power consumption is an important aspect to consider, especially if you're looking for a laptop upgrade as even the best ultrabooks can have mediocre storage.

Some SSDs can consume watts of power at idle while better-suited ones sip just milliwatts. Average workload power consumption and max consumption are two other aspects of power consumption, but performance-per-watt is more important. A drive might consume more power during any given workload, but accomplishing a task faster allows the drive to drop into an idle state more quickly, ultimately saving energy.

For temperature recording we currently poll the drive’s primary composite sensor during testing with a 24C ambient.

The MP600 Mini is the most efficient M.2 2230 SSD that we’ve tested to date. It manages to edge out the Rocket 2230 and TN446, with the QLC drives further behind. The older TN436 is in another zip code.

In normal testing, the MP600 Mini topped out at around 53C. You would have to engage in sustained writes to get it up towards throttling temperatures around 80C. The drive does run slightly cooler on average in PCIe 3.0 mode, but it’s likely to be hot in both the Steam Deck and ROG Ally due to their tight environments. There is a valid link between power draw and heat dissipation, and while this hardware is fast, the peak power draw is still quite reasonable.

Test Bench and Testing Notes

We use an Alder Lake platform with most background applications such as indexing, Windows updates, and anti-virus disabled in the OS to reduce run-to-run variability. Each SSD is prefilled to 50% capacity and tested as a secondary device. Unless noted, we use active cooling for all SSDs.

Conclusion

We have nothing bad to say about the MP600 Mini, except maybe that we’d like to see more. 1TB is the most popular capacity for an M.2 2230 upgrade, particularly on the Steam Deck, so it makes sense that Corsair focused on hitting that point. TLC works well at that capacity for maximum performance, and this hardware is quite efficient, to boot. Currently, Corsair can sell this drive at a lower price than the competition, so this strategy works well. If you need 256GB, we recommend the Rocket 2230; at 512GB, we recommend the TN446, and at 2TB, the Rocket Q4 2230 or S91.

Many other options are on the market, including the Solidigm P41 Plus, the WD SN740, and the Micron 2400. These are all technically OEM drives. We would expect none to be as efficient as the drives based on the Phison E21T with Micron flash, whether TLC or QLC. However, they are all viable options if you can get them at a lower price. On the other hand, we would recommend against older drives, including the TN436 and other Kioxia BG5 clones. Generically-branded drives may or may not have newer hardware, but support will likely be limited for them anyway, so they should be considered a last resort - especially if you’re buying from a marketplace like AliExpress.

All in all, the MP600 Mini is a solid all-around drive. You can buy it without worry. It’s easy to get, and it’s priced right. Corsair knew what they were doing when they targeted this drive, although we would like to see even more options in this space.

MORE: Best SSDs

MORE: Best External SSDs and Hard Drives

MORE: How We Test HDDs And SSDs

MORE: All SSD Content

If you’re a Steam Deck owner, you have undoubtedly heard of the Sabrent Rocket 2230, an early retail M.2 2230 NVMe SSD that helped DIYers upgrade once-space-limited Steam Decks. However, even the original Rocket 2230’s 1TB maximum could be tight with modern games and lots of ROMs, but thankfully now you can reach up to 2TB with Sabrent’s follow-up Rocket Q4 2230. 

SD cards and external storage can only do so much with the Steam Deck and other handhelds, and if you want easily-portable, high-performance storage, a good NVMe SSD is hard to beat. The Sabrent Rocket Q4 2230 is an easy way to upgrade your Steam Deck, ROG Ally, or other portable device to 2TB of fast internal storage. It doesn’t cut corners by using old technology, it's relatively fast and efficient, and it's more than enough to get you gaming on the go. To provide 2TB of capacity in the M.2 2230 form factor while being single-sided, it has to compromise by using QLC flash instead of the faster TLC flash. The QLC flash reduces peak and sustained performance, but the drive performs well enough where it matters.

The drive is supported by a normal retail warranty and comes with Sabrent’s copy of Acronis True Image. The drive arrives with a copper-infused heatspreader label, which might require adjustment in some devices (such as with the EMI sleeve in the Deck). Currently, the only readily available TLC-based option in this form factor is the WD SN740, the client version of the WD Black SN770, although we may see more alternatives thanks to the 232-Layer generation of flash.

The Rocket Q4 2230 is very similar to the Crucial P3 Plus and the Crucial P3. These drives are efficient and generally run cool. Like with the Crucial P3 and P3 Plus, the Rocket Q4 2230 opts for a large pSLC cache, which reduces sustained performance compared to a drive like the Solidigm P41 Plus.

The Micron 2400, another popular 2TB choice for M.2 2230 SSDs, uses the same controller as the P41 Plus — the SMI SM2269XT — but comes with Micron’s QLC flash. That controller tends to be less efficient, and it’s possible the Micron 2400 has a larger pSLC cache, making the Rocket Q4 2230 arguably the best all-around QLC-based option for embedded devices at this time. Let's see how it performs. 

Specifications

The Rocket Q4 2230 is currently available only at 2TB for $220 on Amazon. Price swings are common in the SSD market, and the competition's pricing also varies.

Although Sabrent appears to have left the door open for a 1TB SKU, the 2TB is the only one currently available. It peaks at 5,000 / 3,200 MB/s for sequential reads and writes and 480K / 750K random IOPS. Write speeds can be lower for QLC, although this rated value is below what the drive can achieve, and it has enough speed to saturate the PCIe 3.0 bus present in the Steam Deck.

The drive is warrantied for five years with registration, and the drive can absorb up to 450TB of written data. This endurance rating is more than sufficient for the drive’s use in a device like the Steam Deck. Worries about QLC endurance are not warranted; Micron’s 176-Layer QLC is normally rated for 1,500 program/erase cycles with up to 100,000 in pSLC mode.

Software and Accessories

The Sabrent Rocket Q4 2230 comes with a downloadable, Sabrent-specific version of Acronis True Image which is useful for cloning and imaging. This can be nice to have, especially when working with the ROG Ally. Sabrent also has its own SSD toolbox, which will presumably offer firmware updates for the drive, if applicable.

A Closer Look

The Rocket Q4 2230 arrives in a stylish tin, showing off a mixture of white and copper coloring. While not important, the presentation is attractive and does separate it from many OEM SSDs. The drive’s label operates as a heatspreader as it is made of thermally-conductive copper.

The Rocket Q4 2230 sports an SSD controller and a single NAND package. The drive is single-sided, which is important for many embedded devices like the Steam Deck — it can be a challenge to pack in 2TB of flash with this limitation. 

The Rocket Q4 2230 uses Micron’s 176-Layer QLC (N48R) flash, which is convenient for achieving 2TB in such a small package. In contrast, the original Rocket 2230 was constrained to 1TB because it used Micron’s 176-Layer TLC (B47R).

Using 1Tb dies with TLC flash is possible, as the WD SN740 does with BiCS5, but that flash is less efficient. SK hynix also makes 1Tb dies, but given current stacking technology, we will likely see 232-Layer generation flash tackling this role (aside from maybe BiCS6), such as with Micron’s B58R TLC. This flash is already available on many drives, like the Crucial T700. 

The Rocket Q4 2230 is DRAM-less, but the host memory buffer (HMB) feature is supported on Valve’s Steam Deck, ASUS’s ROG Ally, and other devices. This should be sufficient for portable gaming workloads. The drive is also PCIe 4.0, which begs the question: does that make a difference for the Deck? In fact, this drive is somewhat more efficient when restricted to the 3.0 interface, and it’s worth going for new hardware for overall performance and efficiency on a portable platform. Such technology does not often arrive on an older PCIe interface, and with it being backward compatible, there’s little reason not to go with a 4.0 drive.

The Rocket Q4 2230 uses the Phison E21T SSD controller, which we’ve reviewed on multiple products, including the original Rocket 2230 (a.k.a. the Rocket 2230 NVMe 4.0). This controller has a history of providing good performance and power efficiency. The latter is probably more important with battery-powered, portable devices. This controller’s primary competition would include the Silicon Motion SM2269XT on drives like the Micron 2400 and Solidigm P41 Plus, the InnoGrit IG5220, and the Maxio MAP1602. However, the latter two are less commonly found on M.2 2230. The MAP1602, on paper, is faster but can run slower, which offers some flexibility.

Future controllers, like the Phison E27T and Silicon Motion SM2268XT, could usurp its position, although the extra performance doesn’t mean much for the PCIe 3.0 Steam Deck. Efficiency gains would be nice, but these two controllers will still be 12nm. Any gains would more likely come from the flash. One other worthy mention is the Kioxia BG6 with an unknown controller. Kioxia has often used Phison controllers, and the BG6 uses Kioxia’s 162-Layer TLC (BiCS6), giving it some bandwidth uplift. Efficiency remains uncertain.

MORE: Best SSDs

MORE: Best External SSDs and Hard Drives

MORE: How We Test HDDs And SSDs

MORE: All SSD Content

Comparison Products

The Sabrent Rocket Q4 2230 faces off against other M.2 2230 SSDs, including the Rocket 2230, the Corsair MP600 Mini, the older Inland TN436, the newer Inland TN446, and the Addlink S91. We also threw in a 512GB Rocket 2230 at PCIe 4.0. This gives a good range of the 2230 drives on the market, although more are to come. 

The OEM SN740 and Micron 2400 options are not currently represented. Still, the former is essentially a shorter WD Black SN770, which we compared in our original Rocket 2230 review, while the 2400 is similar to a Crucial P3 Plus with the perhaps less-efficient controller found on the Solidigm P41 Plus. The P41 Plus comes in the M.2 2230 form factor, but it’s designed for OEM use.

We’d like to see an M.2 2230 drive with the InnoGrit IG5220 controller, such as the Visiontek DLX4, which we will investigate in the future. We can make some assumptions about it as we tested IG5220 SSDs like the HP FX900, which were later compared to E21T SSDs, including the Silicon Power UD90. We’ve found that the controllers should be largely comparable for 2230 devices. Still, the E21T would give a slightly better overall experience - just in case you find yourself weighing the Rocket 2230 or MP600 Mini against the DLX4 in the meantime.

Steam Deck - General

To gauge general Steam Deck performance, we engage in a variety of typical storage situations with active time measurement. These activities include dealing with the default SteamOS through initial setup, re-imaging, initial booting, and booting in general. For game testing, we see how long it takes to install and boot the popular title Hollow Knight. For a full explanation of our testing procedure, see our Steam Deck storage testing article.

We don’t see any negative anomalies with the Rocket Q4 2230 in general Steam Deck testing. The use of QLC to reach 2TB does not appear to be a major factor when drives are restricted to PCIe 3.0 interface that's present in the Steam Deck.

Steam Deck - KDiskMark and Temperature

One of the most popular, if not the most popular, storage benchmarks is CrysalDiskMark (CDM), which we use in our own Windows-based SSD testing suite. This benchmark relies on Microsoft’s DiskSpd with templated test settings. CDM lets you quickly see how a drive performs against its idealized, rated specifications, and the benchmark can also hint at a more “real world” feel with low queue depth I/O testing.

The Linux flavor of this benchmark, which is easy to install on the Steam Deck, is KDiskMark. The “K” refers to the KDE desktop environment used on the Steam Deck. KDiskMark relies on the Flexible I/O tester, or FIO, instead of DiskSpd. For the temperature portion, we use data from the SMART sensors on the drive, which can be directly polled and from which the maximum temperature can be extracted.

The QLC drives — the Rocket Q4 2230, the S91, and the MP44S — drop a bit in random read workloads at low queue depth. This did not translate to significantly worse general usage results for us but could impact certain games or activities to a small degree. The older TLC-based TN436 was even worse here but still far better than the stock 64GB drive.

The Q4 2230 did not run particularly cool but was within the general ballpark of other Phison E21T-based drives, whether they use TLC or QLC. We can recommend any of them for the Steam Deck.

PC Trace Testing - 3DMark Gaming Storage Benchmark

Built for gamers, 3DMark’s Storage Benchmark focuses on real-world gaming performance. Each round in this benchmark stresses storage based on gaming activities including loading games, saving progress, installing game files, and recording gameplay video streams.

The Rocket Q4 2230 is mediocre to below average in 3DMark, falling behind the newer TLC drives. However, the difference isn't huge. Latencies are better/lower with all drives when run at PCIe 4.0 due to the controllers opening up a faster bus. 

PC Trace Testing – PCMark 10 Storage Benchmark

PCMark 10 is a trace-based benchmark that uses a wide-ranging set of real-world traces from popular applications and everyday tasks to measure the performance of storage devices.

The Rocket Q4 2230 rockets ahead to first place, followed by the very similar S91. This is a matter of capacity, as these are the two 2TB drives. Latencies are again better with a PCIe 4.0 interface, although the difference between the Deck and ROG Ally with these drives may more likely come from the OS being used.

It isn’t easy to guess what a 2TB TLC drive would look like here, but a possible example is the Addlink S90 Lite, which utilizes the same E21T controller. Its performance is within the ballpark of what we get with QLC. This is likely a limitation of the less-dense TLC and would be overcome when using 1Tb TLC. The 2TB SN770, which should score like the 2TB SN740, has results that support this theory. Future drives with Micron’s 232-Layer TLC would also probably score better.

PC Transfer Rates – DiskBench

We use the DiskBench storage benchmarking tool to test file transfer performance with a custom, 50GB dataset. We copy 31,227 files of various types, such as pictures, PDFs, and videos to a new folder and then follow-up with a reading test of a newly-written 6.5GB zip file.

Copying requires writing, so the QLC drives, like the Rocket Q4 2230, are bound to fall behind in DiskBench. None of these drives can saturate the PCIe 4.0 interface, and being so close to the PCIe 3.0 edge means the Q4 2230 doesn’t gain much here by using the faster interface. This is bittersweet as it does mean the drive doesn’t lose much when used in the PCIe 3.0 Steam Deck. It has more dies than the 512GB Rocket 2230, so it can also benefit from its extra capacity, which is a factor for those upgrading from ≤512GB stock drives.

PC Synthetic Testing - ATTO / CrystalDiskMark

ATTO and CrystalDiskMark (CDM) are free and easy-to-use storage benchmarking tools that SSD vendors commonly use to assign performance specifications to their products. Both of these tools give us insight into how each device handles different file sizes.

The Rocket Q4 2230 performs very well in ATTO. This is to be expected, as Phison SSD controllers usually do. It’s easy to see here how much the older TN436, a Kioxia BG5 clone with hardware used by many knock-off brands, falls behind.

The Rocket Q4 2230 performs ably with sequential workloads in CrystalDiskMark (CDM), too, being effectively at or near the top for reads, and both read and write workloads when restricted to PCIe 3.0. Even though it and the S91 have twice the capacity, the amount of dies and internal interleaving is the same as the 1TB TLC drives.

However, the QLC drives drop behind when it comes to write workloads with a PCIe 4.0 interface, despite all drives writing in pSLC mode. Writing to a QLC drive has more overhead as the block size is larger, there’s a bigger capacity ratio for pSLC, and native write speeds are significantly slower. For this reason, reducing the maximum write speeds even in pSLC mode can be wise, particularly as data might remain in pSLC longer to benefit reads, as QLC also has higher read latency. These drives also push up to the limits of the bus, with writes always hitting a wall before reads - due to more overhead, for one - which can hinder QLC more than TLC.

The longer read latency or time to read (tR) of the QLC flash is made evident through the CDM 4KB random read results. The Rocket Q4 2230’s latency is significantly higher than the Rocket 2230’s. 4KB write latency or time to program (tPROG) is much better, which we’ve seen before with Micron’s 176-Layer QLC. Under PCIe 4.0, the peak IOPS - however unrealistic a workload - will be lower for QLC, but the faster 4K writes can feel snappier in very rare cases.

PC Sustained Write Performance and Cache Recovery

Official write specifications are only part of the performance picture. Most SSDs implement a write cache, which is a fast area of (usually) pseudo-SLC programmed flash that absorbs incoming data.  Sustained write speeds can suffer tremendously once the workload spills outside of the cache and into the "native" TLC or QLC flash. 

We use Iometer to hammer the SSD with sequential writes for 15 minutes to measure both the size of the write cache and performance after the cache is saturated. We also monitor cache recovery via multiple idle rounds.

Regardless of the PCIe mode, the Rocket Q4 2230 writes the entirety of its free QLC as pSLC, which is shown by the >500GB cache. The folding to the QLC state after this is very slow, on the order of 100 MB/s during steady state conditions. Not that we necessarily see better things with all TLC flash, as the BiCS on the TN436 is quite inconsistent.

As shown here, write performance will be a factor for you as a buyer because upgrading an embedded device is usually followed by a lot of data writes - imaging/cloning, lots of downloads and installations, and usually a good deal of benchmarking, too. These drives will all eventually recover to pSLC mode. Writes become more consistent after this initial phase, and so the slower native flash modes are less critical. The pSLC cache will shrink with drive usage, but even at 75% full, the Q4 2230 will have at least 125GB of pSLC, which is more than enough.

Power Consumption and Temperature

We use the Quarch HD Programmable Power Module to gain a deeper understanding of power characteristics. Idle power consumption is an important aspect to consider, especially if you're looking for a laptop upgrade as even the best ultrabooks can have mediocre storage.

Some SSDs can consume watts of power at idle while better-suited ones sip just milliwatts. Average workload power consumption and max consumption are two other aspects of power consumption, but performance-per-watt is more important. A drive might consume more power during any given workload, but accomplishing a task faster allows the drive to drop into an idle state more quickly, ultimately saving energy.

For temperature recording we currently poll the drive’s primary composite sensor during testing with a 24C ambient.

The Rocket Q4 2230 is very efficient, although not quite as efficient as its peers. However, the TLC drives are only 1TB. At 2TB, they could paradoxically be either more or less efficient. If utilizing the same density dies, theoretically, as you can’t do that with a single-sided M.2 2230 SSD, the increased overhead may lower efficiency. If using denser dies, as does the 2TB SN770, efficiency can increase because you have the same number of dies to engage but with more addresses per die.

This is not “free” because latency gets worse. Still, it does suggest that Micron’s 232-Layer TLC could probably provide for the best, no-compromise 2TB M.2 2230 drive, as the SN740’s BiCS5 lacks any CMOS under Array (CuA) technology and is already generationally behind Micron’s 176-Layer flash. That’s not to say WD hasn’t done amazing things with the flash, but embedded devices benefit from newer technology more than usual which is obvious when looking at the TN436. The downside is that no QLC or PLC is dense enough for a 4TB drive in this space. We’ll also have to wait for a verdict on BiCS6.

Generally, we would expect BiCS5 to be less efficient than B47R. In our testing, these drives largely peak at 3-4W when something like the 2TB SN740 is rated for a peak of 6.3W, a substantial difference. Our 2TB SN770 reached a peak of 4.91W, which is noticeably less efficient at 1TB and 2TB. In practice, the difference probably isn’t massive as long as you have a newer controller, though - the TN436’s E19T is objectively much less efficient.

It is worth noting that the Rocket Q4 2230 pulls less power when idle than the TLC options, but in the Steam Deck or Ally, you should have the best power states available outside of any turbo modes. Still, there are cases where this could trickle, for example, if you like to really push the battery with full-on gaming. Still, even in this worst-case scenario, the Rocket Q4 2230 would only theoretically be a few percentage points of battery life better at most than something like the 2TB SN740.

As for temperature, it’s worth checking our Crucial P3 and Crucial P3 Plus reviews to get an idea of what to expect at PCIe 3.0 and 4.0, respectively. This is because the P3 and P3 Plus use the same hardware as the Rocket Q4 2230. As tested, the Rocket Q4 2230 did not throttle even at PCIe 4.0, although it did get up to 79C with sustained write testing. Normal desktop use was well below throttling, especially in PCIe 3.0 mode.

Its higher peak temperature than the P3 and P3 Plus likely boils down to the shorter form factor, with the controller and flash being closer together. The drive is bound to run hotter when installed in an embedded device, but thankfully sustained writes are not common. We would suggest a careful installation; if possible, adding thermal padding could help. The Rocket Q4 2230 does have the heatspreader label, which helps a small amount, although, as mentioned before, it requires adjustment with the Deck’s SSD EMI shield.

Test Bench and Testing Notes

We use an Alder Lake platform with most background applications such as indexing, Windows updates, and anti-virus disabled in the OS to reduce run-to-run variability. Each SSD is prefilled to 50% capacity and tested as a secondary device. Unless noted, we use active cooling for all SSDs.

Bottom Line

The Sabrent Rocket Q4 2230 is not a bad SSD, although that’s to be expected as it’s built on the well-performing Phison E21T SSD controller. It doesn’t use TLC flash, but its QLC is some of the best. It’s a worthwhile trade-off to get 2TB in a single-sided M.2 2230 SSD. It wasn't easy to find M.2 2230 drives in retail for a long time, but now multiple drives have sprung up. For example, many Kioxia BG5 clones are sold by random companies on Amazon.

There will be more and better options in the future, but for now, the Rocket Q4 2230 is a quick and easy way to get a lot of fast internal storage, even when it’s not the less expensive option.

Many embedded or portable devices only run at PCIe 3.0, including the very popular Steam Deck. In this mode, PCIe 4.0 SSDs run cooler and more efficiently and still perform excellently. The Rocket Q4 2230 is perfect here, as it only really falls behind with peak performance when compared to PCIe 4.0 mode. Since it’s QLC-based, it does suffer with sustained performance, particularly as it has a large pSLC cache which leads to a significant drop, but thankfully this is something that is rarely a problem with devices like the Steam Deck. With how things are right now, the Rocket Q4 2230 is a good choice if it’s priced right.

There are, of course, other options, such as the Addlink S91, which uses effectively the same hardware. There’s also the TLC-based WD SN740, although this drive was not widely available in retail until Framework picked it up. The Micron 2400 is in similar straits, although it was available earlier and in more places, and now finds a nice retail home with iFixit. The 2TB Team MP44S is another option that has only been available as of mid-July. 

All of these drives will do as well as the Rocket Q4 2230 if you’re looking for a 2TB solution, but the level of support will vary. However, we caution that an M.2 2230 SSD purchase should be made carefully and with some consideration — be wary of cheap no-name drives available on AliExpress, eBay, and other marketplaces.

The Rocket Q4 2230 may also arrive with 1TB of capacity, but there’s a lot more competition there; it would have to be priced aggressively, and the market may not support that. A better option might be to wait until denser TLC allows for more 2TB retail alternatives. This would bring down the cost of the Rocket Q4 2230, which would be a good thing for the market. Older drives like the TN436 can’t hang, but future controllers could allow devices like the Ally to saturate PCIe 4.0 without additional downsides. The Rocket Q4 2230 is a safe bet right now, though, particularly as extra performance isn’t needed yet.

MORE: Best SSDs

MORE: Best External SSDs and Hard Drives

MORE: How We Test HDDs And SSDs

MORE: All SSD Content

In 2020, Chinese chip champion Semiconductor Manufacturing International Corp. was two or three generations behind global leaders. Today, following several rounds of crippling U.S. sanctions against the Chinese semiconductor sector, the People's Republic's chip industry is at least five generations behind — and the gap can expand, according to Gerald Yin, chief executive of Advanced Micro-Fabrication Equipment Inc., a wafer fab equipment manufacturer from China, reports DigiTimes.

The past two U.S. administrations have introduced 15 semiconductor-focused sanctions to keep China's semiconductor production technologies generations behind the global leaders — namely Intel from the U.S., Samsung from South Korea, and TSMC from Taiwan. Back in 2020, SMIC was about to start making chips on its 7nm-class manufacturing process that featured a logic transistor density akin to that of TSMC's N7 and Intel's 10nm. At the time, SMIC was a couple of generations behind TSMC, which was gearing up to start production of chips on its N5 (5nm-class) node.

But the sweeping sanctions imposed by the U.S. government on October 7, 2022, and followed by Japan and the Netherlands in 2023, have set the Chinese semiconductor industry back by at least a decade. Wafer fab tools manufacturers need to obtain a special license to ship equipment that can be used to make logic chips with non-planar transistors on 14nm/16nm nodes and below, 3D NAND with 128 or more layers, and DRAM memory chips of 18nm half-pitch or less to a Chinese entity. This effectively limits China's semiconductor industry to 28nm and thicker nodes — unless it can come up with its own wafer fab tools, or the export licenses are granted.

The ramifications of these restrictions are palpable in the Chinese semiconductor landscape. SMIC, a front-runner in China's semiconductor domain, had once been vocal about its advancements with 14nm and 12nm nodes. Yet, by mid-2023, any mention of these technologies disappeared from its website and other official communications. Despite these setbacks, some insiders are optimistic, viewing the U.S. obstacles as an opportunity for China to bolster its domestic semiconductor equipment prowess.

If this does not happen, the gap between China's semiconductor industry and the global leaders will expand even further, barring other changes in government policy.

The Compute Express Link technology allows for building various devices to address a vast array of workloads, from expanding memory subsystem capacity and performance to offering ultra-fast persistent storage. Being one of the world's leading makers of NAND memory, Kioxia can address the latter. Recently, it demonstrated its 3D NAND and XL-Flash-based CXL solutions at the Flash Memory Summit 2023, as reported by ServeTheHome. 

Kioxia revealed plans to offer two lineups of its CXL products: CXL + XL-Flash-based devices for performance and reliability-centric applications such as in-memory databases and AI inference workloads, as well as CXL + BiCS 3D NAND-powered devices for capacity-hungry applications like Big Data and AI training. In both cases, storage devices use a special controller and CXL.mem protocol for reading and CXL.io protocol for writing to minimize respective latencies.

Regarding demonstrations, Kioxia showed a sample of a 1.3 TB CXL 1.1/CXL 2.0 BiCS 3D NAND-based device in an E1.S form-factor that can be installed into an E3.S chassis for higher performance/thermal capacity. The device uses a PCIe x4 interface (PCIe Gen5, we presume), though Kioxia does not disclose its performance characteristics, perhaps because the development has not finished.

While usage of a low-latency 3D (TLC) NAND-powered device over a PCIe interface with the CXL protocol on top seems like a very plausible idea, usage of an XL-Flash-based storage device promises to be even more fruitful due to the higher performance of XL-Flash compared to commodity 3D NAND.

Kioxia's proprietary 1st Generation XL-Flash is essentially single-level cell (SLC) NAND spread over 16 planes, whereas 2nd Generation XL-Flash is multi-level cell (MLC) flash memory spread over a higher number of planes, which by definition offers lower latency and higher parallelism for read/write operations, thus guaranteeing massively higher performance compares to mainstream 3D TLC NAND. 

Last year Kioxia said that its CXL storage devices would use its 2nd Generation XL-Flash, which promises to outperform the 1st Generation XL-Flash while being more cost-effective and thus enabling higher capacities. 

For now, Kioxia remains tight-lipped when it plans to ship its CXL devices featuring commodity 3D NAND and proprietary storage-class XL-Flash memory. Still, based on the fact that it is displaying some of the former products and does not showcase the latter (at least openly), we can assume that 3D NAND-based devices will be available a bit earlier.

Samsung is at the Flash Memory Summit in California, showing off its latest wares, announcing breakthrough technologies, and discussing some incredible advances. Samsung is often the source of the biggest news stories of these events, and it hasn’t disappointed with its announcement of both a 256TB SSDs and unveiling of its PBSSD architecture, designed for peta-byte scale solutions.

The Samsung presentation at the FMS 203 was all about “How Memory Innovations Shape Products and Services”. And, you guessed it, everything was being framed in the context of being reimagined for “the AI era.” Never worry, as Samsung is here to develop the latest technologies to cope with the “exponential growth of data and its many applications,” attendees were told.

PM1743 With PCIe 5.0

Samsung started by highlighting its PM1743 server SSD, which was first unveiled at FMS last year. This comes in capacities of up to 15.36 TB, has been enhanced with a PCIe 5.0 interface. The interface revamp means the new drive is capable of “achieving twice the power efficiency of its predecessor,” says Samsung. The first users of the new PM1743 have it earmarked for usage in the field of generative AI, such as with ChatGPT, according to Samsung.

The development of the PM9D3a server SSD in standard 2.5-inch form factor with PCIe 5.0 is also complete, said the tech giant. With its 8-channel controller, this drive will be up to 2.3 times faster than its predecessor, the company claimed. The new PM93a will become available up to 15.36 TB first, with models up to 30.72 TB in H1 2024.

256TB SSDs  

In the quest for maximum data storage within the power and volume limits of a single-server rack, Samsung has created a 256TB SSD. It says that the new devices wield QLC NAND and the highest level of integration density to create these devices.

In Samsung’s tests, the 256TB SSD consumed seven times less power than a stack of eight 32TB SSDs (reminder, 8x32=256).

Petabyte Scale PBSSD

Last but certainly not least, Samsung showcased its latest PBSSD architecture at the Summit. It says that this takes the form of a “petabyte-scale ultra-high capacity solution that provides high scalability by varying the capacity depending on the application.”

With such a great capacity in a single device, Samsung and partners like Meta are aiming to make PBSSDs multi-user friendly. It says the platform made use of a technology called Flexible Data Placement (FDP), which has been ratified in NVMe. FDP is a way of optimizing data placement for increased predictability and performance in real-world hyperscale workloads. The software behind FDP is completely open source.

We hope some of this super-capacious flash memory tech is going to trickle down to consumers. Shoppers still find it difficult to find readily available PC SSDs in sizes larger than 4TB.

Phison has demonstrated its what it claims to be the fastest client solid-state drive released to date: the PS5026-E26 Max14um Gen5 SSD with an over 14 GB/s sequential read speed, which is higher than what the best SSDs offer today, at Flash Memory Summit 2023. The unit was cooled down using an innovative system from Frore.

Phison and its partners were first to offer SSDs based on the PS5026-E26 controller with a PCIe 5.0 x4 interface. Those drives significantly outperformed all SSDs that existed at the time, even though they could not demonstrate their full potential due to scarce availability of 3D NAND with a 2,400 MT/s interface. By now, Micron has finally managed to ramp up production of its fastest 3D NAND devices and Phison can demonstrates its PS5026-E26 Max14um Gen5 SSD featuring a sequential write speed of over 14 GB/s. 

As PCIe Gen5 SSDs based on controllers from Innogrit and Silicon Motion are getting closer, Phison wants to remind people that its platform can pack quite a punch, which is why it built an SSD based on its PC5026-E26 controller and Micron's B58R 3D TLC NAND memory chips that can hit a 14,175 MB/s sequential read speed and a 12,471 MB/s sequential write speed as well as 1.79 million random read IOPS and 1.74 million random write IOPS. 

Performance offered by the PS5026-E26 Max14um Gen5 SSD is a bit higher than sequential performance of enterprise-grade PCIe Gen5 drives and even higher than what Sabrent achieved with its E26-based drive a month ago. To hit such read/write speeds, Phison had to tweak firmware and cool down its drive properly. To do so, Phison used Frore Systems's AirJet Mini cooling system that uses tiny vibrating membrane to generate airflow.

For now, Phison does not disclose whether it plans to offer its partners reference designs of its PS5026-E26 Max14um Gen5 SSD, or just let them achieve similar performance levels themselves, though it is logical to expect SSD makers to offer another generation of E26-based SSDs in light of the fact that there are PCIe Gen5 drives featuring different controllers incoming.

SK Hynix late on Tuesday demonstrated its 321-layer TLC NAND memory at the Flash Memory Summit 2023. This new type of flash is going to enter mass production in 2025, but the company is already showing it off to demonstrate that it is ready for the future.

The demonstrated memory device has 1Tb capacity (128GB) and 3D TLC architecture. SK Hynix says that the 321-layer memory IC features a 59% improvement in productivity compared to a 512Gb 238-layer 3D TLC device, which indicates a tangible storage density increase. Meanwhile, the company does not disclose how it produces 321-layer NAND memory as well as approximate per bits costs compared to previous-generation nodes.

If SK Hynix's lowest-capacity 321-layer NAND device can store 128GB of data, then it is reasonable to expect premium devices to feature larger capacity and enable SSDs with capacities way larger than what we have today.

SK Hynix yet has to disclose how it has managed to achieve 321 active layers. While it is more than likely that the company used string stacking technique that places one or more 3D NAND devices on top of each other and connects them, the question is whether it stacked two or more 3D NAND devices on top of each other.

Traditionally, SK Hynix labels its multi-layer NAND memory as '4D NAND' as it uses CMOS under Array (CuA) architecture that places NAND logic under the 3D NAND memory cells array to reduce die size and costs. Meanwhile, similar layout is also used by Micron, which uses the term '3D NAND.'

Speaking of NAND logic, it is noteworthy that SK Hynix did not share interface speed supported by its 321-layer NAND memory, possibly because it is too early to set them as the company only plans to start mass production of such flash sometimes in the first half of 2025.

"With another breakthrough to address stacking limitations, SK hynix will open the era of NAND with more than 300 layers and lead the market," said Jungdal Choi, head of NAND development at SK Hynix, said during a keynote speech. "With timely introduction of the high-performance and high-capacity NAND, we will strive to meet the requirements of the AI era and continue to lead innovation."

Kioxia on Monday introduced one of the industry's first families of solid-state drives for hyperscalers featuring a PCIe 5.0 x4 interface and capacities of up to 30.72TB. The CD8P line-up of SSDs is designed for mixed use and read intensive workloads and feature all the reliability features that one comes to expect from datacenter-grade SSDs designed to work for 24/7.

Kioxia's single-port CD8P solid-state drives come in E3.S and U.2 form-factors and offer a wide range of capacities starting at 1.6 TB and ending at 30.72 TB. As for performance, the manufacturer rates the drives at up to 12,000 MB/s sequential read speed, up to 5,500 MB/s sequential write speed, up to 2,000,000 random read 4K IOPS, and up to 400,000 random write 4K IOPS. 

The new drives are a bit slower than enterprise-grade CM7 SSDs launched a year ago. Those drives are designed for enterprise workloads and have different performance (up to 14,000 MB/s) and feature set requirements (e.g., dual-port, FIPS SED). 

The new CD8P SSDs use Kioxia's proprietary SSD platform consisting of its own NVM 2.0-compliant controller, firmware, and 112-layer BICS 5 3D TLC NAND memory that supports the company's 7th Generation flash die failure protection to ensure reliable data storage, power loss protection, and end-to-end data protection. In addition to regular CD8P models, Kioxia will also offer SKUs supporting Sanitize Instant Erase (SIE) and Self-Encrypting Drive (SED) capabilities. Meanwhile, CD8P will not be available with FIPS SED capability.

It should ne noted that Kioxia will offer its CD8P SSDs in two types of configurations: the CD8P-V for mixed-use workloads (up to three drive writes per day) will come with 1.60 TB, 3.20 TB, 6.40 TB, and 12.8 TB capacities, whereas the CD8P-R for read-intensive applications (up to one drive writes per day) will be offered in 1.92 TB, 3.84 TB, 7.68 TB, 15.36 TB, and 30.72 TB capacities. 

While Kioxia is introducing its new CD8P SSDs at the Flash Memory Summit and plans to show them up and running at the show, it does not disclose when these drives are set to be available. Given that they are aimed mostly at hyperscalers, Kioxia will ship them once its customers validate and qualify them in their environments.

Today at Newegg, you can take home the MSI Spatium M450 1TB SSD for a seriously low price—one of the lowest we’ve seen for a PCIe 4.0 SSD as of late. It has a recommended price of $59 but is marked down to $37. Using the included rebate card will take the offer down even further to $27.

This offer puts the final price at just under 3 cents per GB. It’s not the fastest SSD on the market but it has plenty of performance for casual computing and most gaming needs.

There are other capacities available for the MSI Spatium M450 model including 250GB, 500GB and 2TB but this discount applies specifically to the 1TB edition. All of the SSDs in this line have an M.2 2280 form factor and use 3D NAND memory. They connect using a PCIe 4.0 x4 interface and are powered by a Phison E19T controller.

The purchase is supported by a limited 5-year manufacturer’s warranty from MSI that voids if the drive reaches 600TBW before the 5-year period has passed. Users can also take advantage of Newegg’s 30-day return policy if necessary.

Visit the MSI Spatium M450 1TB SSD product page at Newegg for more details and purchase options.

Yangtze Memory Technologies (YMTC) has teamed up with China-based SSD brands and developers of flash controllers to market solid-state drives based on its 232-layer 3D NAND memory, reports DigiTimes. While the company does not formally have access to wafer fab equipment required to make such flash memory and cannot sell such devices to multinational drive suppliers, it can still sell in China. 

YMTC has traditionally worked with Maxio Technology, a China-based developer of SSD controllers, to design solid-state drive platforms. So it is not particularly surprising that the two companies have a platform based on 232-layer 3D NAND. DigiTimes claims that YMTC has teamed up with other China-based SSD controller firms to build drives featuring its latest memory and their chips, though it does not disclose the names of these companies.

One SSD supplier that markets 232-layer 3D TLC NAND-based drives is Tengyn. The manufacturer started to sell SSDs featuring YMTC's X3-9070 memory a couple of months ago, and since then, their prices have dropped sharply. Like in other parts of the world, there is an oversupply of 3D NAND in China, so low SSD prices are expected. Meanwhile, since YMTC has trouble selling its advanced 3D NAND memory elsewhere, it is probably willing to sell it at a discount in the People's Republic to keep its fabs busy.

YMTC's 232-layer 3D TLC NAND memory featuring its Xtacking 3.0 architecture was developed to address applications like the best client and server SSDs needing high performance and storage density. The memory devices feature a 15.47Gb/mm2 areal density, according to Yole Group, and a 2400 MT/s interface. 3D NAND devices featuring such a speedy I/O can be used to build PCIe Gen5 x4 drives that offer throughput of over 12 GB/s, and such drives are still not common.

Yet, because the U.S. government blacklists YMTC, international SSD makers are reluctant to use its advanced 3D NAND memory for their products due to possible risks.

Unihiker is a $79 Single Board Computer (SBC) from DFRobot. If that name sounds familiar, then it should. DFRobot are the makers of the LattePanda 3 Delta and LattePanda Sigma. From this pedigree we expected great things, but the $79 price point is tough and something had to be cut.

Unihiker has a 2.8 inch resistive touch screen dominating the board, but under the hood we have a competent quad core Arm A35 CPU and 512MB of RAM. This isn’t a desktop PC, but neither is it “just” a microcontroller (in fact the MCU is based on RISC-V)

What can it do? Who is it for? And should we buy one? To answer those questions and learn more about this board we need to put it on the bench.

Unihiker Specifications

Unihiker Look and Feel

Bearing more than a passing resemblance to an early 2000s PDA, Unihiker is small in the hand. The 2.8-inch screen dominates the front of the board. The product page claims that the screen is a touch screen and initially we couldn’t get it to work. Looking through the config options on the device we spotted a calibration tool. A few clicks of the resistive touch screen and we had a working user interface. Moving around the back we have the electronics that make up the board. These include GPIO connections via “Gravity” ports. These ports resemble Grove / Stemma QT connections, but they are not directly compatible, more on this later. Power and data connectivity is provided by a single USB C port. The extra USB A port is for a keyboard / mouse or even a USB webcam.

Unihiker is a Linux PC

It may look like just another screen and microcontroller, but Unihiker is really a quad-core Arm PC with 512MB of RAM and 16GB of eMMC storage. This is by no means a powerhouse, but it is comparable to a Raspberry Pi 3B+. The board has plenty of power considering the projects it can be used with.

The Unihiker website has tutorials covering machine learning and image classification, robotics, weather stations and humble GPIO projects. Being a Linux PC, and with VNC server capabilities, we could use this as a desktop computer, but it is not the intended use case.

Getting Started with Unihiker

Plug in the board and you can elect to try out the ready-to-go projects on the board, or follow the tutorials. A long press of the Home button loads up the menu and from there we tried out the demos. The simple “Hello World” shows off how the screen works, Charts show how to plot data onto the screen. The most interesting demos being the Spirit Level (using the accelerometer) and a Face Tracking demo which requires a USB webcam to track faces and places rabbit ears or a wizard hat on the target.

All of the demos provide a good introduction to the board, and show how versatile it can be. But the menu isn’t just for demos. From the menu, we can set the board up as a Wi-Fi hotspot, get system information, calibrate the touch screen and toggle services. This last option, Services, is where we find a number of extra features. From here we can set up a Jupyter notebook (a web based interactive computing platform), SIoT (using MQTT), file sharing and screen sharing.

Programming Unihiker is primarily done via Python, but there are a few different ways to do it. At the beginner level we have Mind+, a colorful and easy to use IDE that features both block and text based Python programming options. Mind+ can also remotely connect to Unihiker via USB. By doing this we can run our code directly on the board, well you could.

We encountered a few issues writing Python projects in the IDE. Block based projects worked flawlessly, but writing Python directly in the IDE produced syntax errors and an anomalous file path that caused Python to crash. This had us stuck for a while, so we used SSH to connect and ran the same code, no errors. It looks like the IDE was adding illegal characters to the file path. Using SSH we wrote all of the test code for the review, and it all worked with no issues.

Unihiker has a range of “services” that we can activate, and we took a look at Jupyter and SIOT. Jupyter notebooks merge code and documentation into one page. We can explain a programming concept or problem, then use an embedded section of Python code to illustrate.

Using a Jupyter notebook we wrote a quick script to show text on the display, and to blink an LED connected to GPIO23. The code worked, and the notebook was accessible by anyone on our home network. We also tested SIOT, essentially this is a local MQTT server, useful for the classroom and for keeping your data off the Internet. We were able to send messages to and from Unihiker using a mixture of topics.

Using Microsoft’s VSCode with Unihiker is a breeze, we are merely connecting to the board using VSCode’s SSH extension and from there we can write code directly to the onboard eMMC. Running code from a terminal and we can see the output on the screen. Very easy and the Unihiker documentation is easy to follow.

GPIO Access

Unihiker has a GPIO, provided by the onboard RISC-V MCU, but unless you have bought into the Gravity components system, you are left hanging. The Gravity connectors are merely polarized connectors used to make quick connections. Similar in function to Qwiic / Stemma QT / Grove connectors, Gravity connectors are not compatible with the other formats.

Yes you could force a Grove connector into the I2C Gravity ports, but it wouldn’t work as the pinout is incorrect. The three pin versions breakout one GPIO pin (21,22,23,24), 3V and GND. Whereas the four pin versions are reserved for I2C. But what if we want to use conventional components with Unihiker? You could insert jumper jerky strips into the supplied Gravity cables. It works but it will damage the connectors over time. A better means is to use the edge connector located at the bottom of the board.

The edge connector may look a little out of place. We’ve seen them before on the micro:bit and Adafruit’s Clue board. The Unihiker’s edge connector looks to have the same pinout as the micro:bit and this got us thinking about micro:bit breakout boards. Luckily we have a few and used Pimoroni’s Pinbetween to break out the GPIO pins.

A few lines of Python and we had a blinking LED. Modifying the code to control two GPIO pins, we connected a Kitronik motor control board to Unihiker, supplied 6V of power for the DC motor and then saw the motor spring into life.

The onboard sensors (sound, light, acceleration and orientation) are easily used thanks to the pinpong and Unihiker Python modules (both Blocks and text). With the modules we easily queried the values from the sensors and with an additional few lines of Python we had the data updating on the display.

Sadly, GPIO access is where Unihiker falls down. We have to either buy into one set of bespoke components or another. A proper GPIO breakout would’ve been nice, but alas it was never meant to be. Maybe in version 2? Should there be one?

Who is Unihiker for?

This is a great question. A small Linux SBC with an embedded screen has lots of applications. It could be a great user interface for a machine learning project, control devices across a network and gather data from sensors. If we’re looking at it as a GPIO focused board, we feel a little let down.

Sure we have multiple ways to write code (Mind+, SSH, Jupyter, SIOT etc) and we love that, but the only means to access the GPIO is via Gravity ports, or using a micro:bit breakout board. Sure, we can buy into the Gravity format, and for educational purposes this could be an option. For hobbyist users, we will want to access the GPIO using accessories we have on hand and that will be difficult unless we spend a bit more money.

Bottom Line

Unihiker offers a fun experience, once you get your head around the form factor and the means to code it. The face tracking demo is especially interesting. What lets the board down is the GPIO. For the $79 price tag, we appreciate that it cannot be a “one size fits all” board considering it is a Linux PC with a built-in display.

If you are coming into this completely fresh, then the Unihiker board is an interesting introduction to IoT. It is easy to use and the compact form factor lends itself to portable projects. The embedded sensors (sound, light, acceleration) are really easy to access via the Python module, and that means quick results for eager learners. We enjoyed our time with Unihiker, but the GPIO limitations leave us wanting more from the next model.

Should you buy one? If you need GPIO, then a Raspberry Pi Pico W or Raspberry Pi Zero 2 W is lower cost and provides many more possibilities. That said, the touch screen works well, the hardware is solid and the software support means we can build projects quickly.

Users looking to expand their storage on a budget should take a close look at this offer from Newegg on the Solidigm P41 Plus 2TB SSD. This SSD has been going for around $79 as of late but today is discounted to just $69.

We reviewed the Solidigm P41 Plus late last year and overall regarded it as mid-level SSD. It doesn’t have the greatest performance on the market but if you’re looking for more space and don’t want to overspend, this deal is a great opportunity.

Today’s discount applies to the 2TB model but is also available at 500GB and 1TB capacities. All of the drives in this line feature an M.2 2280 form factor and connect using a PCIe 4.0 interface. The 2TB edition can reach read/write speeds as high as 4125/3325 Mbps.

The Solidigm P41 Plus SSD features an SMI SM2269XT and uses 144-Layer Solidigm QLC flash memory. The purchase is supported by a limited 5-year manufacturer’s warranty from Solidigm that will void if the drive reaches 800 TBW. It’s also backed by Newegg’s 30-day return policy.

Visit the Solidigm P41 Plus 2TB SSD product page at Newegg for purchase options. As of writing, it’s not clear how long the deal will be made available.

Today at Newegg, users can find the Team Group T-Force Cardea 1TB SSD at Newegg for its best price to date. This was already a reasonably priced SSD going for around $49 but this discount takes the price down to $34. It might not be the fastest SSD on the market but for users looking to upgrade their storage on a budget, it’s still fairly notable performance-wise.

As of writing, the drive has been placed on backorder but you can still order it for the discounted rate. We’re not sure when the restock will take place but at this price, it should be well worth the wait.

This discount applies to the 1TB model but there are other capacities available in this line, as well, including 500GB, 2TB and 4TB. All of the drives in this line have an M.2 2280 form factor but the 1TB can reach read/write speeds as high as 3500/3000 Mbps.

The Team Group T-Force Cardea 1TB SSD uses 3D NAND TLC memory and connects using a PCIe Gen 4 x4 interface. It’s supported by a limited 5-year manufacturer’s warranty from Team Group as well as Newwegg’s 30-day return policy.

Visit the Team Group T-Force Cardea 1TB SSD product page at Newegg for more details and purchase options.

As demand for PCs and smartphones remains tepid, prices of commodity 3D NAND and DRAM continue to slide, leading to deepening losses for manufacturers. But in its second quarter financial results SK Hynix now says that it sees recovery of the memory industry thanks to increasing demand for HBM2 and HBM3 memory used by artificial intelligence (AI) servers as well as DDR5 SDRAM that are sold at a premium.

"Amid an expansion in generative artificial intelligence (AI) market, which has largely been centered on ChatGPT, demand for AI server memory has increased rapidly," a statement by SK Hynix reads. "As a result, sales of premium products such as HBM3 and DDR5 increased, leading to a 44% sequential increase in revenue for the second quarter, while operating loss narrowed by 15%."

SK Hynix earned 7.306 trillion won ($5.729 billion) in Q2 2023, down 47% year-over-year, but up 44% quarter-over-quarter. The company's net loss achieved 2.988 trillion won ($2.343 billion), an increase of 16% quarter-over-quarter. Meanwhile, due to increased demand for premium types of memory, the company's gross margin increased from -67% in the first quarter to -39% in the second quarter.

"Having passed the trough in the first quarter, the memory semiconductor market is seen to have entered the recovery phase," said Kim Woohyun, Vice President and Chief Financial Officer (CFO) at SK Hynix. "SK Hynix will strive to prop up earnings through its technological competitiveness in high-end products."

SK Hynix also reported an increase in sales of both DRAM and NAND commodity products for the second quarter. Despite a drop in the price of generic DRAM products like DDR4 due to lower demand from the PC and smartphone sectors, the average selling price (ASP) of DRAM augmented due to higher sales of high-end products intended for AI servers. The diminished inventory valuation loss, coupled with the company’s ongoing efforts to cut costs, resulted in a reduced operational loss. 

In the quarterly earnings teleconference, SK Hynix projected a stable demand for AI memory and anticipates a more pronounced effect of production cuts by memory manufacturers, which is expected to lead to price increases. The company also plans to bolster sales of its high-end DRAM products, including HBM3, DDR5, LPDDR5, and 176-layer3D NAND-based SSDs in a bid to increase revenue.

The manufacturer reiterated its commitment to decrease CapEx  budget by at least 50% relative to 2022, but affirmed that the company will continue investing to increase production capacity of high-density DDR5 and HBM3 to facilitate future market growth. 

"The company's stance on consolidated investment, which is to reduce that by at least 50% compared to 2022, remains unchanged, but with resources secured by management efficiency, we will continue to invest to expand the production capacity of high-density DDR5 and HBM3 that will lead future market growth," said Kim.

While the company is expected to ramp production of 3D NAND and DRAM based on its latest process technologies — 238-layer 3D NAND for flash and 1-beta for DRAMs — by the end of the year, due to an excess inventory of 3D NAND, the company opted to further reduce 3D NAND production in the coming quarter.

The Netac NV7000 is a passable PCIe 4.0 SSD that could work in a desktop PC or especially a PlayStation 5, but unlike the drives on our list of best SSDs, its inconsistent results and current pricing leave much to be desired, meaning it will have to compete on the value front to make its mark.

The Netac brand is perhaps known better outside of the U.S., but it is expanding with many products on Amazon and may become a future player here. The NV7000 is one of its flagship drives that will be expanded in the future with the NV7000-t, a DRAM-less drive that can hit 7.3 GB/s in sequential workloads, reminiscent of the Acer Predator GM7.

We’re specifically looking at the 4TB model today, which would be a good fit for the PS5 if you want the maximum amount of internal storage. In fact, that’s where we feel this drive works best, but it has to be priced right to make sense. We caution that lower capacity models of this drive, particularly those found in other regions, could come with different hardware. This drive most commonly uses the InnoGrit IG5236 controller with either Micron’s TLC as on the HP FX900 Pro, SK hynix’s TLC as found on the SK hynix Gold P31, or YMTC flash, as we saw on the Predator GM7. The 4TB drive we have today instead uses the ubiquitous Phison E18 with Micron’s decent TLC.

This hardware combination is relatively safe, but the NV7000 demonstrated some performance inconsistencies. That means it’s probably not the ideal drive to buy if you regularly execute heavier workloads. Let's take a closer look. 

Specifications

The Netac NV7000 is available in the 1TB, 2TB, and 4TB flavors. It’s spec'd up to 7,200 / 6,850 MB/s for sequential reads and writes and 940,000 / 1,000,000 random read/write IOPS. Current pricing is $94.99, $109.97, and $333.99 on Amazon with coupons applied. This is on the high side even when accounting for the included heatsink, but the drive would work well in a desktop PC or the PS5. The drive is warrantied for five years and 640TB, 1280TB, and 3000TB of endurance (TBW) for the respective capacities.

Software and Accessories

Netac offers a download for an SSD toolbox on its website. This application is rough around the edges, but it works. It shows the general health of your SSDs, including SMART information, has a place for “migration,” which should clone a drive, allows for firmware updating, and also has the secure erase function. It also has a toggle for TRIM support. The utility is pretty basic, so we recommend using different software, if possible. 

A Closer Look

The NV7000 comes with an aluminum heatsink that’s attached to both sides of the drive with thermal padding. It’s easy enough to remove, if necessary. The drive itself has an SSD controller on the top side, one DRAM package per side for a total of two, and four NAND packages per side for a total of four. A PMIC is located next to the controller.

The Phison E18 SSD controller has proven to be a winner, although, at some point, it may be superseded by less expensive and more efficient designs. It’s backed by two 8Gb DDR4 chips for metadata, which is 2GB in total. This does not meet the best ratio of DRAM to NAND but is more than enough for this type of drive. The SSD also has eight 4Tb NAND packages, each with eight 64GB dies of Micron’s 176-Layer TLC (B47R). This is a lot of dies to address; something ameliorated with Micron’s denser 232-Layer TLC.

MORE: Best SSDs

MORE: Best External SSDs and Hard Drives

MORE: How We Test HDDs And SSDs

MORE: All SSD Content

Comparison Products

Although the NV7000 has fairly common hardware, we want to test it against a range of drives to see if it has any nuances. This list of SSDs includes the Rocket 4 Plus-G, the Rocket 4 Plus, the Rocket, the Rocket Q, and the Rocket Q4, all from Sabrent. We also have the Seagate FireCuda 530, the Team A440 Pro, the Team Z44Q, and the WD Red SN700.

Trace Testing - 3DMark Storage Benchmark

Built for gamers, 3DMark’s Storage Benchmark focuses on real-world gaming performance. Each round in this benchmark stresses storage based on gaming activities including loading games, saving progress, installing game files, and recording gameplay video streams.

The NV7000 performs well enough in 3DMark, although not quite as well as its peers that share the same hardware. The Sabrent 4 Plus-G leads the overall score category with a 12% higher score. 

Trace Testing – PCMark 10 Storage Benchmark

PCMark 10 is a trace-based benchmark that uses a wide-ranging set of real-world traces from popular applications and everyday tasks to measure the performance of storage devices.

The story is the same with PCMark 10. This drive is clearly a leap over the previous generation of drives and has more bandwidth than less expensive PCIe 4.0 options, but it isn’t compelling against its direct competitors.

Transfer Rates – DiskBench

We use the DiskBench storage benchmarking tool to test file transfer performance with a custom, 50GB dataset. We copy 31,227 files of various types, such as pictures, PDFs, and videos to a new folder and then follow-up with a reading test of a newly-written 6.5GB zip file.

The NV7000 does shine during file copy workloads, however, going toe-to-toe with the exceptional Rocket 4 Plus-G. The 4 Plus-G is a refined version of the original 4 Plus, using Phison’s I/O+ firmware. This firmware is designed for DirectStorage, which includes sustained workloads.

The Seagate FireCuda 530 also recently received this same firmware update, but it is not applied to our test result in the list. The firmware has not been prominent on other E18 drives, however, it is present on all Phison E26 drives like the Crucial T700, the Gigabyte Aorus 10000, the Seagate FireCuda 540, and the Corsair MP700.

Synthetic Testing - ATTO / CrystalDiskMark

ATTO and CrystalDiskMark (CDM) are free and easy-to-use storage benchmarking tools that storage vendors commonly use to assign performance specifications to their products. Both of these tools give us insight into how each device handles different file sizes.

Sequential performance is quite good on the NV7000 in both ATTO and CDM. Random 4KB performance, on the whole, is not bad, but there is a noticeable dip for low queue depth random read latency.

Sustained Write Performance

Official write specifications are only part of the performance picture. Most SSDs implement a write cache, which is a fast area of (usually) pseudo-SLC programmed flash that absorbs incoming data.  Sustained write speeds can suffer tremendously once the workload spills outside of the cache and into the "native" TLC or QLC flash. 

We use Iometer to hammer the SSD with sequential writes for 15 minutes to measure both the size of the write cache and performance after the cache is saturated. We also monitor cache recovery via multiple idle rounds.

The NV7000 writes at high speed in its pSLC mode before dropping down to a slower TLC mode, and then eventually struggles to an extremely slow state that brings down its steady state write performance. Its cache is relatively conservative, similar to the other E18 drives on the chart, so its overall performance is lower than expected. The precise speed is less important than the pattern of inconsistency that we see. In normal use, this won’t be a problem, but it does suggest Netac is doing something different, especially when considering the uneven results from other benchmarks.

We have seen some performance inconsistencies like this before on drives that use lower-quality but still consumer-grade flash. After all, you’re not going to hit these endurance numbers with bad NAND. However, there can be performance issues when stress testing, particularly if you’re trying to ensure sufficient write endurance for warranty. This could lead to higher read latency, too. This is pure speculation, but it will not impact performance too deeply, particularly on the PS5. However, it does mean this drive needs to be a less expensive alternative to a drive like the FireCuda 530, for instance.

We will add that this type of performance result is not restricted to lesser-known brands like Netac, as we also had lower write performance with the 2TB Team A440 Pro. Micron confirmed that the flash used on that drive, which we detected as having lower rated endurance, is completely legitimate and could find its way onto any number of drives. That could ostensibly include drives destined for lower-intensity work, like PS5 drives. Our advice is that if an E18 drive seems too inexpensive to be true, it’s probably one that is best fit for console use.

Power Consumption

We use the Quarch HD Programmable Power Module to gain a deeper understanding of power characteristics. Idle power consumption is an important aspect to consider, especially if you're looking for a laptop upgrade as even the best ultrabooks can have mediocre storage.

Some SSDs can consume watts of power at idle while better-suited ones sip just milliwatts. Average workload power consumption and max consumption are two other aspects of power consumption, but performance-per-watt is more important. A drive might consume more power during any given workload, but accomplishing a task faster allows the drive to drop into an idle state more quickly, ultimately saving energy.

For temperature recording we currently poll the drive’s primary composite sensor during testing with a 24C ambient.

The NV7000’s power efficiency is merely average. We’re used to seeing far more efficient drives these days, but this is not a drive you should buy for a laptop anyway.

Temperature-wise, the NV7000 idled at 33C and hit a maximum of 75C during testing. This is below the throttling temperature and is quite good, considering the workload. This drive appears designed for the PS5 and would be fine with the included heatsink, but it would also work in a high-end desktop PC.

Test Bench and Testing Notes

We use an Alder Lake platform with most background applications such as indexing, Windows updates, and anti-virus disabled in the OS to reduce run-to-run variability. Each SSD is prefilled to 50% capacity and tested as a secondary device. Unless noted, we use active cooling for all SSDs.

Conclusion

The Netac NV7000 is a run-of-the-mill high-end PCIe 4.0 SSD, excelling in no one area and coming with some performance inconsistencies. We don’t necessarily think this is down to Netac, and it may be a wider problem due to how the flash and SSD markets are currently spiraling. It’s become commonplace for even reputable manufacturers to change the underlying SSD hardware whenever it’s convenient, whether that's the controller, flash, or both. This is extremely typical outside of Western markets but is essentially worldwide at this point, with perhaps some nervousness for Maxio controllers and YMTC flash in the U.S.

Netac and other lesser-known brands are making a push into other markets, including Amazon. They often use comparable hardware to better-known brands, but to compete, they still ultimately have to offer drives at a lower price point. It’s not uncommon to get very good prices if you can wait through other avenues like AliExpress, too. It’s worth looking at these drives as, quite honestly, the impressive results we saw with the Predator GM7 show that there’s still room for growth with PCIe 4.0 SSDs. That may include the soon-to-be-released NV7000-t, but the base NV7000 is largely disappointing at this stage.

We would recommend this drive for use in the PS5 to get extra storage that will effectively be as fast as it gets. We'll need to see a price reduction for that to make sense, though. It’s also probably not our first choice for desktop PC or workstation use, given the plethora of better options. We also know the lower-capacity NV700s have shown up with different hardware, and, although largely okay, some caution is necessary — particularly as the alternative IG5236 controller has known issues with YMTC TLC flash. 

If you’re looking for an alternative but sticking with this class of drive, you’re better off with the Sabrent Rocket 4 Plus-G, Sabrent Rocket 4 Plus, Seagate FireCuda 530, or similar drives that have very high sustained sequential write performance.

MORE: Best SSDs

MORE: Best External SSDs and Hard Drives

MORE: How We Test HDDs And SSDs

MORE: All SSD Content

Some of the best SSDs, specifically the PCIe 5.0 drives based on the Phison PS5026-E26 controller, have been crashing instead of thermal throttling when operating without a cooler. However, it is worth noting that all of the impacted drives are designed and marketed to specifically be used with a heatsink, so the conditions that expose the thermal shutdown will not present themselves if the drives are used correctly (in accordance with manufacturer specifications).

Initially, only the Corsair MP700 exhibited this behavior; however, it's now become apparent that the issue is more widespread and affects other Phison E26-based SSDs that are designed for heatsinks when they are used without heatsinks attached.

German news outlet Computerbase discovered that the Seagate FireCuda 540, Gigabyte Aorus Gen5 10000, and Adata Legend 970 also suffer from the shutdown issue. It was to be expected since the PCIe 5.0 SSDs utilize the same Phison E26 controller. The FireCuda 540, Aorus Gen5 10000, and Leged 970 are still on firmware 22. Seagate hasn't gotten back to Computerbase when the new firmware will be available, whereas Gigabyte promised that it'll arrive "soon."

To recap: the issue only occurs when you use a PCIe 5.0 SSD without cooling. When the drive gets too hot, it shuts down to protect the SSD controller, NAND, and data. This shouldn't be an issue if the PCIe 5.0 drive is adequately cooled by the included heatsink or the M.2 heatsink from the motherboard. Regardless, Phison has released a new firmware (version 22.1) that ensures the PCIe 5.0 SSDs throttle — instead of just crashing, which can lead to data loss. 

Firmware 22.1 introduces link-state thermal throttling that essentially reduces the PCIe interface speed — for example, dropping from PCIe 5.0 to PCIe 4.0 or even PCIe 3.0 to lower the temperature of the physical layer (PHY) without throttling the processor clock. This will obviously impact the PCIe 5.0 SSD's performance, but it should also keep it from engaging a shutdown to protect the integrity of the SSD controller. According to Computerbase, the thermal threshold on the new firmware 22.1 is 85 degrees Celsius.

In the beginning, the Crucial T700 didn't have the issue. The PCIe 5.0 would throttle until the drive operated at hard drive speeds but didn't shut down thermally. Computerbase performed more tests on the Crucial T700, and it showed similar failures, so the drive may also need the firmware 22.1 update.

Under the new firmware 22.1, Phison E26-powered SSDs should still provide an acceptable level of performance at high temperatures. Without a cooler, the Corsair MP700 delivered sequential read and write speeds over 10 GB/s and 2 GB/s, respectively. Remember that it's a safety measure in case temperatures get out of hand, and you should always use a cooler with your PCIe 5.0 SSD.

Edit: 7/21/2023 3:45pm PT: Clarified that these SSDs are specifically designed to be used with a heatsink. 

Solidigm has introduced its D5-P5336 drives that boast capacity of up to 61.44TB and feature a PCIe 4.0 x4 interface to enable read speeds of up to 7,000 MB/s. The device is designed to ultra-high-density read-intensive storage applications, such as AI inference on the edge or content delivery. As of now, the D5-P5336 family of solid-state drives that offer extreme capacity and great performance.

"Modern workloads like AI and capabilities like 5G are rapidly reshaping the storage landscape," said Greg Matson, VP of Strategic Planning and Marketing at Solidigm. "Businesses need storage in more places that is inexpensive, able to store massive data sets efficiently and access the data at speed."

Solidigm's D5-P5336 SSD is based on SK Hynix's inexpensive 3D QLC NAND memory as well as a proprietary platform that ensures strong performance and very good endurance in addition to a rather extreme capacity. Solidigm says that its D5-P5336 provides sequential read speed of up to 7,000 MB/s, sequential write speed of up to 3,300 MB/s as well as random read/write speed of up to 1.005 million/38K 4K IOPS, which is rather good given that we are talking about drives that can store up to 61.44TB of data.

"The D5-P5336 delivers on all three — value, density and performance," said Matson. "With 3D QLC, the economics are compelling — imagine storing 6X more data than HDDs and 2X more data than 3D TLC SSDs, all in the same space at 3D TLC speed.

One of the distinctive features of Solidigm's D5-P5336 is their endurance (up to 65 PBW for top capacity drives), which is said to be in line with that of 3D TLC NAND, which is used by competing high-capacity drives that either cannot match D5-P5336 in terms of performance or in terms of capacity.

"For years there has been debate about endurance and reliability of SSDs, QLC in particular, but Solidigm might have ended that debate with the D5-P5336," noted Avery Pham, VP Operations, VAST Data. "Any number of applications will see notable benefits from these drives from AI and machine learning to object storage."

Solidigm will offer its D5-P5336 in three form-factors: U.2/U3 15mm, E3.S 7.5mm, and E1.L 9.5 mm with capacity points ranging from 7.68TB all the way to 61.44TB. Only the U.2/U.3 and E1.L drives will provide the 61.44TB capacity.

SSD's these large typically feature a SATA interface these days since most of them are designed to substitute for hard drives in applications that do not exactly require extreme performance, but rather need to just increase storage density. But applications like AI deployments on the edge have very specific requirements for storage density (as high as possible) and performance (as fast as possible) and SSD makers have to offer something new here. Apparently, Solidigm just beat its competitors to punch with a 61.44TB drive for these emerging applications.

The Chinese company Yangtze Memory Technologies (YMTC) seems to be adapting to new US sanctions realities with a rather mysterious 3D NAND product that combines a relatively low storage density with an extremely speedy interface. In addition, YMTC has once again raised prices for its 128-layer 3D NAND by about 5%, according to DigiTimes. This move comes despite the ongoing slump in the PC and consumer electronics markets in general and the 3D NAND memory market in particular and follows a previous increase in May.

Mysterious New Product

Although stringent U.S. export rules have hindered YMTC's ability to produce 3D NAND flash chips with more than 128 layers, the company is swiftly adopting alternative strategies. In particular, YMTC has initiated large-scale production of its 128-layer NAND flash chips, dubbed 'Wudangshan,' that reportedly relies on the Xtacking 3.0 architecture, according to the report.

YMTC has productized a version of 128-layer flash before with its Xtacking 2.0 architecture, but the step up to the 3.0 revision marks a significant overhaul of an existing product, which normally wouldn't take place. If true, the US sanctions, which hinder YMTC's ability to make flash with higher layer counts, have spurred the company to re-vamp its older flash with performance-boosting tech. If such 3D NAND devices exist, they may combine relatively low storage density with high performance, which could be the company's stop-gap solution before it comes up with a product that offers both high capacity and high performance. 

The key point of Xtacking is to produce a 3D NAND array and various peripheral logic on separate wafers using the most efficient memory and logic process technologies in YMTC's possession. Xtacking 3.0 introduces back-side source connect (BSSC) for the memory cell wafer, which simplifies production flow and reduces costs, according to TechInsights. YMTC developed Xtacking 3.0 for its 232-layer 3D NAND to bring together ultimate storage density (15.47Gb/mm2, according to Yole Group) and a 2400 MT/s interface, therefore offering perhaps the world's best 3D NAND.

 It is theoretically possible to add a CMOS wafer with high-speed peripheral logic to a 128-layer 3D NAND array wafer, but the question is whether it brings benefits for real-world applications because there is only so much one can do with a 128-layer process technology and 3D NAND device capacity from a costs perspective.

In fact, based on Yole Group's findings, YMTC's 232-layer 3D NAND uses string stacking to achieve 232 layers, with 128 layers on the 'first deck' and 125 layers on the 'second deck'. From a manufacturing point of view, this means that the company may simply not add the 'second deck' to get to a 128-layer 3D NAND featuring the Xtacking 3.0 architecture. But, of course, this severely hurts the storage density of the final device and its competitive position.

Yet, it could make sense for the company to produce such memory if it wants to address specific applications and/or as a stopgap before it finds a way to add the 'second deck' without using equipment that it cannot procure or service.

YMTC has not formally confirmed the very existence of 128-layer 3D NAND with Xtacking 3.0 codenamed Wudangshan, so take the information with a grain of salt though.

5% Price Hike

YMTC recently raised its unit price for a 512Gb NAND device from $1.5 to $1.6, following an earlier hike from $1.35 - $1.4 in May. This pricing strategy contrasts with international NAND suppliers who are grappling with attempts to raise prices amidst tepid customer demand. But Yangtze Memory seems to be taking advantage of its position now that Chinese government agencies cannot buy memory from Micron.

DigiTimes claims that thanks to substantial financial backing from the China IC Industry Investment Fund (the Big Fund), YMTC is seeking solutions for upstream equipment and materials. While the report does not divulge whether YMTC is seeking to use domestically produced wafer fab equipment, it is certainly a possibility.

Apparently, the company's stockpile of consumables, accumulated before the sanctions, is anticipated to sustain the mass production of 128-layer 3D NAND products for at least three more years, the report says. Meanwhile, it remains to be seen whether 128-layer 3D NAND will still be competitive three years down the road. Of course, the Chinese government may force certain organizations to only use YMTC's 3D NAND for their systems, which will guarantee stable demand, but it remains to be seen whether that demand will be significant enough to support the company economically.

Following months of negotiations, Western Digital and Kioxia are on the brink of finalizing a merger agreement, according to Bloomberg, which cites sources with knowledge of the matter. The agreement involves spinning off Western Digital's NAND flash business and combining it with Kioxia.

The merger details remain confidential, but the report claims that Western Digital shareholders would gain control of slightly more than half of the joined company. As for the daily management of the combined entity, it has been suggested that it will be led primarily by Kioxia's executive team. However, Western Digital executives would also play a significant role. 

Notwithstanding the advanced negotiations, the final agreement is not yet in place, and the timeline could change. In fact, discussions may conclude without an agreement at all. 

The merged company will have a dual board representation system with both chipmakers contributing members. The company will be based in Japan and initially listed on the Nasdaq, with plans to eventually list in Tokyo. Meanwhile, Bain Capital, a major investor in Kioxia, would receive a special dividend. 

Despite producing 3D NAND at the same fabs, Kioxia and Western Digital, each maintains their distinct business strategies, with Kioxia's primary focus being on sales of the memory chips. In contrast, Western Digital's approach leans towards supplying finished NAND-based products, including some of the best SSDs available.  

As of Q1 2023, Kioxia commanded a 21.5% stake in the NAND memory, while Western Digital secured a 15.2% share, according to TrendForce. On the other hand, Samsung distributed 34% of NAND memory during the same period. With this in mind, the prospective merged entity of Kioxia and Western Digital's flash businesses is set to become the leading supplier of NAND memory worldwide, with over 36.7% of the market share. 

Raspberry Pi has long been the most famous in the world of single-board computers. But there are plenty of alternatives, such as LattePanda and Khadas, and of course, BeagleBoard, which has announced the availability of its RISC-V powered BeagleV-Ahead.

Coming in at $149, the BeagleV-Ahead is powered by a 2 GHz, quad-core RISC-V 64GCV Xuantile C910 CPU and has 4GB of RAM and 16GB of eMMC flash storage on which you can install either Ubuntu or Yocto Linux. The OS is flashed to the board, and we then remotely interface with the board via a terminal or web interface.

The big draw for this board is the RISC-V CPU. In the maker community, RISC-V is gaining popularity, mainly due to its open-source nature and availability in several different price points and packages. This includes Milk-V, which offers three RISC-V machines starting from $9. You can even get RISC-V-powered soldering irons now!

Beagle boards have used Arm-based CPUs in the past, but the switch to an open-source RISC-V CPU is the only significant change between the boards. The distinct BeagleBoard form factor remains, meaning "capes", BeagleBoard's name for add-on boards (think Raspberry Pi HATs or Arduino Shields). Capes connect to the board using the P8 and P9 header, essentially the same as the Raspberry Pi's GPIO. But with the BeagleV-Ahead, we get six analog inputs and a plethora of PWM, I2C, UART, SPI, I2S and good old digital IO.

The BeagleBoard team has a documentation page up; from there, we can delve deeper into the BeagleV-Ahead. There are a few gaps, but this board is just a few days old, so documentation will eventually catch up.

We first reported on this board in 2021, and in the two years since, there has been a pandemic and a global chip shortage which has impacted the delivery time of the BeagleV. That said, it is good to see another RISC-V board being offered to an enthusiastic maker community.

The BeagleV-Ahead is available from several distributors, with an average price of $149. 

The WD Blue SN580 is the consummate budget NVMe SSD, taking a winning formula and making it better. It performs well enough with a fair range of capacities and solid support, making it a solid contender for our list of best SSDs. It’s relatively efficient, and the single-sided design makes it great for laptops. It doesn't bring anything new to the table, but it’s an easy pick-up for a budget drive if you’re looking for a cheap upgrade that won’t disappoint — assuming the price is right.

The success of the Blue SN570 and Black SN770 has prompted WD to revamp the former with the Blue SN580. The SN580 is essentially an updated SN570, much as the SN570 was an updated Blue SN550, which succeeded the Blue SN500. The SN550 was very popular, but WD later changed its flash, altering performance negatively. That brings us back to the SN580, which jumps up to a PCIe 4.0 interface for more bandwidth, uses a new caching scheme for better burst performance, and generally maintains or improves on the SN570 in every meaningful way.

The warranty remains the same, and WD’s software selection is also the same. However, the SN580 is more efficient than the SN570, which is always a welcome development. The drive also borrows from the Black SN850 with a new caching scheme, an improvement for most over the SN570’s relatively small cache. Perhaps the biggest ding against the SN580 is that it’s otherwise uninspiring; it sets no new records and isn’t even as fast as existing 4.0 drives in its class. It’s certainly better than the Black SN750 SE, but that’s not saying much. Still, it is a safe buy. 

Specifications

The WD Blue SN580, or simply SN580 from here on out, is available at 250GB, 500GB, 1TB, and 2TB. This matches the SN570, although the 2TB model came out later for that drive, likely because it required 1Tb BiCS5 dies. Also matched is the endurance at 600TB written per TB of flash - excluding 900TB for 2TB - which is a good amount, especially for a budget drive. The drive does not support hardware encryption and is warrantied for five years. At the respective capacities, the MSRPs are $27.99, $31.99, $49.99, and $109.99. Actual selling prices will likely be lower than these to be competitive against the SN770.

The SN580 can achieve up to 4,150 / 4,150 MB/s for sequential reads and writes and 600K / 750K IOPS for random reads and writes. The former is ostensibly an improvement over the SN570, but the relatively small amount of gain is from the switch to a 4.0 interface rather than any wild changes under the hood. The SN580 is only available in the M.2 2280 form factor, but the OEM variant of the drive - probably the Blue SN5000 - comes in 2242 and 2230 under the SN5000S label.

Software and Accessories

WD provides two downloads for its SSDs. The first is the Western Digital SSD Dashboard, or toolbox, which is useful for monitoring the health of the drive, handling firmware updates, and for engaging other utilities. The second is a version of Acronis True Image for backing up data.

A Closer Look

The SN580 has an SSD controller, one NAND package, and a power management integrated circuit (PMIC). WD has traditionally used Active-Semi PMICs on its Blue SSDs, a manufacturer now owned by Qorvo.

The drive is DRAM-less, but WD has done well without DRAM on the SN570 and SN770 alike. The original SN500 arrived without HMB explicitly enabled, and it also performed fairly well. Possibly this is due to having more SRAM available for metadata and mapping within the ASIC itself. In any case, HMB is adequate for the intended “light” workloads.

The drive uses WD’s proprietary controller technology under the SanDisk label. This four-channel controller is similar to what’s found on many of WD’s other NVMe SSDs, including the Green SN350, Blue SN500, Blue SN550, Blue SN570, and Black SN770. It appears to be running at a lower bus rate than the SN770, perhaps necessary to help differentiate the drives. The flash is the same 112-Layer TLC (BiCS5) used to good effect on the SN570 and SN770.

The SN580 comes with nCache 4.0, which we first saw on the WD Black SN850. This large, hybrid cache differs from the cache on the launch SN570 and previous Blue NVMe SSDs. The original Blue cache was static, which limited its size but resulted in decent TLC write performance post-cache. WD has settled for higher burst performance with the SN580, which helps set it apart from the SN570. This can also improve efficiency as many tasks will finish faster. The SN770’s cache is slightly different as it evokes a middle-ground TLC mode, but we may see an updated mid-range Black with nCache 4.0 in the future.

MORE: Best SSDs

MORE: Best External SSDs and Hard Drives

MORE: How We Test HDDs And SSDs

MORE: All SSD Content

Comparison Products

The WD Blue SN580 is similar to WD’s other NVMe drives, so we included the Black SN750, the Blue SN570, the Black SN770, and the Black SN850X in our test pool. For drives in this class from the prior generation, we have the Samsung 980, and at roughly the same sequential speeds as the SN580, we have the Solidigm P41 Plus. For the last generation of mid-range PCIe 4.0, we have the Sabrent Rocket Q4. For high-end 4.0 DRAM-less, which will become a popular segment soon, we have the Acer Predator GM7 acting as a stand-in. Lastly, for a high-end 4.0 drive, we have the SK hynix Platinum P41.

Trace Testing - 3DMark Storage Benchmark

Built for gamers, 3DMark’s Storage Benchmark focuses on real-world gaming performance. Each round in this benchmark stresses storage based on gaming activities, including loading games, saving progress, installing game files, and recording gameplay video streams.

The SN580 is merely average in 3DMark, but this is not a bad showing for a budget drive. The SN770 is clearly faster, though. 

Trace Testing – PCMark 10 Storage Benchmark

PCMark 10 is a trace-based benchmark that uses a wide-ranging set of real-world traces from popular applications and everyday tasks to measure the performance of storage devices.

The SN580 is also average in PCMark 10 with the SN770 maintaining its lead over it. This is roughly the showing we'd expect given the drives' intended use cases and pricing. 

Transfer Rates – DiskBench

We use the DiskBench storage benchmarking tool to test file transfer performance with a custom, 50GB dataset. We copy 31,227 files of various types, such as pictures, PDFs, and videos to a new folder and then follow-up with a reading test of a newly-written 6.5GB zip file.

DiskBench is one test where we’d expect the SN580 to clearly improve over the SN570 due to the former’s new, larger cache. We’d also expect the SN580 to end up close to the SN770. Both expectations were borne out as the SN580’s copy performance is quite good.

Synthetic Testing - ATTO / CrystalDiskMark

ATTO and CrystalDiskMark (CDM) are free and easy-to-use storage benchmarking tools that storage vendors commonly use to assign performance specifications to their products. Both of these tools give us insight into how each device handles different file sizes.

The SN580 keeps up well in ATTO before flatlining at the limitation of its bus. We see a similar “wall” with Crystal Disk Mark’s sequential results. This is not the drive for you if you’re going for raw bandwidth. Of more interest is the drive’s random performance being average to good, with read latency being very good indeed. It does fall behind the SN770, which is expected due to bus speed differences. 

Sustained Write Performance

Official write specifications are only part of the performance picture. Most SSDs implement a write cache, which is a fast area of (usually) pseudo-SLC programmed flash that absorbs incoming data.  Sustained write speeds can suffer tremendously once the workload spills outside of the cache and into the "native" TLC or QLC flash. 

We use Iometer to hammer the SSD with sequential writes for 15 minutes to measure both the size of the write cache and performance after the cache is saturated. We also monitor cache recovery via multiple idle rounds.

The SN580 writes at over 4 GB/s for over 85 seconds, suggesting a pSLC cache that spans the entire drive. All of the TLC is capable of pSLC mode, which means the cache will typically be about 1/3 of the available user space. The SN580 uses WD’s nCache 4.0, which is a hybrid cache with a small static portion and a very large dynamic portion, a combination that works well for the most typical workloads.

Once the cache runs out, the write speed plummets to a folding mode for a steady state speed of 545 MB/s. This is close to the SN770’s 563 MB/s, which makes sense as they have the same flash and amount of parallelization. The caches are also similar on these two drives, with the SN770 writing faster but for less time, along with it having a middle TLC state. The nCache 4.0 recovers fairly quickly, particularly the static portion, which improves the SN580’s feel over the SN570.

Power Consumption

We use the Quarch HD Programmable Power Module to gain a deeper understanding of power characteristics. Idle power consumption is an important aspect to consider, especially if you're looking for a laptop upgrade as even the best ultrabooks can have mediocre storage.

Some SSDs can consume watts of power at idle while better-suited ones sip just milliwatts. Average workload power consumption and max consumption are two other aspects of power consumption, but performance-per-watt is more important. A drive might consume more power during any given workload, but accomplishing a task faster allows the drive to drop into an idle state more quickly, ultimately saving energy.

For temperature recording we currently poll the drive’s primary composite sensor during testing with a 24C ambient.

The SN580 is moderately efficient, better than the P41 Plus, but not as good as the SN770. This is due to the SN770 having higher peak performance, which enables it to finish the workload more quickly. The SN580 is still massively more efficient than the SN570 and the older drives, in part due to its larger cache. Still, this result isn’t bad, although it’s not on the level of many competing drives. WD markets the SN580 as being low-power for laptops, and while we think it would be good there, especially compared to the SN570, it’s essentially only satisfactory in that respect.

During write saturation testing, the drive’s hottest sensor reached the 74-77C range, which is below throttling. WD rates this drive for an operational temperature of up to 85C, which is a pretty typical limit. The extra large cache helps in building the temperature up, as also seen with the SN770, but it won't get nearly as hot in real-world workloads.

Test Bench and Testing Notes

We use an Alder Lake platform with most background applications such as indexing, Windows updates, and anti-virus disabled in the OS to reduce run-to-run variability. Each SSD is prefilled to 50% capacity and tested as a secondary device. Unless noted, we use active cooling for all SSDs.

Conclusion

The Blue SN580 is a relatively low-effort product from Western Digital that takes no risks, but, on the other hand, provides a sufficiently satisfying experience. It takes from the DNA of the SN570 and SN770 to produce a drive somewhere in-between that is, at its core, an excellent budget SSD. It’s a stretch to call it a PCIe 4.0 drive, but it does bring improved performance and better caching over the SN570 without stepping on the toes of the SN770. In most cases, it will give an equivalent experience to the SN770 at what should be a lower price, although this suggests the SN770 will eventually be replaced.

The SN580 has a pretty decent warranty, and the software support isn’t bad, either. These aren’t major benefits, but when coupled with WD’s mature design it makes for a very reliable drive. It’s a good choice regardless of your intended use. WD does have a history of switching up the hardware on its drives, like with the SN550 and the SATA Blue, but as tested and launched, the SN580 is a solid all-around budget drive. It’s not the most exciting product we’ve seen, but that doesn’t mean it’s not good or that it doesn’t have its place.

We’ve tested a lot of great drives in this segment — that is, DRAM-less, PCIe 4.0 drives in the 4-5GBps range — that deliver a great experience that can even rival high-end drives. The exception may be at 2TB, where many of them swap to QLC, although that doesn’t apply to the SN770, SN570, or now the SN580. That puts the SN580 in a good spot if you’re looking for a 2TB TLC drive on a budget. The small uplift in performance and improved cache design are bonuses over the SN570, and this translates to a better and more efficient experience.

The SN580’s real competition is the SN770 and, in due time, the higher-end DRAM-less 4.0 options that are arriving, modeled by the Predator GM7. The latter may be able to offer a better experience at a comparable cost, although the SN580 has WD’s name. The SN770, for its part, is still a great drive, but it may have to be updated in time. This leaves the SN580 as a safe all-around choice that doesn’t really do anything new or exciting. It just fills a spot in the line-up where it can devastate the less-desirable budget PCIe 4.0 drives like the Kingston NV2. The rest of the picture comes down to pricing.

MORE: Best SSDs

MORE: Best External SSDs and Hard Drives

MORE: How We Test HDDs And SSDs

MORE: All SSD Content

Update 7/14 03:39 PT:

Arduino dropped us a tweet to say that Arduino Uno R4 WiFi is now supported in the Arduino Cloud. We have re-tested and confirmed that we can make a "thing" (Arduino Cloud parlance for an IoT device) and have updated the relevant section of this review. The final score remains the same.

Updated Review: 

Despite its age, the Arduino Uno, the most famous board in the range, is still seen as the go to board for makers. In its first decade it has powered countless projects, both great and small. The Uno has seen many variations, but at its heart has been the Atmel chip, until now. 

The Arduino Uno R4 series features a new Arm-powered heart and the same CPU across two boards. The most basic board is the $20 Arduino Uno R4 Minima, essentially the next generation Arduino Uno, but at the other end of the scale is the $27 Arduino Uno R4 WiFi, which features an onboard ESP32 co-processor. An ESP32 co-processor just for Wi-Fi may seem like overkill, it can run at up to 240 MHz versus the Renesas RA4M1 Arm Cortex M4’s 48 MHz, but the ESP32 has proven to be a rock solid board to build with. We just have to look across our bench to see the many boards based upon it.

Are these new Arduino boards worth the dollars in your pocket, or should you just buy a Raspberry Pi Pico W? To learn that and more we need to put both of the boards on the bench and run a few tests.

Arduino Uno R4 Specifications

Getting Started With the Arduino R4 Boards

Opening the box of a new Arduino is something that we will never forget. From our first R2 board until today. We don’t do unboxing videos and they don’t factor in our review, but the experience is sublime. Minimal packaging that also displays the basic specifications of the boards. The boards come pre-soldered (unlike the Raspberry Pi Pico) and are dark blue in color, in fact they are of a darker hue than our Arduino Uno R2 board. Both boards share the distinctive Uno layout, and that means we can connect Shields (Arduino parlance for add-on boards, Raspberry Pi uses HATs) to our Uno R4, but more on that later. 

Arduinos come pre-flashed with a simple test script. For the Minima we have the classic “Blink” sketch (Arduino parlance for project code) which sees an LED flash on / off every second. For the R4 WiFi we see a 12 x 8 LED matrix dominate the board. The matrix shows a brief Tetris animation before running through a series of geometric animations.

Making GPIO connections is super easy. GPIO pin references are clearly printed on the female headers and the PCB. This is exceptionally useful for those new to electronics. The Raspberry Pi Pico has GPIO references printed on the underside of the board and that can get confusing. Connecting the R4 Unos to our computer relies on a USB C data cable and the Arduino IDE.

Programming the Arduino Uno R4 Boards

The Arduino IDE has been with us since the early days of the boards; heck we’ve been using it for 10 years now. In the last year Arduino has updated its offline Arduino IDE. Now at version 2.1.1, the Arduino IDE is a fantastic improvement over past iterations. Not that we have any issues with the old IDE, it served us well and it still works with these new boards.

Installing the Arduino IDE and connecting the Arduino Uno R4 boards to the computer, and we see the IDE detect and configure the boards for use. It even downloaded the latest drivers and configuration for the new boards. After that we ventured into File >> Examples and selected a series of example sketches for the boards. 

Starting with the Minima, we updated the standard Blink sketch to use a for loop that would blink the LED a few times before stopping, then repeating. Flashing the code took under 30 seconds and all was good. For the R4 WiFi, we chose to use an example that uses the ESP32’s Wi-Fi connectivity along with the R4’s Arm CPU. The onboard ESP32 is an incredibly powerful microcontroller in its own right, but acting as a co-processor it offloads all of the networking that would slow down the 48 MHz Arm CPU. 

The goal of the web example is to create a web server and host a simple HTML page that we can use to control the status of the onboard LED. The example can either be its own Wi-Fi access point, or we can assign it to our existing network. We chose the latter and after adding our Wi-Fi details we then flashed the board and went to Arduino Uno R4 WiFi’s IP address. Clicking the On / Off text then triggered the LED to change state.

With the tests out of the way, we can be a little more adventurous. How about a web controlled motor? Using the example code, with a slight tweak for the GPIO pins, we connected a DC motor to the Arduino via a Cytron Maker Drive motor controller. The Arduino GPIO can only supply 8mA per GPIO pin, nowhere near what a motor requires. With the connections made we flashed the code, then went back to the browser. A press of the text and the motor sprang to life, then we pressed the other text element to turn it off and then we realized that we had forgotten to change the GPIO pin. A quick tweak and flashing the code to the Arduino saw the problem resolved. 

Using Shields with an Arduino, we benefit from ready-made platforms on which we can base our projects. We have Shields for Ethernet, LCD screens, Audio Synths and we even made our own Shield to flash ATTiny85 microcontrollers. These same shields are electrically compatible with the Arduino R4 boards, but the caveat is that the software may require some tweaking. It will eventually catch up, but you may have to get your hands dirty to make things work, something we’ll cover later. 

We connected up a Maplin (a long since closed UK electronics store) branded touch shield and flashed some code to enable the embedded MPR121 capacitive touch sensor. Opening the Serial Monitor and we saw our keypresses being registered. We then connected our Audio Synth Shield, the same one that we hand soldered for our Make Your Own Uno Kit review. Flashing the test sketch and we soon had a “musical instrument” with which to inflict our brand of music on our poor neighbors.

The Arduino Uno R4 WiFi’s 12 x 8 LED matrix is a simple affair. Instead of individual addressable RGB LEDs we have a matrix made from a series of red LEDs. Controlling the LEDs can be done in code, by creating a series of arrays in a file called Animation.h. We can create these using a web application, saving the arrays to a text file which we then paste into the code. There is also a web app which uses web serial in the browser to connect to an Arduino Uno R4 WiFi running a specific sketch. Using this web app we can “paint” the LEDs on or off. Great fun for demonstrating the LED matrix, and surely the spark of an idea for future maker projects.

Arduino Uno R4 WiFi Qwiic I2C Issues

This issue only applies to the Arduino Uno R4 WiFi, and the sole reason for that is the inclusion of a Qwiic connector. Yes the same Qwiic connector offered by SparkFun, Adafruit (StemmaQT), Pimoroni (QW/ST) and Cytron (Maker Port) is present on the R4 WiFi and while this is a most welcome addition, it also brings a problem. 

Normally, Arduinos have just a single I2C connection, but the addition of the Qwiic connection means that the R4 WiFi has two. More is a good thing, right? Well yes it is, but software has to catch up to this innovation. The Arduino Uno R4 WiFi cheatsheet, which we only later found, identified the secondary I2C bus and helpfully told us what to do in order to make it work with our range of StemmaQT add on boards. But it wasn’t quite that simple. 

Arduino relies on the Wire.h library to configure the I2C interface, and we can’t just import it into our code. Instead we have to locate the library for our chosen board, in our case Adafruit’s MPR121 touch sensor code. Handily, hovering the mouse over the correct line shows the file path to the library. Copy the path, open it in Notepad++ and change Wire to Wire1. Save. Go back to the Arduino IDE, flash the code and all is good. Yes this is a little time consuming and not for newcomers, but as we said earlier, the software support will catch up once the R4 WiFi gets into more hands. 

With this workflow in hand, we managed to test a VCNL4040 proximity sensor, IS31FL3741 RGB matrix and an LTR390 UV sensor. We also tested a SparkFun Qwiic Shield, providing four Qwiic connectors, all routed via the SDA / SCL pins of the Arduino. Which meant that we had to undo our changes to Wire.h in order to make it work.

Arduino IoT Cloud and the Arduino Uno R4 Boards

The offline Arduino IDE isn't the only means to write code for your new Arduino Uno R4. The Arduino IoT Cloud is an online service to create your own IoT applications (Things) using an online Arduino IoT. The free tier is a little restrictive, but enough for users to cut their teeth.

Originally we encountered a problem in that Arduino IoT did not detect our R4 boards even though we followed the instructions and installed a bunch of software. That isn’t the end of the world, as the Web Editor, an online IDE, works exceptionally well. We flashed code to both of the boards and everything went well.

Arduino Uno R4 WiFi Support Added to the Arduino Cloud

A few days after this review, Arduino announced support for the Arduino Uno R4 WIFi on its Arduino Cloud service. We went back and re-tested by creating an IoT "Thing" with a web controlled LED. Pressing a button on a dashboard would trigger an LED to turn on and off.

Setting up the Arduino Uno R4 WiFi for use with the Arduino Cloud is pretty easy. The only section which may phaze users is when we need to update the firmware, requiring us to download and run a .bat file. This triggered Windows Security and initially prevented use. Clicking "More Info" and then "Run anyway" did the trick. After that we ran through the online wizard, created a "thing", linked the Arduino Uno R4 WiFi as a device, created a dashboard, and the Arduino Cloud auto-generated some boiler-plate code. This code served as the basis for our project, a quick function handler to read the state of the dashboard button, and some LED control code and we had a web controlled LED.

With the Arduino Cloud, the Arduino Uno R4 WiFi is a great platform for your first IoT experiments. The free tier is enough for most users and educators, more advanced users will jump onto one of the subscriptions for increased services. With the Arduino Cloud we can create web apps to control devices and components connected to the Arduino Uno R4 WiFi, and we can receive data from the device and view it in a web interface.

Who are the Arduino Uno R4 Boards Aimed at, and Why? 

The Arduino Uno R4 Minima is clearly aimed at new users. The $20 price tag, ease of use and form factor will benefit those looking to learn electronics. It is the Arduino that a new generation will claim as their first machine. 

The Arduino Uno R4 WiFi is only $7 more and with that we get Wi-Fi and Bluetooth (not at the same time) and the Qwiic connector. This little board, once it receives some software love, will be what educators, hobbyists and professionals will use as the heart of their projects. Just look at how popular the Arduino Uno was in its first decade. Sure it was never a powerhouse, it couldn’t run an OS, or connect to networks. But the community adopted it and made it much more.

Should we Just buy a Raspberry Pi Pico W?

There is no clear-cut answer here. The Raspberry Pi Pico W is much smaller and cheaper than the Arduino Uno R4 boards. The Raspberry Pi Pico is under $5, and the Pico W under $8. They can both be coded using the Arduino IDE and have a massive support base. That said, if you have built your career using Arduino and have all of the kit and knowledge, then the R4 boards are an obvious step along the path. 

Arduino boards have a proven track record in the education sector, largely down to their ease of use and excellent documentation. But Raspberry Pi has the same track record, and its Pico documentation is exceptional.

Bottom Line

It has some flaws, for now, but the Arduino Uno R4 series of boards will soon become the de facto standard on which learners, makers, educators, and professionals base their projects. If you are new to Arduino, the Minima’s $20 price tag is quite enticing, but pay the extra and get the $27 Arduino Uno R4 WiFi as the inclusion of Wi-Fi is an incredible learning opportunity. That said, the Minima will be in our bits box, ready for service when we need to write ATtiny85 firmware or make some musical device / light show for the holiday season.

The adage about some things being too good to be true never gets old, as demonstrated by TechTuber Roman 'der8auer' Hartung in his latest video. The extreme overclocking expert was alerted by a friendly subscriber about some very attractively priced Samsung branded 4TB SSDs on the AliExpress marketplace. To cut a long story short, as you can see from the video title below, these drives were fake.

Der8auer's skepticism with regard to these AliExpress advertized SSDs being genuine was not just aroused by the incredible pricing — this particular 'Samsung' 4TB SSD was €40 (about $44). Another warning about whether this was a genuine product came from the model name/number. The 'Samsung 980 EVO' doesn't even exist in Samsung's product catalogs, never mind a 4TB version.

Pondering over the packaging, many other blunders or inconsistencies from the knock-off makers were apparent. On the front of the 'Samsung 980 EVO' we see a single-keyed M.2 device, which should be an NVMe device, however, the advertised transfer speeds would have tallied with an M.2 SATA device. Indeed, unpackaging the drive showed it was M.2 SATA. On the rear of the packaging was a mishmash of copied slogans, terms, and warranty promises.

After unboxing, der8auer continued his visual inspection, raising various peculiar flags for a 'Samsung' drive. Peeling the sticker showed that whoever put together this M.2 device erased any identifying markings off the SSD controller chip. Despite being able to read some code numbers on the flash NAND chips, of which there were just two (on this single-sided SSD), der8auer couldn't find any technical information about them.

Moving onto testing, der8auer decided to put his new 'Samsung 980 EVO 4TB' drive through some system info and benchmarking tools. Upon connecting the M.2 device, Windows showed it as a 3.72 TB free volume. CrystalDisk Info agreed with this initial assessment, and the info fields were fully populated.

Der8auer's attempt to verify the capacity of the new SSD hit a practical roadblock when it became apparent that completing this process would take at least 24 hours. Yes, it started to go very slowly after an initial burst of speed. Switching to the same tool's benchmarking tab showed why any capacity verification task would be "painful and slow," with sequential read and write speeds of 36.25 MB/s and 0.84 MB/s, respectively. It is hard to find such slow flash storage nowadays, so this is kind of a negative achievement.

Other Storage Products Which You Should Definitely Avoid

As a footnote, der8auer noticed some other irresistible SSDs while browsing for the headlining 'Samsung 980 EVO 4TB'. He couldn't stop himself from adding the following to his basket:

• An external 'Seagate' SSD with an advertized 64TB capacity,
• A similar 'Seagate' external SSD with an incredible 128TB claimed capacity,
• And another M.2 drive, this time claimed to be a 'Samsung 980 PRO 4TB' (a Samsung SKU which isn't shipping yet).

The above bullet-pointed external drives were found to contain a small PCB with 64GB microSD cards glued into a socket. However, the purported Samsung 980 PRO 4TB (€50) offered an almost acceptable performance of 2,473 MB/s reads and 1,057 MB/s writes until the lack of cache became apparent. Again, the capacity (4TB in this case) was faked. Moreover, the so-called 980 PRO 4TB was probably "best case 1TB" in capacity, said der8auer.

When Sabrent earlier this year announced plans to build a PCIe Gen5 SSD with a read speed of 14 GB/s, we were impressed. Today Sabrent sent out a screenshot with proof that its upcoming Rocket X5 can achieve such a level of performance and even said that it is working to achieve even higher speeds to topple the best SSDs currently available.

Sabrent's Rocket X5 can hit sequential read speeds of 14,179 MB/s and sequential write speeds of 12,280 MB/s in CrystalDiskMark. As for random performance, Sabrent's prototype drive can hit 1.56 million 4K read IOPS and 1.669 4K write IOPS, comparable to high-end enterprise-grade SSDs.

Speaking of enterprise-grade SSDs, we are talking about Samsung's PM1743 PCIe 5.0 x4 SSD, which achieved approximately 13.8 GB/s throughput with Intel's Alder Lake processor in the IOMeter benchmark in late 2021. Apparently, Sabrent can reach even higher speeds with its Phison PS5026-E26-based drive, albeit by using very fast 3D TLC NAND with at least 2400 MT/s interface. 

"The new Rocket X5 is still in the tunning process, and we expect performance to improve even better," a statement by Sabrent reads. "This still needs to be a finished product, but we are moving forward with the tech we have at hand now. To name a few items, many things can change, including the name and label and the heat sink. As we work with Phison very closely, new tech is on the way, and Phison assists us in pushing what we have now to achieve groundbreaking results."

Sabrent said it had conducted experiments with various thermal solutions aimed at keeping these extreme drives cool. The company indicated that the current thermal solution is specifically tailored to cool these next-gen SSDs efficiently, and its fan does not produce any high-frequency noise.

"Stay tuned to Sabrent for news on our other top-performing products as we approach the release," the company said. "Our test lab is very busy these days with many new groundbreaking products we will release soon."

TrendForce, the high-tech market intelligence firm, has published an update on the NAND flash market. Its latest forecasts see NAND flash's average selling price (ASP) falling between 3 and 8% in Q3 2023. However, for PC enthusiasts and DIYers, the news is even better. Client SSD pricing is expected to fall between 8% and 13% during the same period. In other words, those in the market for PC SSD storage could perhaps save an extra 10% or more on today's prices if they wait a few weeks.

NAND Flash chips are a major component in PC SSDs, products like smart devices (UFS and eMMC) and memory cards. The pricing of this essential storage device component will continue to fall significantly in the current quarter, according to the analysts at TrendForce.

As usual with its market intelligence reports, TrendForce provides some background and reasoning for its predictions. Currently, NAND makers are "making concerted efforts to scale back production," notes the analyst firm. Manufacturers thus hope to create artificial scarcity and pump up NAND pricing. However, TrendForce reckons that these efforts will be in vain and aren't enough to address the state of oversupply in Q3 2023 – even with the traditional seasonal demand surge expected in the coming months. TrendForce predicts pricing of NAND and NAND-based products to slide.

Looking closer at the segmented figures TrendForce provides, we see that Client SSD pricing looks likely to improve – for customers, not NAND producers. Last quarter was a great period for PC enthusiasts, DIYers and upgraders, with SSD ASPs sinking between 15% and 20%, according to the tabulated data. The continued drop of between 8 to 13% is also well worth waiting for if you are looking to do a new build or for an upgrade and you aren’t in a mad rush to complete it.

If you follow our news and deals, you will know there have been plenty of SSD bargains in recent weeks. We can be pretty confident that the deals are set to continue, probably improving, based on TrendForce data.

With NAND ASPs dropping, it isn’t only PC SSDs (AKA Client SSDs) that will be more keenly priced. TrendForce outlines how products like Enterprise SSDs, mobile storage (eMMC and UFS), and other devices have been impacted and are predicted to react. So, for Tom’s Hardware readers, the expected drops in smartphone and tablet components might be welcome to reduce upward price pressures on such products.

Last but not least, TrendForce doesn’t predict NAND ASPs will continue to drop steeply after Q3. It expects to see a rebound in high-end 3D NAND pricing first but doesn’t give any firm prediction on whether NAND ASPs will continue to fall or rise in Q4 2023.

DigiTimes reported that ASML is mulling the release of a special version of its deep ultraviolet lithography (DUV) tool that will comply with the latest U.S. export rules and that could be shipped to Chinese customers without a license. The device would enable companies like SMIC and Hua Hong to make chips on 28nm-class nodes and thicker, but it will not allow them to use more advanced nodes. However, the company has since denied this report.

Updated (7/6): ASML wrote us a note, saying that the DigiTimes report is incorrect. 

"ASML complies with all applicable laws and regulations, and that it is not designing a system specifically for China and will not respond further to rumors and speculations," ASML said in a statement.

The lithography tool in question is the Twinscan NXT:1980Di, and currently this is the least advanced immersion scanner that ASML still makes. The machine has 1.35 numerical aperture optics and is capable of <38 nm resolution, which is enough for 7nm-class and even more advanced nodes. In fact, this scanner — originally released in 2016 — it was used by TSMC to develop its 7nm-class process technology.

Cutting down this device and increasing its minimum supported resolution to prevent SMIC and other Chinese chipmakers from creating sub-28nm process technology is theoretically possible. Keeping in mind that the lion's share of SMIC's revenue comes from production nodes that are thicker than 28nm, it is likely that Chinese companies would still be interested in procuring such tools.

The latest export regulations mandate American firms and individuals to obtain licenses for exporting tools and technologies capable of manufacturing logic chips with non-planar transistors on 14nm/16nm nodes and below, 3D NAND with 128 layers or more, and DRAM memory chips of 18nm half-pitch or less. The same rules apply to non-American companies that export components from the U.S., which is the case with ASML and the Twinscan NXT:1980Di.

The latest Dutch export rules require ASML to obtain an export license to sell its Twinscan NXT:2000i scanners to Chinese companies. ASML has not officially created a limited version of the Twinscan NXT, but if the export regulations continue to restrict sales of advanced technologies to China, such a move seems likely.

The Raspberry Pi Pico's RP2040 has been embedded in a plethora of form factors, many of which feature in our list of best RP2040 boards.
SB Component's $4.50 Micro RP2040 may look familiar. It bears a striking resemblance to Solder Party's RP2040 Stamp which sees the GPIO pins broken along the perimeter of the board, but with Micro RP2040 we have an onboard USB C interface, but it came at a cost.

The sacrifice is a handful of GPIO pins. We don't get the full 40 pins from the RP2040, but that said, who uses all of them anyway? Software debug (SWD) pins are present, so you can make your own debug probe, or connect it up to an external debug probe to step through your code. The design of the board follows the Raspberry Pi Pico's castellations, which means it can be surface mount soldered, perhaps using one of the best soldering stations, into a project. The product page also claims that it can be used in a breadboard, but the bottom row of GPIO pins will have to soldered pointing upward, otherwise they will short each other out in the breadboard.

The form factor packs a lot in a small package, making it useful for creating your own USB HID devices. A version with Wi-Fi and Bluetooth would be a killer product, but you can only pack so much into the form factor.

The key difference between the Micro RP2040 and RP2040 Stamp is the USB interface. Micro features a USB C interface built in to the board, whereas Stamp required an off board USB to serial converter or an Arduino Uno shaped carrier board.

Micro RP2040 is available via SB Components for approximately $4.50. Technical details and pinout diagrams can be found in a GitHub repository.

Chen Nanxiang, the head of Chinese 3D NAND champion Yangtze Memory Technologies (YMTC) that got into the U.S. trade blacklist late last year, asked producers of wafer fab equipment to buy back tools the company cannot use. In his keynote speech at Semicon China conference he also warned that the global semiconductor industry is entering an era of turmoil and disarray due to increasing geopolitical tensions.  

Buy It Back

According to Bloomberg, Chen Nanxiang, chairman and CEO of YMTC, gave a speech in Shanghai, China, that directly addressed machine suppliers, and how they should buy back the legally purchased equipment and components if they cannot be used.

While many companies are slowing down spending on wafer fab equipment these days, makers of chip production equipment have huge backlogs, so selling off fab tools may not be a problem for YMTC, if it really wants to get rid of them.

YMTC got hit by the U.S. government twice last year. In October, YMTC was blocked from acquiring wafer fab equipment (WFE) with American technology necessary for producing 3D NAND with 128 or more layers as part of the broad sanctions imposed by the U.S. government on the Chinese semiconductor industry. The restrictions also oblige U.S. citizens to obtain a license from the U.S. Department of Commerce to support development or production of said memory devices in China. As a result, four prominent chipmaking tool companies — ASML, Applied Materials, KLA, and Lam Research — ceased their work with YMTC due to the requirement of obtaining relevant export licenses from the DoC.

In December, YMTC got into the U.S. DoC Entity List and can no longer procure equipment, software, and other technologies from U.S.-based companies (or other companies selling technologies developed in America) — unless the latter secure a special export license from the DoC. Such export licenses are reviewed with a presumption of denial.

As a result of the two hits, YMTC cannot obtain, install, and use the latest tools it acquired from WFE producers that it needed to make 232-layer 3D NAND, a type of memory that could power some of the best SSDs with a PCIe 5.0 interface. Furthermore, the company has troubles servicing its existing tools. Some analysts believe that Yangtze Memory might have to abandon production of 3D NAND and turn itself into maker of specialty 2D NAND or even a contract chipmaker.

Turmoil and Disorder

Chen stated that globalization is effectively dead, and the previously established rules, balance, and harmony have been disrupted. This shift is predicted to greatly impact the supply chain, industry practices, and business models. The head of YTMC believes that key aspects of globalization such as competition, innovation, and the free movement of talent and resources have been affected. 

According to reports via Nikkei, Chen believes that globalization is dead and that government interventions and that existing rules and past balances are now broken. Chen continues, stating that the semiconductor industry could see a period of turmoil. A period which could see a tremendous and profound impact on the supply chain and industry practices.

The chief executive and chairman of the troubled 3D NAND maker highlighted the importance of global cooperation in the semiconductor industry. He emphasized that the complex chip supply chain involves 25 countries and has indirect ties with another 23. 

Chen also indicated that semiconductor markets in China and the U.S. — which benefited from contributions from foreign investors and inventors — do not solely belong to their respective countries, but are global assets. Similarly, he brought example of NAND flash memory, which was invented by Toshiba in Japan back in the 1980s, and which is perfected by South Korea-based Samsung and SK Hynix.