Intel's next-gen Nova Lake family is expected to land on a new socket, but it will also bring forth a bunch of new chipsets with it. We've already covered them before, but new details have surfaced about the flagship offerings: Z970 and Z990. Both of these will allegedly feature the same PCH that's 22% smaller but more power-hungry than the current-gen Z890 platform, consuming up to 14W at the top-end when fully saturated.

Before we discuss that, a low-resolution picture of the Z990 PCH has also been leaked, with Jaykhin claiming it measures out to 25 x 24mm for the package, while featuring an 11.15 x 6.5mm die. That comes out to 72.5mm² for the die area, while we're looking at 600mm² for the package area. For comparison, Z890's package area was 658mm², and the die area would work out to 92.9mm², constituting a 22% shrinkage for the die and an 8.8% smaller overall package.

This is despite the higher power draw suggested for Z990, which is 14W. However, the platform will only hit that upper limit when running multiple PCIe 5.0 devices simultaneously; otherwise, the base power for the Z990 PCH is just 7.9W — still 1.9W more than the 6W base power of Z890. Even the cut-down Z970 apparently features a 6.4W base power draw. Both chipsets have the same maximum operating temperature of 113°C, 5 degrees higher than Z890.

When you have just a single GPU slotted in the motherboard, it's wired up directly to the CPU and doesn't use any of the downstream PCIe lanes stemming from the chipset. The same goes for a single PCIe 5.0 drive for Z970 and up to two PCIe 5.0 SSDs for Z990. But the moment you add more PCIe 5.0 devices, they're routed through the chipset, which makes it consume more power to maintain signal integrity.

We haven't officially seen any LGA 1954 motherboard yet, but there were some early prototypes at Computex, which is where the leaked PCH image could've come from. Anyhow, Nova Lake is expected to carry up to 52 cores at the top-end, so it makes sense such powerful motherboards are required to handle the silicon. Previous rumors have pointed toward an insane 700W PL4 for the flagship NVL-S part with dual compute tiles.

It's a weird fact of the modern PC industry, but the motherboard "chipset" or "Platform Controller Hub" (to use Intel's terminology) is basically just a PCIe-attached I/O hub that exposes additional PCIe, USB, SATA, and platform services. Contemporary CPUs are built as SoCs, so all the essential functions to run the machine are integrated into the CPU itself. That's why companies have started creating what are effectively I/O breakout boards by slapping motherboard chipsets onto PCI Express cards, including the one pictured above from WisdPi, as well as one coming from Minisforum, as spotted by PC Watch at Computex.

In the case of the AMD Promontory 21 chipset, it really is just a PCI Express I/O controller. In fact, AMD created the X670 chipset by simply wiring up two Promontory 21 chips in series. It's the exact same chip, just used in pairs on the fancier motherboards, so you end up with piles of I/O, though it's all running over the same PCI Express 4.0 x4 link to the CPU.

We've reported on cards like this before; this past December, we highlighted an open-source project for exactly this purpose, and of course, Asusd and ASRock have shipped expansion boards for specific motherboards that "upgrade" them from B650 to X670 by adding the second Promontory 21 chipset. All the way back in 2023, we also covered AMD engineers showing off the same thing in a Gamers Nexus video. Indeed, the very word "Promontory" refers to a projection from a larger body, such as a high cliff or, perhaps, an add-on card.

So, the WisdPi "PROM21 All In Expansion Card" is a half-height PCI Express 4.0 add-in card that offers a whopping four M.2 slots, five USB 10 Gbps ports, six USB 2.0 ports, and an OCuLink port that supports PCI Express 4.0 or can breakout into four SATA ports. That's more connectivity than some motherboards offer, which is no surprise, since the card costs $199 by itself, with no cables. WisdPi is primarily a Raspberry Pi shop, and as you'd expect, the card is compatible with Raspberry Pi devices that expose PCI Express, but that's simply because it's actually compatible with any PCI Express device, from AMD, Intel, Arm, or whoever.

The card spotted by PC Watch at Minisforum's Computex booth is similar in that it also offers four M.2 slots, an OCuLink connector, and at least one high-speed 20 Gbps USB port. However, it lacks the SATA functionality and the other USB ports of the WisdPi offering. Also, unlike WisdPi's card, it seems that it will come with a dedicated cooling unit (with a blower fan and a shroud) for fast SSDs. That's a little ironic considering the card seemingly limits drives to PCIe 4.0 x2 if you install all four, but "cooling" is better than "no cooling". No word on how much the Minisforum offering will cost, but it will apparently be available in Q3 of this year.

It's almost a shame these cards aren't more commonplace. For folks with machines based on entry-level motherboards, one of these cards could be just the ticket to expanding the I/O. Particularly on older-generation motherboards that usually had only one M.2 slot (if any), a card like one of these could enable significant storage expansion now that M.2 SSDs are the only type still being actively produced. You can check out PC Watch's photos of the Minisforum "PCIE TO 4" card at their post over here.

We already know that Intel pins a lot of hopes on its Nova Lake processors and hopes that they will put it back on the map for high-end desktop CPU enthusiasts, but until today we did not know almost anything about the company's next-generation 900-series chipsets that will support Nova Lake CPUs. On Monday Jaykihn, a leaker who tends to know a lot about Intel's plans, published a table describing specifications of Intel's B960, Z970, Z990, Q970, and W980 platforms.

Two things that strike the eye with the new family of chipsets is the lack of the H910 platform for entry-level PCs as well as the presence of the Z970 platform for inexpensive desktops with overclocking capability, which is an all-new category of platforms. For demanding users that plan to overclock their CPUs, Intel will offer its Z990 chipset (which supports overclocking using both multiplier and BCLK), whereas those who do not plan to overclock can go with the W980, which is officially positioned as an entry-level workstation solution and therefore supports vPro technology and manageability features. In addition, Intel will have the Q970 chipset for performance-mainstream desktops. Interestingly, the Q970 will be the only chipset that will not support memory overclocking among the 900-series chipsets.

The range-topping Z990 and W980 chipsets will offer 48 PCIe lanes (including 12 PCIe 5.0 lanes from the chipset and 16 PCIe 5.0 lanes from the CPU), two Thunderbolt 4/USB4 ports supported by the processor, 5 USB 3.2 20 Gbps, and 10 USB 3.2 10 Gbps ports. By contrast, inexpensive B960 and Z970 will only support 34 PCIe lanes (including 16 PCIe 5.0 lanes from the CPU and 14 PCIe 4.0 lanes from the chipset), one TB4/USB4 port supported by the processor, two USB 3.2 20 Gbps ports, and four USB 3.2 10 Gbps ports, according to the leak.

Intel's 900-series chipsets are expected to support Intel's Nova Lake processors that are said to use the LGA1954 socket and pack up to 52 cores, including up to 16 high-performance Coyote Cove cores, up to 32 energy-efficient Arctic Wolf cores, and four ultra-low-power cores. In addition, the new CPUs are projected to feature Xe3 integrated GPU and media engine from the Xe4 GPUs.

Although Intel has reaffirmed Nova Lake will arrive before the end of the year, the company hasn't shared any details about the processors yet. Before we see them, we expect to see a minor range of Arrow Lake Refresh chips.

Even the best motherboards often present connectivity and expansion challenges, particularly in compact form factors. A new PCIe AIC (Add-In Card) is an effective solution. There's an open-source project titled “AMD B650 Southbridge Expansion Card,” hosted on OSHWHub, that provides an interesting option to enable substantial expansion of your motherboard’s storage capabilities.

It is not the first time a manufacturer has developed an AIC based on an AMD chipset. Previously, ASRock introduced the “X670 Xpansion Kit,” a specialized expansion card designed for its B650 LiveMixer motherboard. This add-in card, which also uses the B650 chipset, elevates the B650 LiveMixer to the feature set of an X670-class motherboard.

For context, the X670 motherboard is equipped with two AMD Promontory 21 chipsets, while the more cost-effective B650 motherboards include only one. By incorporating an additional chipset via the AIC, the B650 LiveMixer can provide features and capabilities comparable to those of higher-end X670 models. Hardware enthusiast Uniko's Hardware quickly reminded us that Asus used a similar approach, placing an X670 chipset on an M.2 daughterboard that plugs into the ROG Strix X670E-I Gaming WiFi, a mini-ITX motherboard. In contrast, the AMD B650 Southbridge Expansion Card is a community-driven project hosted on OSHWHub, a platform dedicated to sharing and collaborating on hardware designs.

The AMD B650 Southbridge Expansion Card interfaces with the host system via a standard PCIe 4.0 x4 connector. Notably, this expansion card is compatible with both Intel and AMD platforms, provided the system includes an available PCIe 4.0 x4 expansion slot. The expansion card features four SATA III ports, two PCIe 4.0 x4 slots, and one USB 3.2 Gen 2x2 header.

Benchmark results confirm that all storage ports deliver excellent performance. PCIe 4.0 SSDs achieved sequential read speeds just above 6,900 MB/s and write speeds exceeding 6,400 MB/s. SATA drives demonstrated consistent sequential read and write speeds above 500 MB/s. The USB 3.2 Gen 2x2 port recorded sequential read speeds above 2,000 MB/s and write speeds over 1,900 MB/s.

Schematics, a parts list, and comprehensive instructions for building the AMD B650 Southbridge Expansion Card are available at no cost. The author notes that you must flash the card with a special firmware for proper functionality. While the firmware is not publicly hosted, you can obtain it for free by joining the creator's QQ group.

According to the author, you can replicate the AMD B650 Southbridge Expansion Card for approximately 300 yuan (about $42.88). Although the expansion card is not commercially available, OSHWHub has integrated manufacturing services through its sister company, JLCPCB. It allows you to order custom hardware directly using the design files shared on OSHWHub, and JLCPCB will manufacture it.

According to a report from the Korean outlet Hankyung, Samsung Foundry, the manufacturing division of Samsung Electronics, has reportedly secured a notable client: none other than Intel. The foundry has ostensibly obtained 8nm orders to manufacture Intel's forthcoming Platform Controller Hub (PCH).

The report states that Samsung and Intel are in the final stages of mass production of Intel's chipsets. Consequently, it is plausible that the forthcoming 900-series chipsets for the LGA1954 socket, which will accommodate the Core Ultra 400S (codename Nova Lake), will leverage Samsung's 8nm process technology. There exists a longstanding history of collaboration between Samsung and Intel, as the former has previously produced chipsets and other low-value chips for the latter.

Samsung presently manufactures certain Intel chipsets utilizing the 14nm process node at the foundry situated in Austin, Texas. Meanwhile, Samsung's 8nm process node is at its manufacturing plant in Hwaseong, Gyeonggi Province. Consequently, it appears that Intel's forthcoming chipset production may be relocating back to Korea. This strategic move to choose Samsung seems logical, considering Intel's intent to diversify away from TSMC, which is currently experiencing a persistent shortage.

You don’t really need cutting-edge manufacturing processes for chipset production, so 8nm is perfectly fine. Nonetheless, it’ll be interesting to see what kind of benefits Intel can reap from Samsung’s 8nm process node, whether they come in the form of improved features, better power consumption, or thermals. One thing’s for sure, though. The transition from 14nm to 8nm could give Intel some bragging rights over AMD since the latter’s current 800-series chipsets are still on the 14nm process node. Then again, who is to say that AMD’s next chipset will not surprise us with a node shrink?

Since its introduction in 2017 and commencement of mass production in 2018, Samsung's 8nm manufacturing process has achieved a satisfactory yield level, thereby attracting significant clients. The foundry previously secured a contract with Nvidia to produce the custom System-on-Chip (SoC) for Nintendo's Switch 2 console, which is experiencing strong sales. Landing a deal with Intel represents a notable achievement for Samsung as well. Despite Intel's declining processor market share to AMD, the Blue Team remains the dominant industry player, holding over 75% of the market share, approximately.

Samsung's production capacity is approximately 350,000 wafers per month. Specifically, the 8nm process node yields between 30,000 and 40,000 300mm (12-inch) wafers monthly. This figure accounts for roughly 11% of Samsung's total capacity. Nevertheless, as the number of clients the foundry acquires increases, so does the demand for photomasks, thereby creating a mutually beneficial situation for all stakeholders within the supply chain.

Assuming the report is accurate, Samsung will commence full-scale manufacturing of Intel's 8nm chipsets in the upcoming year. Intel has already confirmed that Nova Lake will be launched either before the conclusion of 2026 or shortly thereafter. Given Intel's customary practice of releasing its Z-series chipsets initially, the premium Z990 chipset will probably serve as the inaugural product of the Samsung-Intel 8nm partnership.

In tech, the old is often replaced to accommodate the introduction of the new. AMD has confirmed to Tom's Hardware that the chipmaker is effectively discontinuing the B650 chipset. The objective is to facilitate the transition to the new B850 chipset.

"AMD is working with channel partners to transition the B650 chipset to B850, offering improved connectivity and expanded PCIe Gen 5 support. With faster storage, more flexible expansion, and advanced networking capabilities, the B850 chipset provides a future-ready AM5 platform for gamers, creators, and professionals. The transition is already underway, with existing B650 stock at retail being depleted over the coming quarters," an AMD representative told Tom's Hardware.

AMD introduced the B650 chipset in 2022 to provide consumers with an affordable entry point into the AM5 platform. The keyword here is "affordable," as the B650 chipset isn't the entry-level SKU for AMD's 600-series chipset. Subsequently, the chipmaker launched the A620 chipset a year later, targeting buyers who seek only essential features.

With the launch of the AM5 platform in 2024, the premium X870E and X870 chipsets were the only ones available. The B850 chipset was released earlier this year, in January. It was public knowledge that the B850 chipset would eventually replace the B650 chipset, which had been on the market for nearly three years. It was only a matter of time before AMD publicly announced it.

AMD B850 Specifications

AMD uses the Promontory 21 chips for its 600-series chipsets and continues to do so on the latest 800-series chipsets. The x50 SKUs have historically shared closely similar and sometimes even identical specifications, which has led some enthusiasts to call AMD out for merely slapping a new sticker on the same chipset. In the case of the B650 and B850 chipsets, however, there are some slight differences in terms of features.

For starters, there was non-existent support for PCIe 5.0 on the primary x16 expansion slot on the B650 chipset. On the B850 chipset, however, it's optional, so it'll depend on the motherboard vendor whether it wants to implement the feature. Secondly, PCIe 5.0 support on the M.2 slot was optional for the B650 chipset, but it's mandated for the B850 chipset.

The discontinuation of the B650 chipset doesn't necessarily mean we won't see new B650 motherboards or the end of support from AMD's partners. Depending on existing inventory, motherboard manufacturers could still release more B650 motherboards.

AMD has halted production on its B650 chipset, according to Chinese tech forum Bobantang. The B650 has been the midrange chipset for the AM5 socket since its introduction in 2022 and has been one of the most popular options for PC users upgrading to AM5 processors.

In the original forum post (accessed via ITHome), the author claims that AMD has begun the shutdown stage of B650 production, and has informed its board partners of the change. While mass production has ended, supply of the chipset and motherboards using it is expected to remain high beyond this production halt.

Intel has finally announced its budget-focused chipsets for its latest Core Ultra 200 series desktop CPUs. B860 and H810 are aimed at the mid-range and entry-level price segments, with motherboards sporting these chipsets starting at $99 and $129, respectively.

B860 is the newest chipset in Intel's B-series lineup, sporting 14 lanes of PCIe 4.0 connectivity (not including the CPU lanes), four SATA 3.0 ports, 1Gb Ethernet, Bluetooth 5.3, and Wi-Fi 6 support. Memory support is purportedly limited to 6400 MT/s. USB-based I/O consists of up to 16 USB 3.2 ports and 12 USB 2.0 ports; however, bandwidth for those 12 ports can be changed to power dual 20Gbps USB ports, four 10Gbps ports, or six 5Gbps ports.

H810 is the new entry-level chipset in Intel's lineup, sporting lower PCIe connectivity, USB connectivity, and memory capacity. The chipset features four SATA 3.0 ports, Wi-Fi 6, Bluetooth 5.3, 1Gb ethernet, dual eSPI ports, and four SATA 3.0 connectors. Memory frequency support is rated at 6400 MT/s; however, DIMM slot capacity has been reduced to 1 DPC (DIMM per channel), meaning H810 only supports two memory slots at most. PCIe connectivity is rated at Gen 4 speeds with 8 lanes in total (again, not including CPU lanes). H810 supports up to 10 USB 2.0 ports with options to switch the ten USB ports down to dual 20Gbps ports or four 5Gbps ports, and up to four USB 3.2 ports.

As previously iterated, H810 motherboard pricing will start at $99, and B860 motherboard pricing will start at $129. Intel revealed four vendors for each motherboard series, with H810 comprised of Asus, Gigabyte ASRock and Colorful options. B860 was advertised with Gigabyte Maxsum, MSI and Asus ROG options. Inevitably we will see H810 and B860 motherboards from all of Intel's board partners, however these four main board partners for each chipset will probably be the first to release H810 and B860 motherboards to the market respectively.

Alleged specifications for Intel’s next-generation chipsets have leaked, showing eight new SKUs. According to the X post by hardware leaker Jayjihn, these are the features of the upcoming 800-series chipsets.

Intel’s Arrow Lake will slot into the LGA1851 socket. Although the next-generation processors will have the exact dimensions as the previous Alder Lake and Raptor Lake chips, they will have 151 more pins, which are physically incompatible with current Intel motherboards. Therefore, Arrow Lake will arrive with a new wave of Intel motherboards.

The Intel 800 series chipset list includes three new consumer desktop chipsets, two laptops, two workstation chipsets, and one enterprise desktop chipset. According to the leaker, there isn’t an H810 SKU for this generation of Intel chips, but we’ll have to wait until launch day to be sure. It would appear that the H870 chipset has replaced the H810 as the entry-level chipset. The H870 chipset has the latest PCIe lanes (24) out of the lot and lacks features such as PCIe and SATA RAID. The chipset doesn’t support a base clock or memory overclocking, either.

As expected, the Z890 chipset will spearhead the 800-series chipset family. It has the most PCIe lanes (48) out of the mainstream 800-series chipsets. Z780 supports CPU and memory overclocking. The flagship chipset also features different PCIe 5.0 configurations, enabling 1x16 + 1x4, 2x8 + 1x4, and 1x8 + 3x4 setups.

Although the leaker has included the W890 chipset in the table, he highlights that it isn't for Arrow Lake. Instead, the W890 chipset is for Intel's next-generation Xeon HEDT platform. Intel currently has the W790 chipset and will introduce a Sapphire Rapids Refresh very soon. Therefore, the W890 chipset will likely have a new socket to house Granite Rapids-X chips. You might also notice that the workstation W890 only has two memory channels. Still, the leaker said these are just preliminary specifications, with the mentioned chipset expected to have four channels in its final form.

These Intel 800-series chipsets will land alongside the Arrow Lake -S processors, but they’re not the only architecture expected to use this series. While we don’t know how many generations they will support, many Intel chipsets typically support two to three generations of chips. For instance, the Intel 600-series chipsets supported 12th Gen Alder Lake, 13th Gen Raptor Lake, and 14th Gen Raptor Lake Refresh chips.

AMD is preparing to launch its next-generation desktop CPU and motherboard series. The current flagship consumer motherboard chipset from AMD is the X670 (E), which we’d expect to be succeeded by the X760 (E). However, recent leaks suggest AMD will skip ahead and dub the next chipset X860 (E), running in parallel with Intel's 800 motherboard chipset series like the Intel Z890 flagship. Both x86 CPU vendors will thus simultaneously be offering 800 series chipsets - which shouldn't but may cause more confusion among customers.

Most major motherboard manufacturers will display their newest AMD and Intel motherboards at Computex 2024 in early June. On the AMD side of things, expect these motherboards to be compatible with the upcoming Ryzen 9000 series of CPUs, codenamed Granite Ridge.

According to Benchlife, the chipsets used with the Ryzen 9000 CPUs will be called the 800 series. This is the same as Intel’s new motherboard chipset, so the major CPU manufacturers will both have flagship motherboards using 800-series nomenclature.

It will be important to remember, though, that the similarity will mostly end with the numbering scheme. Each manufacturer’s newest motherboards will feature entirely different sockets. They’ll also support different processors. Just like motherboard / CPU combinations going back as far as I can recall, you won’t be able to socket an Intel CPU in an AMD motherboard, or vice versa.

Of course, this naming can always change between now and the actual release. AMD has, after all, made numerous changes in the naming for its Strix Point mobile Zen 5 platform. At one point, it was rumored to be named the Ryzen 9050 series. The latest rumors, however, claim that has changed to the Ryzen AI 300 series.

Despite the upcoming new AMD chipset, it seems some motherboard manufacturers are still improving upon the first-generation AM5-socket motherboards. Benchlife received a new, as yet unreleased, Gigabyte B650E Aorus Pro X USB4. While the motherboard sounds like it would be obsolete almost upon delivery, the specifications for the board tell a different story.

In fact, Benchlife notes that the “B650E Aorus Pro X USB4 motherboard matches the specifications of most AMD X870 motherboards.” It also features a reinforced PCIe slot supporting up to a maximum weight of 58 kg. I can’t for the life of me figure out why you’d need your PCIe slot to withstand so much weight, but there you have it.

Gigabyte hasn’t officially announced the new motherboard yet, but a teaser video on X (formerly Twitter) makes it clear something is coming soon.

The year was 2020, and Huawei released its HiSilicon Kirin 9000, an application processor for smartphones that used TSMC's N5 (5nm-class) process technology and packed 15.3 billion transistors — just before the company got listed on the Entity List of the U.S. government and lost access to TSMC. In 2023, the company launched its Kirin 9000S: a version of the Kirin 9000 made by SMIC using its 2nd-gen 7nm-class process technology. Nanoreview.net this week tested the new system-on-chip and its verdict was not exactly favorable.

The new Huawei HiSilicon Kirin 9000S processor can keep up and even leave behind its predecessor when it comes to general-purpose CPU workloads, but when it comes to power efficiency and graphics workloads, it is considerably behind the Kirin 9000 that is now more than three years old. The results are not particularly surprising as it is hard to beat a 5nm processor with a 7nm SoC unless you sacrifice power efficiency and cost.

Nanoreview.net tested both the original Kirin 9000 and the Kirin 9000S in popular benchmarks, including AnTuTu 10, Geekbench 6, 3DMark Wild Life, and numerous mobile games.

Huawei's HiSilicon Kirin 9000S demonstrated a similar total score (around 900,000 points) to the original Kirin 9000 in AnTuTu 10, but it was 33% behind in GPU performance than its ancestor. Surprisingly, despite lower clocks and the same core count, the Kirin 9000S was 4% faster in the single-thread Geekbench 6 workload and 17% faster in the multi-thread Geekbench 6 workload. The Kirin 9000 is also 20% faster than the Kirin 9000S in 3DMark Wild Life, which is probably because the company could not build a GPU that would be as fast as a 24-cluster Arm Mali-G78 (1536 stream processor) working at 759 MHz. 

Since SMIC's 2nd Generation 7nm technology is clearly much less advanced than TSMC's N5 production node, the Kirin 9000S is clearly significantly less efficient than the Kirin 9000, so expect smartphones based on the new SoC to offer shorter battery life unless they are equipped with a higher capacity battery.  

Although the new HiSilicon Kirin 9000S is tangibly slower than its processor, it still seems to be a quite good SoC for smartphones, so eventually Huawei might come up with a chip that will beat its 2020 product. The only question is whether it will be competitive against offerings from Apple, MediaTek, and Qualcomm.

The first die shot of Huawei's HiSilicon Kirin 9000s system-on-chip (SoC) published at Baidu reveals a monolithic die with a massive 5G modem, a huge image signal processor (ISP), a new neural processing unit (NPU) and custom CPU and GPU cores. Huawei's Kirin 9000S system-on-chip powers Huawei's new Mate 60 Pro smartphone. The chip is made by China-based SMIC using its 2nd generation 7nm-class process and stacking, then provided to Huawei in violation of US sanctions. 

Huawei’s HiSilicon Kirin 9000S SoC is a rather intricate SoC with four high-performance Taishan V120 cores and four Arm Cortex A510 energy-efficient cores (in two dual-core clusters), which shows that developers decided to take a different approach than Apple and packed more high-performance cores and fewer energy-efficient cores. By contrast, Apple's A17 Pro features two performance and four efficient cores. The SoC also features a quad-cluster Mailiang 910 GPU, which reportedly operates at a maximum clock of 750 MHz. While we do not have official information about the architecture that powers the GPU, Huawei likely uses Arm's Mali technology for its GPU.

While the CPU and GPU cores take up a substantial portion of Kirin 9000S's die size, it is noteworthy that the SoC also packs a large 5G modem and a huge ISP to enable advanced imaging capabilities.

According to TechInsights, the application processor is made on SMIC's 2nd Generation 7nm process technology, and it is a bit surprising that Huawei decided to integrate a modem rather than spend more die area on additional CPU and GPU capabilities. 

Back in the day, Huawei's HiSilicon arm was known as a leading designer of smartphone system-on-chips that challenged application processors from Apple, Qualcomm, and Samsung. Because of the U.S. sanctions against Huawei, HiSilicon lost access to leading-edge production at TSMC, and the company had to put advanced SoC development on pause for a few years as it needed to wait for China-based SMIC to develop technologies comparable to those from TSMC.

As it turns out, the company not only released a rather interesting SoC with a novel approach to integrate more high-performance cores and fewer energy-efficient cores, but it has also managed to pack in a 5G modem, a chip that could be made separately to improve yields.

Although the Institute of Electrical and Electronics Engineers (IEEE) yet has to formally ratify the Wi-Fi 7 (802.11be) specification, the technology is coming to market. Intel has already unveiled its first Wi-Fi 7 controllers and adapters, which will be coming to the market in various forms this year.

Intel currently lists two M.2-2230 draft Wi-Fi 7 adapters (as noticed by @ghost_motley): the Intel BE200 and the Intel BE202. Both adapters support 2x2 TX/RX streams, use 2.4 GHz, 5 GHz, and 6 GHz bands, and the BE200 has a maximum speed of 5 Gbit/s, which is significantly below the maximum transfer rate supported by the standard. The exact differences between the two parts is unknown at this point, but Intel claims that the BE200 is Wi-Fi 7 pre-certified. Both Intel's BE200 and BE202 support PCIe and USB interfaces and can be used for desktop motherboards and laptops.

The BE200 will be used by the upcoming Gigabyte Aorus Z790 Master X motherboard (PCB revision 1.2), as noticed by @momomo_us. Meanwhile, some other versions of the platform will use Qualcomm's Wi-Fi 7 QCNCM865 (PCB rev. 1.0) and MediaTek's Wi-Fi 7 MT7927, RZ738 (PCB rev. 1.1). To take advantage of Wi-Fi 7, users will also need to use Wi-Fi 7-compliant routers and access points.

Wi-Fi 7 promises a maximum aggregated bitrate of 40 Gbit/s, positioning it as a potential successor to wired Ethernet for the majority of users. To achieve these impressive rates, Wi-Fi 7 will harness three frequency bands: 2.40 GHz, 5 GHz, and 6 GHz, will expand channel width to 320 MHz, and incorporate 4096-QAM. However, many client devices will likely operate at slower speeds, as Intel's BE200 example demonstrates.

The technology will also build upon the foundations set by its predecessors. It will mandatorily support features like MU-MIMO (Multi-User Multiple-Input Multiple-Output) and OFDMA (Orthogonal Frequency-Division Multiple Access), which were introduced and supported by Wi-Fi 6 and Wi-Fi 6E. These enhancements are expected to optimize the efficiency and capacity of wireless connections.

Intel originally envisioned Wi-Fi 7's application in bandwidth-demanding tasks, especially in augmented reality (AR) and virtual reality (VR) headsets that used Intel's WiGig technology. While the IEEE plans to formally adopt the IEEE 802.11be specification by 2025, Intel and others are optimistic about the technology's performance and have already launched their first Wi-Fi 7-compliant controllers and adapters.

That said, it will likely be years before Wi-Fi 7 devices are ubiquitous. Full certification of the spec isn't expected until sometime in 2024, with the major rollout of compatible client devices coming after that. So if you recently bought a Wi-Fi 6 or 6E router, it should serve you well for now. But a year or two down the road, you may want to consider upgrading -- especially if you have a need for wireless speed and a desire to be on the cutting edge of consumer tech.

AMD has curiously taken a decidedly low-key approach to launching its value-centric A620 motherboards, burying the news of its A620 announcement late on a Friday night, but the company has finally shared the full details and slide deck with us about its new range of value-geared motherboards. To reduce pricing, the motherboards will not support the full range of Ryzen 7000 processors at their full power levels by default — none of the Ryzen 7000 X-series processors are guaranteed to be fully supported by default. These boards also eliminate some features of the more expensive AM5 motherboards, like overclocking and faster USB connectivity, in exchange for a lower price point.

Even though AMD's Ryzen 7000 lineup currently tops our list of the best CPUs for gaming, the supporting AM5 motherboard platform and required DDR5 memory have earned a reputation for egregiously high pricing. AMD even recently ran a promotion that gave you up to $125 off if you purchased a chip, motherboard, and memory, but that has expired today. 

That makes for a well-timed release of AMD's budget A620 platform with at least one motherboard that will retail for $85, but early indications point to pricing exceeding $100 for several new boards. Unfortunately, that means the overall A620 lineup will likely also be more expensive than expected. 

We don't have US pricing for most of the A620 motherboards, but while ASRock has one board for $85 and another for $99, ASUS announced EU pricing for three of its motherboards, with conversion to USD netting pricing of $151, $161, and $183. That's above what we expect for budget-class boards with reduced feature sets. We'll follow up with more US pricing as we receive it, but it seems there will only be a few truly budget-class boards with this generation of the A-series motherboards.

AMD's B-series motherboards are designed to support every Ryzen 7000 model at its full TDP range, sometimes resulting in seemingly ridiculously overpowered boards for the lower-end chips. This results in superior performance and forward compatibility, but also high pricing for B-series motherboards compared to Intel's competing chipsets. In contrast, Intel doesn't require its B-series motherboards to support the full peak power limits of its more power-hungry high-end chips, so vendors can cut back on power delivery (VRMs, etc.) to reduce pricing. Naturally, this results in reduced performance with the highest-end Intel chips on its budget boards.

AMD is now taking a similar approach with AM5. The A620 motherboards are designed to support chips with a 65W TDP, meaning models with a peak power consumption of 88W (PPT). Motherboard makers have the option to make more expensive models with support for higher power levels, but at its base, the A620 spec allows for a peak of 88W of power delivery.

For those base models, you can install chips with higher TDP ratings into an A620 motherboard, and it will boot if the BIOS supports it, but the chip will not operate at its full peak power consumption (PPT). This means the highest-end chips will lose some performance in heavily-threaded applications due to VRM limitations on some boards, but AMD expects the reduced power delivery will not impact gaming much.

In either case, this means that most bargain-basement A620 motherboards will only fully support non-X Ryzen 7000 models, as even the X-series Ryzen 5 chips all come with a 105W TDP. Not only do the boards not even support the base 105W TDP for the lowest-end Ryzen 5 X-series chips, but the 142W peak (PPT) is also 20% higher than the supported peak 88W PPT. That's a significant 76W delta.

This approach falls outside AMD's standard AM5 policy thus far, but it mostly makes sense. Higher-end chips aren't a good fit for this class of lower-end motherboards, and the lower power delivery will ultimately reduce pricing for budget builds. Users can choose to step up to higher-end A620 models if they want support for an X-series chip to run at its full TDP.

Additionally, this is the same approach that Intel uses for its lower-end fare. However, the fact that the X-series Ryzen chips all start at 105W means that the lowest-end A-series boards are significantly limited in their appeal. 

Overall, the new motherboards will fall into a lower price range than their B-series counterparts, but as you can see in the full slide deck we've included above, there are other sensible tradeoffs in features, with the boards stepping back from B650's one USB 20Gbps and six 10Gbps ports to two USB 10Gbps and two 5Gbps ports.

A620 provides a x16 PCIe 4.0 connection but doesn't allow that to operate in a 2x8 mode like B650. A620 also supports a direct-to-CPU x4 PCIe 4.0 connection for an M.2 SSD port, but doesn't have the option for a PCIe 5.0 M.2 port like B650. This is a nice step up from the x16 PCIe 3.0 supported on the previous-gen A520, though.

Of course, many bargain-basement A620 boards will only have one M.2 slot, but A620 has the same support for four SATA ports as B650. That's also two more SATA ports than the previous-gen A520 chipset. 

As before, the A620 chipset doesn't support manual overclocking, so changing voltages and frequencies or auto-clocking via Precision Boost Overdrive (PBO) is off-limits. Unsanctioned BCLK overclocking was possible on at least one A520 board, but for now, it's anyone's guess if that will be possible with A620. The A620 motherboards support memory overclocking via EXPO profiles up to DDR5-6000, but we're unsure if manual memory tuning is allowed.

From what we've gleaned, these motherboards use the Prominitory 21 (PROM21) chipset, the same silicon used for AMD's X670 and B650 motherboards. The AM5 platform leverages a new chiplet-based approach for its chipsets, sometimes using two PROM21 chips for higher-end chipsets, but A620 will only include one chip. Surprisingly, AMD didn't develop lower-end chipset silicon for these budget boards, but re-use helps reduce design costs. It can also deliver cost benefits from higher-volume production. We have seen signs of a PROM22 chip that will come later, but with the same features as PROM21.

Motherboards are coming from all of the usual suspects, including ASRock, ASUS, Gigabyte, MSI, and Biostar, but US pricing hasn't been announced for all of the A620 boards yet. We'll follow up with more details as they become available.

Computex 2022 is fast approaching, and it was pretty much assumed that AMD would give us some early details on its next-generation Ryzen CPU platform. However, Gigabyte today confirmed the news in a press release, stating that it will come to Computex bearing gifts in the form of AMD X670 motherboards for Ryzen 7000 processors.

The company will showcase four new SKUs, which will be vying for a spot on our best motherboards for gaming list: the X670 Aorus Xtreme, X670 Master, X670 Pro AX, and the X670 Aero D. Gigabyte also confirmed that all of the motherboards will come equipped with a PCIe 5.0 slot for next-generation graphics cards and a new M.2 interface for the incoming crop of PCIe 5.0 SSDs (which will support bandwidth of up to 14 GBps).

We should also mention that the arrival of AMD 600-Series chipsets will usher in the first new CPU socket design for Ryzen desktop processors since AM4 launched in early 2017 (and eventually spanned five processor generations). The new LGA1718 AM5 socket will first serve as a home for Zen 4-based Ryzen 7000 "Raphael" processors built using a new 5nm process node.

In addition to the aforementioned native PCIe 5.0 support, Ryzen 7000 processors will support DDR5 memory, which first debuted with Intel's 12th generation Alder Lake processors in 2021. In what could serve as a wrinkle for customers looking to cut costs when upgrading to new Ryzen 7000 processors, our supply chain sources indicate that new X670 and B650 motherboards will exclusively support DDR5 memory. Naturally, this requirement will hit enthusiasts in their wallets, given the price premium of DDR5 over legacy DDR4 memory.

This is a departure from what Intel did with Alder Lake and its accompanying 600-Series chipsets. Alder Lake processors can support DDR4 or DDR5 memory, and many motherboard manufacturers have opted to produce DDR4-capable designs to satiate enthusiasts that have already invested in DDR4 modules. 

Although there is sure to be plenty of excitement around AMD's incoming Ryzen 7000 processors and supporting motherboards, it's also likely that Computex 2022 will also see the announcement (or at least a teaser) of RDNA 3-based Radeon RX 7000 Series graphics cards. 

"High-performance computing plays such an essential role in our daily lives, and AMD is committed to always push the envelope on performance and innovation," said Dr. Lisa Su earlier this month. "At this year's Computex, AMD will share how we accelerate innovation with our broad ecosystem of partners."

It appears that Apple has kept a closely-guarded secret regarding its new M1 Max silicon. New pics of the underside of the chip reveal that it may actually have an interconnect bus that enables Multi-Chip-Module (MCM) scaling, allowing the company to stack together multiple dies in a chiplet-based design. That could result in chips with as many as 40 CPU cores and 128 GPU cores. Apple has yet to confirm provisions for chiplet-based designs, but the M1 Max could theoretically scale into an "M1 Max Duo" or even an "M1 Max Quadra" configuration, aligning with persistent reports of various chiplet-based M1 designs in the future. 

Apple has managed to impress the world not once but twice already with the performance of its Arm-based M1 CPUs. The company's latest M1 Max chip is a force to be reckoned with by itself - the chips' gargantuan 57 billion transistors enable Apple to scale up to 10 CPU cores and either 24 or 32 GPU cores (depending on the configuration you get), all in a single 5nm chip. Adding accommodations for a chiplet-based design would theoretically multiply compute resources, and thus performance. 

The interconnect bus would allow Apple to scale its chips by "gluing together" the appropriate number of M1 Max chips. But, of course, it's not simply a matter of flipping one M1 Max chip and aligning it with the second one; Apple would still have to use specific interposer and packaging options for a chiplet-based design.

Interestingly, Apple's M1 Pro chip (which fits between the M1 and M1 Max SoCs) lacks the interconnect bus — it's actually located in the extended half of the M1 Max (a beefier version of the M1 Pro). This likely means that Apple only expects users that need the additional graphics compute power in its M1 Max (such as graphics or television studios) to require further performance scaling via this chiplet design philosophy. 

Marrying two 520 mm^2 Apple M1 Max dies in an "M1 Max Duo" chip could deliver up to 20 CPU cores and 48 or 64 GPU units. It would also require an appropriate doubling of the system's memory to 128 GB. Memory bandwidth should also scale in such a system, up to 800 Gb/s. That seems doable within the current M1 Max design, although the 10,040 mm^2 of Apple silicon would be more expensive, of course.

Going for the "M1 Max Quadra" solution with 40 CPU cores and 128 GPU cores would be even more complicated. Perhaps an added I/O die, as the source suggests, is the correct solution, but possibilities abound. Apple could also sustain enough inter-die bandwidth via an I/O technology akin to AMD's Infinity Fabric. Whether or not the larger chips would require an I/O die remains an open question, as other leaks have suggested the design will be expanded in a monolithic design.

It's ultimately unclear how Apple would choose to handle memory bandwidth scaling - and any solution would have very increased platform development costs throughout. But then again, these theoretical "M1 Max Duo" and "M1 Max Quadra" products would cater to a market that cares more about performance and power efficiency than cost. 

If you believe the chatter around the water cooler, Intel's upcoming Alder Lake lineup of processors is coming our way as soon as October. And, of course, a processor needs to find a home in a motherboard. Intel's Z690 will be the top-end chipset for 12th Generation Core processors codenamed Alder Lake, and according to a report by Performance Computing Inquisitor, the new boards will come with cutting-edge memory and PCIe connectivity. As with all leaked information, even though these details do line up with what's generally known about the Alder lake processors, do take these new bits with a grain of salt. 

For starters, the leaked information claims the Z690 chipset supports both DDR4 and DDR5 DRAM. That means that it is up to the motherboard maker to choose which memory it decides to support on the motherboard, and it will ultimately result in mixed motherboard offerings from board partners like Asus, MSI, and Gigabyte, among others. In addition to the mixed memory support, the Z690 chipset has Gear 2 or Gear 4 modes for DDR5. This allows the memory controller to run at half or quarter speed, respectively, depending on the data rate, to boost overall throughput. 

The CPU outputs 16 Gen5 PCIe lanes, while the chipset provides up to 12 PCIe Gen4 lanes and up to 16 PCIe Gen3 lanes. This setup purportedly allows for higher-end SSDs to be attached, while also keeping GPUs on the newer PCIe bus standards.

Intel has decided to upgrade its Direct Media Interface (DMI) connection, which now uses 8 PCIe Gen4 lanes for communication between the processor and chipset, as pictured in the diagram above. It is worth pointing out that the chipset supports solid USB connectivity with up to four USB 3.2 2×2 (20Gb/s) ports. And for WiFi, there are CNVio modules that support Intel Wi-Fi 6E, and possibly even Wi-Fi 7. However, the latter is less likely in the near term.

As a general reminder, all of the aforementioned information should be taken with a grain of salt. Until Intel discloses further details, the specifications are uncertain and will remain that way until the launch later this year. 

The WD Black SN850 is easily one of the best SSDs that money can buy right now. But owners had been reporting a performance loss over 40% when installing the SSD in a M.2 slot that's connected to AMD's X570 chipset. Fear not however, Western Digital has identified the problem, and the company tells us a solution is en route.

According to the manufacturer's investigation, the SN850 suffers from a drop in write performance on some X570 motherboards when the maximum payload size (MPS) is configured for 128 bytes. The option basically dictates the maximum transaction layer packet (TLP) that goes through the PCIe controller. Obviously, a low value will cripple devices with higher MPS capability, since they're forced to operate at lower MPS settings.

For now, it's recommended that SN850 owners install their SSDs on a M.2 slot that's connected directly to the Ryzen processor. If all your M.2 slots are already occupied, you may just have to wait for Western Digital's new firmware, which eliminates the MPS restriction on the SN850.

Western Digital assures us that the new SN850 firmware will be available for download on or before July 12, 2021. The update process is fairly simple, since all you have to do is launch the Western Digital Dashboard software, and it'll prompt you for the update.

You can find Western Digital's full statement to us below.

Western Digital's statement: 

The WD_BLACK SN850 NVMe SSD can experience a decrease in write performance when connected to a chipset M.2 slot on certain motherboards, specifically when max payload size (MPS) is set to 128 bytes (128B). To resolve this issue, Western Digital will release a firmware update that eliminates a restriction in our product for this setting of MPS, expected to be available by July 12, 2021.

For optimum SSD performance, Western Digital recommends that customers download the latest BIOS version and drivers for their system and connect the WD_BLACK SN850 directly to the processor/CPU M.2 slot.

WD_BLACK SN850 customers can access/download the latest firmware via Western Digital Dashboard software. Once the firmware update is available, customers who open the Dashboard software will be prompted to make the update.

It appears that Intel's 600 series chipset — built for Alder Lake — will not support the Gen 5.0 standard, per a report from HardwareTimes. According to a PCI-SIG certification, Intel's future Alder Lake-supported chipsets will max out at Gen 4.0 speeds running in an x4 configuration, meaning the only Gen 5.0 support Alder Lake will see is through the CPU lanes alone.

We don't know why Intel cut Gen 5.0 support from the 600 series chipsets, whether the it's cost issues or capability issues. Either way, this could be a positive strategy for Intel to keep Alder Lake motherboard prices at a minimum.

When Gen 4.0 first came out, we saw a large jump in motherboard prices due to the motherboards requiring much higher quality materials and far more PCB layers. These are both required to ensure Gen 4.0 speeds are stable. This is especially true of AMD's X570 platform which has full Gen 4.0 support on both the CPU lanes and chipset lanes. We can only imagine the same thing will happen with Gen 5.0, and probably be worse, since Gen 5.0 is significantly faster than Gen 4.0.

But with Intel supporting Gen 5.0 only on the CPU, motherboard prices may not be as expensive as they could be. It's much easier for motherboard manufacturers to build boards around one or two Gen 5.0 PCIe slots than to build the entire board to support Gen 5.0.

Plus, most consumers and prosumers are rarely saturating Gen 4.0 speeds right now, even on Gen 4.0 NVMe SSDs, and we don't expect this to change over the next few years. So this Gen 5.0 issue shouldn't be a problem for most buying into the Alder Lake platform next year.

According to IC Insights, Samsung is soon expected to make a comeback in the semiconductor industry to once again overtake Intel as the largest semiconductor manufacturer in the world. Predictions believe this change will take place right around Q2 of 2021 (which isn't far from now).

Intel has long been a dominant payer in the semiconductor industry and currently holds the longest run as the number 1 semiconductor manufacturer in the world, starting in 1993 and lasting all the way through 2016.

It took 23 years before Samsung finally displaced Intel of its position in 2017, thanks to a growing supply of memory sales during that time. It was a good time for competition, and finally proved Intel could be beaten by another competitor in the semiconductor industry.

It should be of no surprise that Samsung was the company to beat Intel; over the past decade, Samsung has become a mega-corporation in the tech industry, becoming the worlds leading memory and NAND flash manufacturer, as well as producing many other devices such as TVs, phones, and smart home appliances.

But Samsung's lead was short-lived -- after just two quarters, the company suffered a 17% loss in revenue due to a sharp decline in memory sales allowing Intel to regain the number one position in 2018.

Luckily for Samsung, Intel's sales have mostly flatlined since 2020, leading to a minor decline in revenue. This has allowed Samsung, with its slow but continuous increases in revenue, to almost match Intel in sales performance over the past few months.

If this trend continues, Samsung should once again displace Intel as the lead semiconductor manufacturer.

In a move that flies in the face of our new telecommuting-heavy culture, TSMC will apparently be prioritizing automobile chip production as we move into 2021. This news comes from the Taiwan Economics Ministry (via Reuters) and was announced a day after a meeting between Minister Wang Mei-hua and the company’s senior executives.

While TSMC told the minister that its production capacity is currently full, it also noted that it will “optimize” chip production to hopefully free up capacity, and that it will prioritize auto chip production if that capacity opens up.

“Other than continuously maximizing utilization of our existing capacity, Dr. Wei also confirmed in our investors’ conference that we are working with customers closely and moving some of their mature nodes to more advanced nodes, where we have better capacity to support them,” TSMC told Reuters in a statement, referring to comments TSMC CEO C.C. Wei made in an earnings call earlier this month.

“If production can be increased by optimizing production capacity, it [TSMC] will cooperate with the government to regard automotive chips as a primary application,” the Taiwan Economics Ministry added.

According to Reuters, much of the pressure for this decision comes from German Economy Minister Peter Altmaier, who requested in a letter to Wang that Taiwan help ease a shortage of semiconductor chips in the auto sector, as the shortage is stalling Germany's pandemic recovery.

We’re uncertain yet how much of a problem this presents for computer enthusiasts. While companies like AMD and Nvidia rely on TSMC to produce their chips, TSMC’s comments about “moving some of their mature nodes to more advanced nodes” implies that the company may find some way to continue to move forward on new projects with these partners. At the very least, current capacity will not be affected.

Intel, notably, also relies on its own fabs to produce its chips (at least for now), and while that has come back to bite the company before, it also leaves it largely unaffected by decisions such as these.

Nvidia and Intel also have fingers in the car market, so it’s unlikely that this decision would hurt either company’s revenue that much.

The key deciding factors here are how long this shift in priorities will last and how demanding it will be, however. TSMC is set to spend $28 billion building its capacity this year, meaning the company may well be able to meet both its automotive and computing clients without much hassle.

After all, auto chips in 2020 only accounted for 3% of TSMC’s sales. Granted, the first half of the year saw reduced orders from automotive clients, but even a 27% auto chips sale jump in the second half of the year could not push auto chips above the 3% mark on total TSMC sales.

“They dropped their orders due to various reasons when demand was low amid the pandemic,” a senior Taiwan government official told Reuters. “But now they want to boost their production.”

But with current sales so low, is it even possible to boost it enough to make tech enthusiasts start sweating?        

It appears new 500 series chipset motherboards for Intel's next-generation Rocket Lake processors are fast approaching, according to Weixin, a Chinese news outlet. Weixin claim that Intel might be releasing new Z590, B560, and H510 via its board partners before the Rocket Lake launch on January 11. Weixin also believes the Rocket Lake CPU launch will happen sometime in late February or early March. As this is a rumor, take everything with a pinch of salt.

If this is true, then the motherboard release will coincide directly with CES 2021 which also begins on January 11. But we still don't know exactly why Intel is releasing Comet Lake supported motherboards a full month earlier than the CPUs intended release date. Potentially this could be good news if you are building a new Intel system in January and you want to use the new 500 series boards before upgrading to Rocket Lake a few months later. This could also enable the 500 series boards to be updated and for bugs to be addressed. But still, this would-be very niche and it would be highly doubtful lots of people will upgrade to a 500 series board before the CPUs launch. Not to mention, Intel's 400 chipset boards like Z490 should be forwards compatible with Rocket Lake including having PCI-E 4.0 support which Rocket Lake CPUs will start supporting.

We can get an idea of what these new 500 series boards will support via our what we know about Intel's upcoming Rocket Lake CPUs. The new chips will be supporting PCI-E Gen 4 out of the box, AVX-512 support, Thunderbolt 4, and more PCI-E lanes, specifically four more lanes dedicated to a single M.2 NVMe slot for storage, just like AMD's Ryzen processors; and an upgrade in bandwidth for processor to chipset communication, from four lanes to eight lanes. Plus updated ports like HDMI 2.0b and USB 3.2 Gen 2.2 (20G). So Intel's new 500 series boards will have to support all these features out of the box.

Besides these, we still don't know exactly what features H510, B560, and Z590 themselves will offer over existing 400 series chipset motherboards, we'll have to wait for that information once Intel talks about these boards next month during CES 2021.

AMD just updated its new StoreMI V2 to support AMD's 400 series chipsets like the X470 and B450 platforms. AMD initially only supported the updated version of the storage software on the more expensive X570 platform, but promised to expand support to older platforms throughout the course of the year. Now the company is delivering on its promise to bring StoreMI V2 to less expensive 400-series platforms. The software is free, which is especially helpful for budget PC builders that usually have B450 boards and run small SSDs + HDD combos, which is where StoreMI comes in handy the most.

StoreMI is a storage tiering solution that allows you to merge your SSD and HDD (or slower SSD with faster SSD) into one virtual drive. StoreMI moves data between both storage solutions, so it analyzes your data and moves frequently-used data to the faster SSD. In effect, the technique merges the capacity and pricing advantages of HDDs with the speed of the SSDs. That results in faster overall system performance, like snappier game and application loading.

AMD's original StoreMI software had a few issues. Most notably, it was a tiered solution, meaning it transferred data from the slow SSD/HDD to the faster SSD storage, but didn't keep a copy of the data on both drives. This has the advantage of making the full storage capacity from both drives usable, but if StoreMI's virtual partition ever glitched out or died, you would lose all the data stored on the faster drive. 

StoreMI V2 addresses the problem by moving to a caching system, which copies all data from the slower drive to the faster drive. That means there are two copies of any data held in the SSD. This lowers your storage capacity for the StoreMI volume to the size of the slower drive, however this solution is far safer because it's bulletproof if your StoreMI virtual partition gets corrupted – you will still have a copy of the data on the slower drive.

You can download the new StoreMI V2 Tech here. It is available for all 400 series, 500 series, X399 and TRX40 AMD platforms.

12 Entry-level and budget-friendly A520 desktop chipset based motherboards have been registered by ASRock for regulatory approval at the EEC. As a successor to the previous entry-level A320 chipset, the A520 is designed with Ryzen 3 and Athlon processors in mind and comes with a number of compromises over more expensive, feature-rich motherboards. They lack CPU overclocking capabilities of more expensive boards, and there is no support for PCIe 4.0 so we have to rely on sixteen PCIe gen 3 lanes for graphics. I/O connectivity is also more limited than the higher-end counterparts with 4 SATA ports, and 9 USB 3.1 Gen 2 ports. All of the boards are expected to be Micro-ATX and Mini-ITX form factors.

These are the models revealed in the regulatory approval register.

• A520M Pro4
• A520M Pro4 R2.0
• A520M Pro4 R3.0
• A520M-HDV
• A520M-HDVP
• A520M-HDV R2.0
• A520M-HDV R3.0
• A520M-HVS
• A520M-HVS R2.0
• A520M-HVS R3.0
• A520M-ITX/ac
• A520M/ac

These newly registered boards are expected to arrive some time in Q3 2020 and will replace the older A320 chipset series of boards. If you're looking at building a gaming system, your best bet will be to stick with the new B550 chipset, or X570 for high-end systems as these platforms support PCI-Express 4.0 and overclocking. For general-purpose use, or very-tight budget builds, the A520 boards will be worth considering.

One month ago, AMD released a new Ryzen chipset driver that had a few issues during installation. Then, last week the company dropped a new version that was supposed to fix these problems, which for the most part it did. However, it turns out that this new driver created other problems for some users, as spotted by members of the ComputerBase forum.

Although we haven't experienced any issues on our own systems, it appears that the new installer, despite running successfully, doesn't always update the GPIO driver to the latest version. There are also reports of increased CPU power consumption and incorrect operating voltages, along with other issues with Ryzen Master.

ComputerBase is now working with AMD, asking its users to send over a handful of users' log files, which ComputerBase forwards to AMD's driver team after anonymizing to figure out why the installations aren't going as expected.

• %userprofile%\AMD_Chipset_IODrivers.Log
• %userprofile%\Device_ID.log
• c:\windows\inf\setupapi.dev
• c:\windows\inf\setupapi.log
• c:\windows\dpinst.log

It's important to note that not nearly everybody is experiencing problems. But as with any software, there is always bound to be a group of people that get unlucky; however, the last couple drivers have apparently been causing a few more problems for users than normal.

If you are experiencing issues, you might want roll back to an older version if the problems are impeding proper operation of your system.

The AMD Ryzen Chipset Driver in question is version 2.04.04.111, which is built for AMD chipsets ranging from the A320 series up to X570 on the mainstream platform, as well as A399 and TRX40 Threadripper platforms. You can find the download page here. You'll have to fill in your system's details, and if you're having problems, look for the "previous drivers" link hidden below, skipping versions 2.04.04.111 and 2.03.12.0657.

Broadcom announced a series of chipsets for the new Wi-Fi 6E extension of 802.11ax that introduces Wi-Fi for the 6GHz band, and claims it is the industry's most comprehensive Wi-Fi 6E portfolio.

Broadcom announced four chipsets for residential applications and four for the enterprise. The most powerful chips have support for 4x4 dual-band Wi-Fi with 160MHz channel support. There are also two 3x3 tri-band SKUs with 80MHz channels and a 2x2 SKU with Arm processor. The 6GHz band allows for a maximum of 1.2GHz channel bandwidth, which could be through seven 160MHz channels.

Greg Fischer, senior vice president and general manager of the Broadband Carrier Access Products Division at Broadcom, said this about the chipsets in a statement:

"As momentum accelerates around availability of 6 GHz, Broadcom is excited to be on the forefront of Wi-Fi technology paving the way for ecosystem adoption of Wi-Fi 6E. With the industry's broadest portfolio of Wi-Fi 6E silicon, we will enable our customers to build a variety of products that unlock the tremendous potential of 6 GHz spectrum. This announcement demonstrates Broadcom’s continued leadership and unwavering commitment in driving the next Wi-Fi evolution for enterprise and residential WLAN as well as mobile devices."

Broadcom said it has already started sampling the chips. Wi-Fi 6E on 6GHz was announced last week and allows for 1.4Gbps at 7m with low latency.

Just two days ago, we reported that MSI announced two new X299 motherboards, including the new Creator X299. As spotted by VideoCardz, it appears that MSI has a similar board for AMD’s new Ryzen 3000 platform: the Creator TRX40.

The Creator TRX40 will be aimed at the high-end desktop Threadripper chips, which are expected to come with up to 32 CPU cores. MSI's slip follows other listings from Gigabyte of its TRX40-equipped motherboards.

Recent listings at the USBIF standards committee point to three new chipsets for the Threadripper platform: TRX40, TRX80, and WRX80, with the latter being rumored to target the workstation market, while the TRX40 and TRX80 chipsets slot in for enthusiasts. The TRX40 chipset paired with 3rd-Gen Threadripper chips is also expected to have quad-channel memory support and boast heaps of PCI-Express 4.0 lanes.

This kind of platform would make sense for the Creator series boards, as these are focused on expandability and storage. As such, we should expect lots of memory support and PCI-Express lanes for storage.

The leak for the MSI Creator TRX40 comes from a promotional page, although it appears to have been removed now. The promotion was simple: if you register your product and leave a review, you would be eligible for a 25$ steam voucher.

Sadly, so far, that’s all we know about this motherboard. We hope to learn more about the TRX40 based motherboards soon.

The European Commission today announced that it's fined Qualcomm €242 million (roughly $271 million) for "abusing its market dominance in 3G baseband chipsets" by selling its products below cost in an effort to force one of its competitors, Icera, out of business. Qualcomm said in a statement that it plans to appeal to the General Court of the European Union with the hope of reversing the commission's fine.

Baseband chipsets are integral to smartphones, tablets and other devices meant to connect to cellular networks. Making a smartphone without using a baseband chipset would be like building a pirate ship without sails; it won't be able to serve its intended purpose. There's a lot of booty to be had for any company able to seize a sizable portion of the market, which is exactly what Qualcomm did.

The European Commission explained how Qualcomm established its dominance in the baseband chipset market in today's announcement:

"Today's decision concludes that Qualcomm held a dominant position in the global market for UMTS baseband chipset between 2009 and 2011," it said. "This is based in particular on Qualcomm's high market shares of approximately 60% (almost three times the market share of its biggest competitor) and the high barriers to entry to this market. These include the significant initial investments in research and development to design UMTS chipsets and various barriers related to Qualcomm's intellectual property rights."

None of that is illegal, the commission said, because Qualcomm assumed leadership of the baseband chipset market fairly. The fine results from Qualcomm using its control over the market to sell products below cost to its largest customers (Huawei and ZTE) between mid-2009 and mid-2011. According to the commission, that "predatory pricing" was supposed to prevent Icera from competing for those customers.

Qualcomm general counsel Don Rosenberg responded to those allegations in a statement. “The Commission’s decision is based on a novel theory of alleged below-cost pricing over a very short time period and for a very small volume of chips," he said. "There is no precedent for this theory, which is inconsistent with well-developed economic analysis of cost recovery, as well as Commission practice."

Rosenberg also said Icera wasn't harmed by Qualcomm's practices because it was acquired by Nvidia for $367 million in 2011. Nvidia then "continued to compete in the relevant market for several years after the end of the alleged conduct." Qualcomm wants to "expose the meritless nature of this decision," he said and will "provide a financial guarantee in lieu of paying the fine while the appeal is pending."

The European Commission is hardly the first organization to accuse Qualcomm of anti-competitive practices. The company has also been scrutinized by the Federal Trade Commission, among other regulators around the world, and engaged Apple in a global battle over unfair license terms before the companies agreed to a multi-billion dollar settlement in May.  This is pretty standard for Qualcomm by now.

In a couple of previous articles, we put together comparisons of chipsets for readers to get a high-level idea of what the competing platforms have to offer users. So far, we've covered mainstream enthusiast chipsets in our AMD’s X470 vs. Intel’s Z390 feature, as well as the more budget-oriented options with AMD B450 vs. Intel B360, taking things like memory support, overclocking ability, I/O Interface and storage technologies into account, as well as motherboard selection and pricing.

This time, we're going to tackle the current high-end desktop (HEDT) platform chipsets, AMD's X399 and Intel's X299, along those same lines. As we've seen with the other chipsets, both will offer lots of options in terms of features and price points. As these are boards designed specifically for high-end hardware, prices do tend to start off higher than mainstream boards.

Regardless of your board choice, buying into one of these platforms will cost more due to the supported CPUs and quad-channel memory support. HEDT here really does mean high-end desktop, so your build budget needs to be ready if you want to step up to this class of computing. For this reason, many HEDT builders or buyers are pro-sumers or have a need for the abundance of cores, PCIe lanes, memory bandwidth, and other features that AMD's Threadripper and Intel's Core X platforms have to offer. If you're on the fence about which one is best for you, we'll dig deep into the chipset features below to find out if there's a clear winner in this battle.

CPU Support

Released in June of 2017, Intel’s X299/Basin Falls platform has been out for almost two years now. The X299 platform was released in conjunction with Skylake-X CPUs which, at the time, consisted of seven processors ranging from a 6c/12t part in the Core i7-7800X ($999) up to the 18-core/36-thread flagship Core i9-7980XE ($1999). All processors were built on the 14nm process, with TDPs from 140W to 165W. Intel CPUs supplement the 24 PCIe lanes from the chipset and add either 28 or 44 lanes depending on the installed CPU.

Intel also refreshed these Skylake-X based CPUs in late 2018 and brought out six refreshed chips with higher base clocks and similar peak boost clocks with core counts remaining the same. There is a single i7 (i7-9800X) and five i9s from the i9-9820X up to the flagship 18c/36t Core i9-9980XE ($1,999). In total, Intel has 15 different CPUs for the platform including the i7-7740x and i5-7640x which are Kaby-Lake based 4c/4t and 4c/8t parts. Note that the Kaby Lake X parts have been officially discontinued by Intel for nearly a year, but they're still readily available.

On the AMD side, team red's HEDT platform consists of seven CPUs dubbed Ryzen Threadripper. The Threadripper CPUs range from 8c/16t to 32c/64t monsters with a TDP of 180-250W. Clock speeds on the Threadripper range from 3.4GHz and boost up to 4.4GHz depending on the model. First-generation Threadripper uses a 19xxX naming convention, while the refreshed chips are 29xxX or 29xxWX. The entry level Threadripper 1900X is $549 (although when we wrote this is was on sale for $300), while the flagship 32c/64t 2990WX is $1800. The AMD CPUs also supplement their chipset sourced PCIe lanes (four) adding 60 lanes directly connected to the CPU.

Winner: AMD

Speaking solely about the chipset and its CPU support, both are able to handle the latest processors each platform has to offer. AMD's platforms typically last longer than a single generation or two unlike Intel who, at least on the mainstream platforms, seems to change with nearly every generation. The X299 platform has seen a couple of minor CPU updates, but this will be the end of the road it seems. There isn't any verified information if Zen 2 based Threadripper CPUs will work on X399 based motherboards, but it stands to reason it would given we haven't heard anything yet about a new HEDT chipset or socket, and AMD has stated that new third-gen Threadripper CPUs will arrive in 2019. With that in mind, while this area is a close call, the nod goes to AMD for platform longevity.

Memory Support

Moving on to memory support, both AMD and Intel platforms use a quad-channel setup capable of supporting 128GB of RAM on motherboards that have eight DRAM slots. Memory speed support starts out at DDR4 2400/2666 for Intel and 2666/2933 for the TR4 platform. Memory speeds for AMD top out around DDR4 3600, depending on the motherboard used. Intel on the other hand is able to support faster memory and is typically not as finicky. For example, on the more overclocking-centered boards like the Asus Apex and Omega boards, speeds are listed up to DDR-4133 for the six-core CPUs and above (even faster for the Kaby Lake four-core CPUs). Clearly the "your mileage my vary" adage is applicable here. ECC memory can be used on some boards, though without the ECC enabled.

Memory compatibility issues have plagued AMD’s Ryzen architecture since its release and limited even the mainstream platform to around 3200 MHz early on. But with AEGESA updates/platform maturation, this has improved. With the release of Ryzen 2/Zen+, these restrictions and compatibility issues have been resolved for the most part, but in general the AMD TR4 platform is more particular and doesn’t peak as fast as Intel. X399 and TR processors also support ECC memory. For a prosumer platform it can certainly be beneficial to run ECC where data corruption, errors, and system failures need to be avoided.

Winner: Intel

Though both platforms support quad channel memory, it is Intel whose compatibility and speeds have it stand out above AMD here. While it is true these platforms are not generally pushed to their limits by their owners, out-of-the-box compatibility is key, and Intel has more flexibility on that front whereas with AMD, users can be better served by selecting off the QVL list. If your use case requires ECC RAM for accuracy of results, the AMD platform has that ability across all CPUs, while Intel does not.

Overclocking Ability

Unlike our battle of the budget chipsets, both the X399 and X299 chipsets support overclocking, along with any CPU that fits in the socket. The big difference here between the platforms is the CPU's ability to overclock and the motherboard’s ability to support a high-powered CPU and keep the VRMs cool.

On the former, Intel CPUs can overclock more from their base clocks. The biggest factor in holding them back is heat. These processors output a lot of it when all cores and threads are overclocked and in use. For example, using P95 Small FFT on an i9-9980XE at 4.3 GHz (all cores and threads), the system pulled a bit over 700W from the wall. It isn’t easy cooling these CPUs, nor is it easy to supply them with clean power. A robust VRM and heatsinks on motherboards are required for best results. To that end, Intel board partners released updated motherboards as they learned more information about how much power can be used by these processors. Memory overclocking for Intel is also more fruitful because the platform generally reaches much higher speeds than with X399.

I have not personally overclocked Threadripper, but our CPU reviewer has, most recently with second-generation chips like the 2950X and 2990WX Both generations tend to top out not far past 4GHz. Again, with 180W TDP CPUs and pushing all cores and threads, chances are users will be heat limited on the CPUs before the motherboard gets in the way. That said, most of the boards appear to have solid VRM's and should be able to overclock without a lot of fan fare, and AMD's auto-overclocking Precision Boost software is a nice feature that simplifies the process. Overclocking memory is possible on the X399 platform, though it typically tops out at much lower values than on the Intel HEDT side.

Winner: Intel

Although both platforms can overclock the CPU and memory, Intel has refreshed their boards and has a larger selection of motherboards (which we discuss later in a bit more detail) better equipped to handle the rigors of clocking these high-wattage CPUs to their cooling limit. We saw updates from board partners on that platform adding some more heft to the VRM heatsinks and in some cases, upgrading them altogether. What puts X299 over the top in overclocking ability is the chipset/platforms ability to overclock memory higher than AMD.

I/O Interface Technology

Intel’s bump from X99 to X299 was quite a change. X299 brings in high-speed I/O (HSIO) lanes (this configuration came from the mainstream platform), allowing a lot more bandwidth from the chipset to the CPU which yields greater flexibility compared to before. AIBs can now select which features to implement off the new lanes, too. The X299 platform supports up to eight SATA 6 Gbps ports, 10 USB 3.0 ports, and 14 USB 2.0 ports. USB 3.1 Gen 1 or Gen 2 (5 Gbps or 10 Gbps) support will come through add-on chips. The X299 platform also supports Intel Optane Memory as well as bootable RAID and Intel RST.

As far as PCIe lanes for Intel, depending on the CPU it can have either 28 or 44 lanes with the board deciding how most of these are split up. When using one of the Kaby Lake CPUs, the PCIe lane count drops to 16 - the same as the mainstream platform. With a 44 lane CPU, many boards are able to support quad SLI and Crossfire, assuming the board has the slots. That said, some port sharing is still possible on this platform depending on the setup, meaning plugging in some devices will disable other ports or slots.

For the AMD TR4 platform, USB 3.1 Gen 2 support is native to the chipset and it can also support up to 20 USB 2.0 and 3.0 ports with up to twelve SATA 6 Gbps ports. For PCIe lanes, Threadripper and the X399 platform gives users a total of four PCIe 3.0 lanes and eight PCIe 2.0 lanes with the TR CPU adding 60 more lanes which then eclipses the Intel platform. Users can also RAID the M.2 modules and receive full speed without paying for a the VROC key as is required to get full speed on X299 (see here).

PCIe lane availability on X399 spans across all of its CPUs, unlike Intel where PCIe lane count varies by the CPU used. With the sheer number of lanes available, there are no ‘dark’ channels, ports, or lanes that get switched off when others are used. In other words, you can build multi-GPU setups and install multiple PCIe NVMe drives without port sharing.

Winner: AMD

The X399 platform takes this contest with its number of PCIe lanes as well as that value remaining the same across all the Threadripper CPUs. This configuration coupled with ‘free’ RAID capabilities for ultra-fast M.2 storage and its ability to eliminate port sharing gives AMD the crown easily.

Storage Options and Technology

Digging into storage options and technology specifically, Intel's X299 chipset includes more SATA ports (8 total), additional network ports including 10GB, PCIe storage (M.2), and more. The DMI 3.0 pathing (PCIe 3.0 x4) brings the HSIO concept we’ve seen in the mainstream platform to HEDT as well. Outside of the framework, we know the SATA ports can support RAID 0, 1, 5, 10 modes, while for the M.2 PCIe drives, a physical VROC key is be needed to enable full performance RAID modes on those modules.

X299 supports Intel Optane memory, utilizing 3D XPoint (which is non-volatile, unlike RAM) for drive acceleration, similar to Intel’s RST caching technology. Intel Optane memory shines brightest when accelerating a mechanical drive.

Similarly, AMD's X399 platform includes its StoreMi technology, which allows the use of up to 256GB of an SSD and some RAM to speed up slower mechanical drives and improve performance. StoreMi combines an SSD and HDD so the PC sees it as one large volume. The software works by storing the most frequently accessed files on the faster drive, so when that task is requested again, it will be accessed on the faster part of the storage array.

X399 supports both SATA- and PCIe-based RAID arrays in RAID 0, 1, 5, and 10. The big difference here is its ability to run RAID over PCIe without the need for a paid ‘key.’ Intel limits this feature on its HEDT platform to it from competing with its enterprise (Xeon) platforms that cost much more.

Winner: AMD

Though both chipsets offer users a lot of storage flexibility, only one of these offers the whole show without paying for additional keys: AMD. The X399 chipset gives users plenty of native SATA ports and plenty of PCIe lanes for additional storage across their entire lineup of CPUs. With Intel on the other hand, users will need to pay closer attention to their current and future storage needs when choosing a CPU to make sure it has the PCIe lanes available when/if needed. That said, if placing already fast PCIe NVMe drives in RAID isn’t your thing, then with the right CPUs, both offer plenty of storage options. AMD, though, can offer more without the issue of losing other ports.

Motherboard Selection and Pricing

For the X299 platform, Intel board partners have brought a total of 54 boards to the party in various sizes ranging from Mini-ITX to E-ATX. There are selections from ASRock, ASUS, EVGA, GIGABYTE, MSI, and Supermicro. Out of the major partners, only Biostar doesn’t have a board on this platform. In total, X299 currently offers one Mini-ITX board (ASRock X299E-ITX/ac), five MicroATX, 40 ATX boards, and eight E-ATX sized boards. Pricing on X299 boards range from around $160, up to $750 (Asus ROG Rampage VII Extreme Omega) with a mean somewhere in the low-$300 range.

For X399, there is a total of 15 motherboards ranging in size from MicroATX to E-ATX. There are no Mini-ITX boards for X399. Four board partners offer X399 based boards, GIGABYTE, ASUS, MSI, and ASRock. We do not see Biostar here either, nor Supermicro on the AMD HEDT side. There is one Micro-ATX board, the ASRock X399M Taichi, eight ATX boards, and six E-ATX boards offering a full range of features. The X399 motherboards start out around $240 at the low end and end, and top out around $620 (Asus ROG Zenith Extreme Alpha) with a mean in the same low $300 range.

Winner: Intel

Potential buyers should be able to find a board they like from either platform from MicroATX on up, with both X299 and X399 offering motherboards that will work for just about any situation. However, if you to build a small form factor (SFF) Mini-ITX PC with these HEDT parts, you'll have no choice but to use Intel/X299, as X399/AMD does not have a board of that size. Although pricing for X299 starts out lower than X399, it ends higher, with the average price of boards somewhere in the $300 range. But the nod here goes to Intel, as X299 has a lot more options, in every form factor, with lower starting prices.

Bottom Line

Both platforms offer a heck of a lot of CPU processing power and lots of high-end components. AMD has the core count and price-to-performance title, while in most cases (AMD SMT can be more efficient in some applications), Intel takes the overall performance title when comparing like systems (with the same core/thread count). Both can offer users the latest in connectivity through native support or third-party controllers/additions.

If overclocking is important, Intel's HEDT CPUs tend to give users a bit more headroom, although both usually get tripped up just past their boost clocks anyway, for different reasons. On both of these platforms, robust VRMs and VRM cooling is important for the best results. Memory overclocking on Intel boards can also reach higher speeds in general, and though AMD has improved, Intel is less finicky about module support. Though if ECC is needed, the AMD platform is your only answer here.

In the end, both platforms tied in our roundup here with three sections going to Intel (Memory support, Overclocking advantages, and Motherboard Selection and Price), and three to AMD (CPU Support, I/O Interface Technology, and Storage Options/Technology). The truth of the matter is, users really can’t go wrong with either unless they are trying to save some money (AMD) or go all out on performance (Intel). It really depends on what the purpose of the build is and what other hardware is selected as to which platform will suit the user best as they both have their pros and cons and each platform can be a viable solution.

More Face Offs

• AMD B450 vs. Intel B360: Mid-Range Platforms Square Off

• AMD X470 vs. Intel Z390: High-End Platforms Square Off

• AMD Ryzen Threadripper 2990WX vs. Intel Core i9-9980XE: Which HEDT CPU Is Best?

• AMD Radeon RX 590 vs. GeForce GTX 1060: Which Mid-Range GPU is Better?

• AMD Ryzen 7 2700X vs Intel Core i7-9700K: Which CPU Is Better?

• AMD Ryzen 2 vs. Intel 9th Gen Core: Which CPU Deserves Your Money?

Contrary to some speculation, AMD will likely continue to work with ASMedia Technology to build its chipsets, upcoming 500-series included, Taiwanese media Digitimes reported today. 

According to the Taiwanese publication, the integrated-circuit design company owned by Asus is "expected to land contract design orders for all mainstream PCIe chips from AMD even after the U.S. chipmaker rolls out X570 motherboard chipsets that support PCIe 4.0". Our own motherboard sources have told us that AMD could continue to incorporate ASMedia elements into parts of the chipset, such as USB 3.1 support. 

ASMedia is a pretty well-known name in the motherboard field. The company is mostly responsible for providing third-party USB, PCIe and SATA controllers for both AMD and Intel motherboards. As a matter of fact, ASMedia was the one who designed and produced AMD's Promontory chipsets, such as the 300-series and 400-series for the chipmaker's Zen processor microarchitecture. So when word of AMD designing the X570 chipset in-house got out, many thought AMD would ditch ASMedia.

AMD processors have been exposed to a few vulnerabilities in the past through ASMedia's chipsets. Correcting the issues takes a while as AMD has to go through ASMedia to do so. Therefore, having a more direct control over the design process could be a reason for AMD to go with an in-house design.

However, there's just one inconsistency with Digitimes' report: the time frame. The article says ASMedia will not complete the tape-out for PCIe 4.0 until the end of the year. However, the penultimate paragraph claims ASMedia will roll out tape-outs for PCIe 4.0 by the end of of the year.

AMD is expected to launch its Ryzen 3000-series processors roughly around Q3. One of the new chips' selling point is support for the PCIe 4.0 interface, so naturally AMD will want X570 motherboards available to customers to house them, one way or another. 

When purchasing a new computer--especially one that you’re building yourself—there are a lot of components to consider. And one of the pieces of the puzzle that often gets overlooked is the chipset.

Typically, your chipset choice is dictated by the CPU that you choose, and many people don’t put much thought into the differences between the chipsets of each platform. We already know that AMD’s Ryzen processors represent a great value compared to Intel’s latest Core processors, but what happens when you factor in the disparities between the hardware that allows the processor to communicate with the rest of the system?

AMD and Intel both offer a range of motherboard chipsets that cover all price points, from the entry level to enthusiast-class hardware. Today we’re looking at the flagship chipsets from each company on their respective mainstream platforms, AMD’s X470 and Intel’s Z390, to see which one offers the best combination of features for a high-end build.

CPU Support

Intel’s Z390 supports the company’s 8th and 9th generation Core processors, which gives you a wide selection of CPUs to pair with it. On the low-end, the chipset supports Intel’s dual-core Celeron G4900 and G4920, and Pentium Gold G5400 through G5600 CPUs. The chipset also supports Intel’s brand-new Core i9-9900K, Core i7-9700k, and Core i5-9600K, and the entire line of last year’s 8000-series i3, i5, and i7 processors.

Intel’s Z390 features the LGA 1151 rev2 socket, which Intel introduced with the 8th generation Core CPUs and 300-series chipsets. The socket layout is identical to the Z170/Z270’s LGA 1151 rev1 socket, but it features different power delivery specifications which render the Z390 motherboards incompatible with 6^th and 7^th generation processors (at least without some major tweaks in the BIOS and Intel's Management Engine).

AMD’s X470 chipset also supports multiple generations of CPUs. AMD X470 motherboards feature the company’s AM4 socket which is compatible with AMD’s entire lineup of Ryzen processors short of the high-end Ryzen Threadripper series. 

AMD’s new chipset pairs with the company’s first- and second-generation Ryzen 5, and Ryzen 7 processors, including the flagship Ryzen 7 2700X. The X470 chipset is also compatible with AMD’s lower-end processors, such as the Ryzen 3 lineup, including the Ryzen 3 2200G and Ryzen 5 2400G processors with Vega graphics, and the 7th generation of the company’s budget APUs.

Winner: AMD. This was a close call, because Intel and AMD's platforms both offer support for a wide range of processors in different price and performance segments and across two generations today. However, AMD's plan to stick with the AM4 socket until sometime in 2020 means eventual Ryzen 3000 CPUs will also almost certainly work with X470 chipsets, though you'll likely need a BIOS update. This makes X470 more appealing to builders or buyers who want to add more CPU performance in the future without having to swap out motherboards. 

Memory Support

AMD’s X470 motherboards support dual channel DDR4 memory with a maximum of four DIMM slots. Memory speed compatibility is dependent on the construction of the motherboard, the type of memory in use, and how many DIMMs are populated. If your motherboard’s PCB features six or more layers, it will support two sticks of single rank DDR4 at up to 2933MHz. But if the board has four layers, support tops out at DDR4-2667. Dual-rank memory tops out at 2677MHz on motherboards with two DIMM slots and 2400MHz when two of four DIMM slots are in use. The maximum supported memory speed when four DIMMs are installed is 2133MHz for single rank memory, and 1866MHz for dual rank memory.

Intel’s Z390 motherboards also support dual-channel DDR4 memory configurations with up to two DIMMs per channel for a maximum of four sticks of memory. Officially, Intel’s platform supports memory speeds of up to 2666MHz. Intel’s platform isn’t picky about dual and single rank memory modules.

If you limit your consideration to the manufacturers supported features, the X470 is compatible with faster memory than the Z390. However,  Z390 tends to be more forgiving with overclocked memory than AMD’s Ryzen platform. Despite Intel’s unwillingness to increase the officially supported memory clock speed, you should have no trouble installing much faster memory modules on a Z390 motherboard.

You can overclock your memory on the X470 platform, but you’ll have better results on Intel’s platform. There’s a reason that companies like G.Skill reserve their fastest memory modules for Intel’s platform.

Winner: Intel. On paper, AMD’s X470 motherboards support faster memory, but Intel’s Z390 will accept faster memory modules without introducing detrimental problems. Intel boards also work better with RAM eXtreme Memory Profile (XMP) overclocking presets (not surprising giving XMP is an Intel spec), making achieving higher memory speeds easier. AMD's partners have debuted workarounds, like A-XMP, but these typically aren't as robust as XMP profiles.

Overclocking Advantages

There's a whole lot to be said about AMD versus Intel on the overclocking front, but little to none of it has to do with the chipset. Intel chips tend to have more overclocking headroom, but far more AMD CPU models are unlocked for overclocking. And most of AMD's chipsets are allow for overclocking, while Intel only allows for overclocking on its high-end Z390 (and older Z370) chipsets.

But all of that is outside the scope of this story, which is just about AMD's X370 and and Intel's Z390 chipsets, both of which allow for overclocking, provided of course that you have an unlocked CPU to pair with your motherboard.

Winner: Tie. While there's much to debate about overclocking on AMD versus Intel, X370 and Z390 specifically are fundamentally tied here, in that both chipsets allow you to adjust clock speeds when paired with an unlocked CPU.

I/O Interface Technology

AMD’s X470 chipset features that same I/O capabilities as its predecessor. Like the X370 chipset, AMD's X470 supports two USB 3.1 Gen2 ports, four USB 3.1 Gen1 ports, and six USB 2.0 ports.

AMD’s Ryzen platform supports up to 28 PCIe lanes, but the CPU controls most of those lanes. Ryzen CPUs feature 20 PCIe 3.0 lanes, with 16 lanes dedicated to two 16x/8x PCIe slots for graphics cards, and four lanes to split between SATA and NVMe data transfer interfaces.

The X470 chipset adds another eight general-purpose PCIe 2.0 lanes, which board manufactures can dedicate to additional M.2 slots, 5/10GbE network interfaces, or other PCIe-based devices.

Intel’s 9th Generation Core processors and the new Z390 chipset feature more PCIe lanes than AMD’s Ryzen platform, but Intel’s chipset handles the bulk of the PCIe communication. Intel’s 9th generation Core CPUs feature 16 PCIe lanes, which are dedicated to the x16/x8 slots for GPUs or PCIe-slot SSDs. All other PCIe devices share the Z390 chipset’s 24 lanes.

Intel’s chipset features PCIe revision 3.0 lanes, whereas AMD’s chipset lanes are revision 2.0.

Intel also equipped the Z390 with more native USB ports than AMD’s X470. Z390 motherboards can offer up to 10 USB 3.1 ports, with as many as six of those being USB 3.1 Gen 2 ports. The specifications can support 10 USB 3.1 Gen 1 ports, and also address up to 14 USB 2.0 ports. But the total number of ports on the board can’t exceed 14 between all USB generations.

Intel’s Z390 chipset features integrated Intel Wireless-AC, with support for wireless Gigabit Wi-Fi. But that doesn't mean every board will implement wireless connectivity. Integration makes adding Wi-Fi less expensive, but plenty of board makers have chosen to leave it out to lower costs and further segment their product stacks.

Winner: Intel. The Z390 chipset offers a few more USB ports, and the specification includes built-in Wireless-AC. It also provides 12 additional PCIe lanes, which gives motherboard makers the ability to add more advanced features and components with their boards.

Storage Options and Technology

AMD’s X470 chipset offers up to six SATA ports for hard drive and SSD connections. Four SATA ports are linked to the PCIe bus and share four PCIe lanes with an optional NVMe interface. The chipset also supports two optional SATA Express ports--not that we've ever actually seen SATA Express drives for sale.

AMD’s X470 chipset spec also includes free access to AMD’s StoreMI technology, which allows you to use a small SSD and RAM to accelerate mechanical hard drives. StoreMI can combine a mechanical drive and an SSD so that your computer treats them as one volume. AMD’s software stores the frequently accessed files on the faster drive to ensure your regular tasks can be accessed as quickly as possible.

StoreMI also includes a RAM Cache feature, which enables you to dedicate up to 2GB of your system memory to further accelerate file access times.

AMD’s X470 chipset supports RAID arrays in striped (RAID 0), mirrored (RAID 1), and stripped with mirroring (RAID 10) configurations.

Intel’s Z390 chipset specifications also include six SATA 6Gbps ports for mechanical hard drives and SATA SSDs. Intel provides support for up to three SATA Express ports, and up to three PCIe M.2 slots for NVME SSDs.

Intel also offers acceleration technologies that speed up data access. Intel Rapid Storage Technology (RST) can combine a small SSD with a mechanical drive and make them operate as a hybrid drive. Like AMD’s StoreMI technology, RST identifies the most frequently used data and stores it on the SSD volume for faster access.

RST isn’t just an SSD caching technology, though. It also includes power management policies that help maximize the performance of your drive during heavy multitasking use cases. RST also offers advanced RAID management for up to six drives in striped (RAID 0), mirrored (RAID 1), striped with parity (RAID 5), and stripped with mirroring (RAID 10) configurations.

Intel’s Z390 chipset also supports Intel’s Optane Memory, which is another hard drive acceleration technology. Optane Memory uses protocols like the ones that underpin Intel’s RST technology, but in combination with ultra-fast 3D XPoint memory, which is much faster than a traditional SSD and isn’t volatile like traditional RAM.

Winner: Intel. AMD’s X470 chipset offers ample drive connectivity, and it includes free access to AMD’s new StoreMI caching technology. But Intel’s platform supports more drives and offers Optane Memory acceleration as an option. It is possible to pair an Optane Memory module with an an X470 motherboard using StoreMI, but it isn't officially supported by Intel, and we've seen conflicting reports of performance. 

Motherboard Selection & Pricing

AMD’s X470 and Intel’s Z390 are the top-tier chipsets for each respective company’s flagship mainstream processor platforms, but that doesn’t mean that you’ll pay an arm and a leg to get access to them. We found examples of X470 motherboards for under $120, and Z390 boards start at about $115.

Of course, you could pay much more for a top-shelf board. AMD X470 motherboards top out at around $300, but you could pay as much as $899 for a Z390 board if you’re so inclined.

The selection of X470 motherboards is somewhat limited. We found 21 boards available from five manufacturers, including Asus, ASRock, Gigabyte, MSI, and Biostar. Eighteen of those are ATX boards, while three examples feature the Mini-ITX form factor. We have not seen any Micro ATX X470 boards in the wild (these are generally relegated to the lesser B450 chipset).

Older Z370 boards are a little bit easier to find, even though X470 boards have been available for months and Z390 just hit the market this fall. We found more than fifty Z390 motherboards available, with five Mini-ITX boards, five Micro-ATX, and three EATX board within the bunch.

Z390 boards are manufactured by Asus, ASRock, MSI, Gigabyte, NZXT, and Supermicro. EVGA usually makes Intel motherboards, but it has not announced any Z390 products yet.

Winner: Intel. AMD covers the critical price points between $130 and $300, which is good enough for most buyers. But Intel’s partners have gone above and beyond with the options available for Z390 motherboards, giving builders many more options to choose from.

Bottom Line

AMD’s X470 offers all the features that the average builder needs. But Intel’s Z390 offers options for absolutely anybody--save perhaps those looking for the extra ports and PCIe lanes offered up by the high-end desktop platforms (X399 for AMD and X299 for Intel). If you’re interested in extreme overclocking, if you plan to plug in lots of storage devices, if you use a slew of PCIe expansion cards, or if you value high-speed memory modules, Intel’s enthusiast chipset is superior to AMD’s.

That said, if you value an upgrade path and you don't plan on pushing the limits of the previously mentioned features, AMD's X470 is still plenty capable. And for the budget-conscious, don't overlook the B450 chipset. It offers the majority of the features you'll find on X470, with board prices starting at just $60. And AMD's B450 chipset supports CPU overclocking, something Intel's competing B360 chipset does not.

More Face Offs

• AMD Ryzen Threadripper 2990WX vs. Intel Core i9-9980XE: Which HEDT CPU Is Best?
• AMD Radeon RX 590 vs. GeForce GTX 1060: Which Mid-Range GPU is Better?
• AMD Ryzen 7 2700X vs Intel Core i7-9700K: Which CPU Is Better?
• AMD Ryzen 2 vs. Intel 9th Gen Core: Which CPU Deserves Your Money?
• AMD Ryzen 2 vs. Intel Coffee Lake: What's the Best CPU Platform?
• AMD vs. Intel: Which PC Build is Better for Under $500

If you didn't win one of the 8086 Core i7-8068K CPUs Intel gave away during a brief sweepstakes, it's time to make other plans. In the U.S. that means Amazon, Newegg, or Micro Center for most. The first two have processors in stock as of today, both selling for around $425, or $25 less than the MSRP.

For those still in Taipei you can take your pick from two stores in the old computer mall with pricing starting around 14300NT (around $479) plus a 2% credit card fee. Just look for the glass display of retail boxed CPUs on the second floor for the better deal.

"The 8086 "sweepstakes" will be open to residents of the U.S., China, Germany, Canada, France, the UK, South Korea, Taiwan, and Japan. Each of those countries have 500 up for grabs, except Germany, China, and the U.S., which will have considerably more processors to give away (1,000, 2,000 and 2,086 respectively)."

Rumor has it that Intel will only produce 50,000 Core i7-8086K 40th anniversary processors, with the first 8086 handed out through the sweepstakes that covers nine countries. The first processor with a factory set 5GHz turbo clock feeding frenzy has already started. Good luck getting your orders in.

The release notes for a version of Intel Rapid Storage Technology (pictured above) leaked some interesting tidbits of information about some upcoming Intel desktop platforms.

First of all, the document mentions the much-discussed Z390 chipset that will allegedly replace the dated Z370 as the highest-end mainstream desktop chipset for Coffee Lake. There have already been plenty of leaks confirming the existence of Z390, but the new document states that the chipset will also support Cannon Lake processors. Unlike Intel’s current Coffee Lake processors that are fabricated on a 14nm process, Cannon Lake will be the first processors to use Intel’s long-in-development 10nm process.

The first Cannon Lake processors were set to release in 2017, but that didn’t happen. Early this year, Intel confirmed that it had shipped a small amount of 10nm parts, but the company has yet to make any official announcement of a Cannon Lake product.

The second big reveal in the document is the existence of a successor to the X299 chipset. A new high-end desktop (HEDT) chipset called the X399 is coming. That name is the same as AMD’s existing X399 chipset that the Threadripper processors use. When AMD took the name “B350” before Intel did, Intel chose to use “B360” instead. It looks like Intel isn’t backing down this time, so we don’t know how this situation of overlapping names will work out.

Intel’s X399 is listed as supporting Coffee Lake and Cannon Lake. Currently, X299 is the home of Skylake-X and Kaby Lake-X parts; there are no Coffee Lake-branded parts for that platform. The previously known successor to Skylake-X was Cascade Lake-X. Because the name Cascade Lake comes from the Xeon product line, it’s possible that Intel has chosen to rename those parts as Coffee Lake-X.

The X299 chipset and its socket, LGA 2066, is only a year old, so we don’t expect X399 to use a different socket.

Gigabyte’s X470 salvo consists of only high-end and enthusiast Aorus motherboards. There are three in total: the Gaming 7, the Gaming 5, and the Ultra Gaming. All three boards have dual M.2 slots, Gigabyte’s most advanced RGB lighting, and premium onboard audio.

X470 Aorus Gaming 7

Starting with the flagship, the X470 Aorus Gaming 7 comes absolutely loaded with RGB. Beyond the usual heatsink, I/O cover, and back-glow lighting accents, the Gaming 7 has RGB-lit DIMM slots and PCIe slots. Fortunately, all that bling is backed up with some pretty over-designed hardware. The Gaming 7’s power circuitry is cooled by one of the biggest motherboard heatsinks we’ve seen in a while. The extreme cooling extends to the M.2 slots, which are both on the front side of the board and are cooled by dedicated heatsinks. As if it just wanted to outdo its competition, Gigabyte used metal reinforcement on all of the Gaming 7’s PCIe x16 slots and even its RAM slots.

As expected, Gigabyte didn’t hold back on the value-add features, either. The Gaming 7 has integrated wireless like some other X470 motherboards, but it uses Intel’s new 1.73Gb/s-capable wireless solution that launched with the Coffee Lake mainstream chipsets. On the audio front, Gigabyte, like Asus, uses a dedicated ESS Technologies DAC chip and a dedicated headphone amplifier for improved sound.

X470 Aorus Gaming 5 and Ultra Gaming

Slightly down the line from the Gaming 7 are the Gaming 5 and Ultra Gaming. In terms of features, both of these are closely related, mainly separated by some bling factor. For starters, they both use the same power circuitry that is a step below the Gaming 7’s. The Gaming 5 loses the RGB-lit RAM slots but keeps the lit PCIe slots, whereas the Ultra Gaming has neither. Both boards lose the dedicated heatsink for the second M.2 slot, and the Ultra Gaming further lacks any integrated wireless solution. The Gaming 5’s integrated wireless is the same as the Gaming 7’s.

Features common to all of Gigabyte’s new X470 motherboards are the capability of full PCIe 3.0 x4 bandwidth on both M.2 slots, the inclusion of front and rear USB 3.1 Gen2 Type-A and Type-C ports, and the inclusion of two of Gigabyte’s voltage-stable USB 3.0 ports for long cable runs. All of the motherboards also fully support the Gigabyte RGB Fusion lighting ecosystem, which can use digital and static RGB LEDs. However, Gigabyte’s ecosystem uses different RGB headers from Asus, MSI, and Asrock.

Gigabyte didn’t announce pricing and availability of its new X470 motherboards.

Biostar launched the X470GT8 and X470GTN motherboards to pair with AMD’s 2nd-gen Ryzen processors. The GT8 is an ATX motherboard competing in the enthusiast gaming market, whereas the GTN is an ITX motherboard for the mid range of the small-form-factor market.

Unlike its competitors, Biostar brought only two motherboards to the Ryzen Pinnacle Ridge launch. Starting with the larger of the two, the X470GT8 has all the usuals, including two reinforced PCIe 3.0 x16 slots, four DIMM slots, six SATA 3.0 ports, and ALC1220-powered audio. Most competing motherboards have two PCIe M.2 slots, but the GT8 has only one, which is tucked behind a dedicated heatsink.

Aesthetically, the GT8 looks similar to Biostar’s other flagship motherboards. It adopts an RGB-lit rear I/O shroud and RGB-lit heatsinks. Biostar hasn’t migrated to digital-RGB lighting yet, so the board features just four-pin 12V RGB headers. The GT8 includes a general selection of ports on the rear I/O, including three video outputs (DVI-D, HDMI, and DisplayPort), USB 3.1 Gen2 type-A and type-C ports, and gigabit Ethernet provided by an Intel controller.

The GTN adopts a similar layout to other X470 ITX motherboards we’ve seen. In addition to the standard configuration of two DIMM slots and one PCIe 3.0 x16 slot, the GTN has the usual four SATA 3.0 ports and a rear I/O layout that’s nearly identical to the larger GT8. The GTN only loses a DisplayPort. Usually, motherboard manufacturers place secondary, slower M.2 slots of the back of the board only, but the Biostar GTN’s only M.2 slot is on the back. Although the slot has the full speed of a PCIe 3.0 x4 connection, a fast SSD located here stands a chance of being throttled due to overheating. Unlike the GT8, the GTN’s networking is powered by Realtek, and the audio uses an ACL892 codec.

Biostar didn’t announce pricing and availability for the X470GT8 and GTN.

As we’ve noted before, Intel’s Coffee Lake desktop lineup was sorely under-populated until the launch of the second batch of 300-series chipsets. These include the H370, B360, and H310. The new chipsets join the existing Z370, which launched with the first round of Coffee Lake products in late 2017, to complete the Coffee Lake product stack.

As is usually the case with a platform-related launch, there’s a smorgasbord of motherboard announcements. As we’ve already done with the new Coffee Lake laptops, we’ve compiled a list of all the new 300 series motherboards.

Asrock

Asrock thought this round of motherboard launches was the right time to correct a weakness in its existing motherboard lineup--not having enough RGB lighting. Asrock’s new Fatal1ty H370, B360, and B360M motherboards are the company’s first to feature digital RGB lighting support. The boards implement Asrocks new Polychrome lighting system, which uses the three-pin header that is shared by the Asus Aura and MSI Mystic Light lighting ecosystems. The Fatal1ty boards also all feature higher-end audio, provided by an ALC1220 chip, 10-phase power circuitry, and dual M.2 slots.

Next in the lineup are the Pro boards, which include the H370, B360, and B360M Pro4. These boards are largely the same as the Fatal1ty boards, but they lack RGB lighting and use a lower-end audio solution. Following the Pro4 boards are the a group of budget-oriented mATX boards that include the B360M-HDV, H310M-HDVP, H310M-DGS, and H310M-G/M.2 boards. The H310M-G/M.2 surprisingly features Polychrome RGB lighting, but the boards are otherwise the most basic of the product stack.

Rounding off Asrock’s new motherboards are a trio of ITX boards: the H370M-, B360M-, and H310M-ITX/ac boards. They all come with Intel AC Wi-Fi solutions built-in, but they don’t use the new Intel 9560 wireless card.

Asrock had many, many new SKUs to launch:

Asus

Asus, likewise, released a truckload of new motherboards, but an H310-based board wasn’t part of the haul. The H370 boards include the new ATX ROG Strix H370-F and ITX H370-I motherboards. Both feature RGB lighting and have dual M.2 slots. The ATX TUF H370-Pro loses out on the Strix boards’ more advanced power circuitry, but it integrates Intel’s new 9560 Wi-Fi card for gigabit wireless and Bluetooth 5.0. Finally, the MATX Prime H370M-Plus rounds out Asus’ H370 lineup as the most basic choice.

Asus’ B360-based motherboards are segmented in the a similar way. There’s the ATX ROG Strix B360-F and ITX B360-I boards and the wireless-equipped TUF B360-Pro ATX board. Instead of a Prime series B360 board, Asus has the ATX ROG Strix B360-H and MATX B360-G motherboards. These are basic gaming boards that still include extensive cooling and pretty cool PCB designs.

Biostar

Biostar jumped the gun and launched its new 300-series boards early. The company didn’t launch an H370-based board. Instead, it launched three B360 and one H310 motherboard. All of the boards are fairly basic.

The ATX B360GT5S and MATX GT360GT3S both have two M.2 slots. The most basic B360MHD and H310MHD MATX boards lack any M.2 slots and also lose two RAM slots.

Colorful

Colorful is a relatively new player in the consumer motherboard market, and it brought only one motherboard to the launch party--the Battle Axe C.B360M-HD MATX motherboard. It’s a pretty basic motherboard that has only one M.2 slot and lacks wireless, USB type-C ports, or any RGB lighting capability.

EVGA

Likewise, EVGA also only launched one motherboard, the H370 Stinger ITX. It's a basic but high-quality board that has solid power delivery circuitry, dual M.2 slots, and includes Intel’s 9560 WiFi and Bluetooth solution. Unfortunately, the board lacks any USB type-C ports.

Gigabyte

As one of the big four motherboard OEMs, Gigabyte delivered H370 and B360 motherboards galore. The company’s higher-end Aorus Gaming 3 motherboards, which include the H370, H370 WiFi, B360, B360 WiFi, and B360M WiFi, all include dual M.2 slots, Gigabyte’s RGB Fusion lighting, and extensive power circuitry cooling. The WiFi motherboards use the new Intel 9560 wireless solution. There’s also the lesser B360N Gaming WiFi ITX motherboard, but it uses an older Intel 433Mb/s WiFi card.

Gigabyte’s other 300-series motherboards include less interesting, non-Aorus-branded H370, B360, and H310 motherboards. Perhaps the only two that stand out in this bunch are the H370N WiFi and the B360N WiFi. Both of these pack the newer Intel 9560 WiFi card. The former even has dual gigabit Ethernet jacks and a USB type-C port, making it a good building point for NAS/HTPC.

Here are the rest of Gigabyte's B360, H370, and H310 motherboards launching today:

MSI

Last but not least, MSI also delivered 19 new H370, B360, and H310 motherboards. The H370 Gaming Pro Carbon is the highest-end of the lot. It features dual M.2 slots, extensive cooling, and RGB back-glow. Like MSI’s other recent RGB-enabled motherboards, the Gaming Pro Carbon has both three-pin digital and tradition four-pin traditional RGB headers. There’s also a B360 Gaming Pro Carbon with similar features. At a step down from the Gaming Pro Carbon series is the Gaming Plus series, which has H370, B360, and H310M versions. These boards don’t have much RGB lighting integrated onboard directly, but they still have RGB headers. In case you don’t fancy the red and black look, there are white Arctic versions of the B360 and H310M Gaming.

The H370M and B360M Bazooka and B360M Mortar motherboards don’t really fit cleanly into any category. The Mortar, which has a titanium version, has better cooling, power circuitry, and integrates RGB lighting, but it’s limited to the lower-end B360 chipset. The Bazooka boards, on the other hand, seem to slot in above the MATX Gaming Plus boards but below the Bazooka. Finally, MSI has four consumer H310-based motherboards and two crypto-mining-focused H310 motherboards. The H310-A Pro has six PCIe x1 slots and one PCIe x16 slot. The more extreme H310-F Pro ups the PCIe x1 count to 12.

Here's the rest of MSI's fresh new lineup:

It’s a tough time to be a PC builder or gamer. Graphics-card prices are astronomical, RAM sits in a similar celestial orbit, and HTC’s new Vive Pro VR headset will set you back $1,200 if you don’t have an existing set of sensors and accessories for HT's previous-generation headset. If you want to play on the leading edge, or anywhere close to it, it will cost you.

Indeed, from the PC-component perspective, about the only thing working in favor of increased performance and affordability at the moment is the CPU market.

The Lay of the CPU Land

The launch of AMD’s Ryzen chips last year, which brought eight-core CPUs into the mainstream for the first time, was followed later by Intel’s six-core Core i7-8700K. All of a sudden, the number of cores and threads available for your computing dollar was never better.

And that ratio may get better still—AMD’s next-generation Ryzen chips are expected to arrive in April. Leaked reviews of the AMD Ryzen 7 2700X and Ryzen 5 2600 seem to indicate that they are a decent upgrade over the initial Ryzen chips from 2017—especially when it comes to the latency issues that plagued last year’s Ryzen chips in some games.

What’s Intel’s Next CPU Step?

With new Ryzen processors about to roll out, our interest--of course--turns to what Intel will offer up as a response, and as a replacement for chips like the Core i7-8700K. (That key "mainstream flagship" CPU debuted just about six months ago, in early October of 2017.) Let’s take a look at what we know (or at least what we think we know) about the putative Core i7-9700K, and what it will need to compete with AMD’s increasingly tough competition.

Multiple rumors and leaks strongly indicate that the ostensible Core i7-9700K will sport eight CPU cores and 16 threads, in contrast to the six cores and 12 threads on the Core i7-8700K. If this turns out to be true, it will be a first for Intel on a mainstream-grade CPU platform—as opposed to the enthusiast-targeted Core X-Series platform, which goes all the way up to 18 cores (if you can afford to drop a couple of grand on a CPU). But the eight cores on the Core i7-9700K, if that turns out to be accurate, only just matches what AMD has to offer in terms of cores and threads on its mainstream Ryzen CPUs; it doesn’t surpass the competition.

10 Nanometer (or Not)?

Of course, a shift to the long-promised 10nm process node would help Intel move further ahead of AMD without necessarily adding more cores. But that’s looking increasingly unlikely with Intel’s next round of desktop chips. Intel’s Gregory Bryant (senior vice president and general manager of the Client Computing Group) proclaimed at CES 2018, back in January, that 10nm chips were shipping to customers. But documents released weeks later seem to confirm that those 10nm “Cannon Lake” parts that shipped in 2017 were low-power, dual-core parts without integrated graphics. These aren’t the Cannon Lake chips that enthusiasts and gamers are looking for.

In fact, it’s looking likely that the Core i7-9700K will be a so-called “Coffee Lake Refresh” chip built around Intel’s 14nm++ process. Why do we say that? For starters, we have heard persistent rumors and leaks around an upcoming Z390 chipset. Most recently, a listing for "motherboard-specific sensor info for MSI B360/H310/H370/Z390-Series," showed up in a recent AIDA64 update. (AIDA64 is a benchmark and diagnostic suite often used to test pre-release hardware.)

What a Chipset Tells Us

We now know that the other 300-series chipsets mentioned above are real—in fact, we’ve tested a few motherboards based on them. So there’s no reason to think that a Z390 chipset won’t arrive soon—at Computex 2018 in July, perhaps.

Now, the “3” in the Z390 name almost certainly indicates that the chipset is aimed at Coffee Lake CPUs, just like the other 300-series chipsets. What does that tell us? If Intel knew it was going to have 10nm "Cannon Lake" desktop chips ready sometime in 2018, it seems extremely unlikely—though certainly not out of the question—that the company would be launching yet another high-end chipset for its current-generation processors halfway through the year.

If we were placing bets, we’d put our chips (pun intended) on the likelihood that Z390 was designed specifically for a line of Coffee Lake Refresh chips, which would include the Core i7-9700K. Alternately, the Core i7-9700K could be a limited stop-gap between chip generations, just like the pair of "Devil’s Canyon" chips were back in 2014.

Can Intel Push Coffee Lake Higher?

Of course, even assuming all the above leaks and assumptions are true, there is quite a lot we still don’t know about Intel’s future mainstream flagship chip. This wikichip page has a fairly detailed spec list based on existing leaks and rumors. It may prove to be accurate, but it doesn’t say anything about clock speed—base clocks or turbo-boost rates. And if the Core i7-9700K is indeed a Coffee Lake chip that doesn't end up featuring architecture or major process-node changes, clock speed (plus core count) would tell us most of what we’d need to know about the performance of this upcoming chip.

That said, the Core i7-8700K already has a top Turbo Frequency (at stock settings) of 4.7GHz. It seems likely that, for Intel to set the new chip apart from its predecessor, the stock Turbo clocks would have to hit 5GHz or higher. If Intel can do that, while pushing up the base frequency to about 4GHz and keeping thermal demands in the same 95-watt range of the Core i7-8700K, that indeed would be impressive.

A Lot Comes Down to Pricing

Of course, we don’t know what Intel or AMD will charge yet for their next-generation CPUs. AMD has been extremely price-aggressive since the launch of Ryzen last year, often undercutting Intel’s comparable parts by a significant amount. With its finances now solidly in positive territory for the first time in years, AMD may continue to pressure Intel on chip prices. After all, the CPU underdog is making money and gaining market share with this strategy. So why change it?

But CPU prices aside, Intel also faces problems of affordability when it comes to the platform as a whole. AMD’s AM4 boards have been out for a year, and the company says that new CPUs will be backward-compatible with existing AM4 boards (after a BIOS update, of course). Intel’s chips, meanwhile, have not been backward-compatible with previous-generation boards in recent years. And on top of that, Team Blue only just released its lesser 300-series chipsets, after six months of essentially forcing consumers interested in 8th-Generation Core processors to overspend on motherboards—particularly upgraders and builders who don’t care about overclocking, Optane storage, or multiple-graphics-card setups.

Intel Has Options

So, where does Intel stand in the desktop CPU space with next-generation chips like the Core i7-9700K, and what does it need to do to gain ground on its newly resurgent competitor?

Offering more cores for less money would be a good start. That seems unlikely in the short term. Offering up more-affordable motherboards that work with more than one processor generation wouldn’t hurt, either. Both tactics seem to be working for AMD, and Intel clearly has the coffers to cut its margins thinner than AMD can.

Another interesting option for Intel to compete in today’s world of ever-overpriced graphics cards would be a socketed version of the "Kaby Lake-G" CPU that we just tested in the Intel NUC 8 VR. A custom build around a Kaby Lake-G chip would allow many users to game without the added expense of a graphics card, as well as making truly compact mainstream gaming systems more viable than they have been in the past. Given the rise of titles like Fortnite and Overwatch—which aren’t that graphically demanding—the market could be ripe for just that kind of system. And Intel clearly has some kind of designs on the GPU market, as evidenced by the company’s hire of AMD’s Raja Koduri late last year.

But while Koduri is certainly an interesting hire, a socketed Kaby Lake-G is extremely unlikely, given that AMD makes the Vega graphics in those chips. It was surprising to see AMD selling Intel its graphics silicon for comparatively low-volume products such as the new Intel NUC and high-end convertibles like the Dell XPS 15 Convertible. But we highly doubt that AMD is desperate enough to cut into its own core desktop-CPU and graphics businesses by helping Intel compete against itself and its own socketed graphics-equipped products, like the "Raven Ridge" AMD Ryzen 5 2400G.

It's been six long months since Intel first debuted its Coffee Lake processors and the 300-series chipset. Unfortunately, Intel released only six Coffee Lake processors in the initial salvo, but now it has added some more heft to its Coffee Lake lineup. Intel also only released the Z370 motherboards, leaving enthusiasts without a value motherboard option to pair with Intel's locked processors. 

Now we know why: Intel's initial chipsets were merely an iteration of the previous-gen 200 series, meaning they featured the same specifications, while today's launch finds the company releasing an entirely new lineup of chipsets based on the Cannon Lake PCH.

First, let's cover the remainder of the Coffee Lake lineup.

New Coffee Lake Processors

Intel offers standard processors that feature normal TDP ranges, as we see with the top three processors in the chart below, as well as T-Series "power-optimized lifestyle" models that come with lower TDP ranges. As such, those processors are designed for low power applications and feature reduced clock speeds and overall performance. As we can see, that reduced performance doesn't come with a reduced price tag, thus making T-Series processors thoroughly uninteresting for the enthusiast crowd.

Intel's new additions don't hold many surprises, other than the fact that the release is relatively narrow and doesn't include any new standard Core i7 models. The new Core i5 and i3 models are expected iterations of the existing 14nm++ Coffee Lake lineup. The Core i5 models come with a 6C/6T design, and the Core i3 model has the expected 4C/4T architecture. Much like their Coffee Lake predecessors, the Core i5 models support up to DDR4-2666 while the Core i3 models support DDR4-2400. The only real surprise comes in the form of the Core i3-8300's 62W TDP, which is a 3W reduction compared to other Coffee Lake Core i3s.

The Core i5-8400 is a popular processor with enthusiasts even though it comes with a locked multiplier, but the Core i5-8600 may be the processor to beat. It features a 500MHz reduction in base frequency compared to the Core i5-8600K, but it also comes with the same 4.3 GHz boost frequency. All the other key specifications, aside from the 65W TDP, are identical to the Core i5-8600K that comes with a $44 premium. The Core i3-8300 also makes for a compelling offering that bridges the huge price gap between the -8100 and -8350K. As per Intel's new policy, it is no longer sharing multi-core Turbo specifications.

Intel Expands The 300-Series Chipsets

Intel's new chipset is a far more interesting topic. It's based on the unreleased Cannon Lake PCH and comes with native support for USB 3.1 Gen2 (up to six ports). Intel uses the 14nm process for the new PCH, whereas the Z370 features a 22nm process node. Logically, that means the new chipset should have a lower TDP rating, but Intel hasn't shared specifics.

Intel also doubled Wi-Fi capabilities with support for Wireless-AC 2x2 160 MHz. Intel integrated the MAC into the chipset, but it still requires a PHY for operation, thus reducing costs and avoiding additional FCC requirements for the chipset. Intel offers its own CNVi module as the other half of the integrated solution, but the company also supports third-party solutions. This speedy connection, which debuted with the Gemini Lake processors, provides more bandwidth than a wired connection with up to 1,733 Mbps, but it requires a router that supports the feature.

As expected, the H-, B-, and Q-series chipsets do not support overclocking, and they cannot split the 16 CPU-based PCIe 3.0 lanes across multiple devices. The H310 chipset doesn't support Optane memory nor Intel's Smart Sound technology. The Q370 chipset is the lone vPro-compatible offering.

The H370 supports the maximum 24 PCIe 3.0 lanes, which Intel pares back to 20 for the H370, 12 for the B360, and six PCIe Gen 2.0 lanes for the H310. The H370 and Q370 chipsets support the full 30 HSIO (High Speed Input/Output) lanes, while the B360 is pared back to 24 lanes, and the H310 only supports 14 lanes.

All the chipsets feature a varying number of maximum supported USB Gen 3.1 ports, which is further split into the number of allowable Gen 1 and 2 ports. Intel also differentiates the chipsets by removing the Rapid Storage Technology feature entirely from the H310 chipset and placing various restrictions on the number of supported PCIe storage ports for each chipset. Naturally, that also has an impact on supported RAID levels. Intel also completely disabled support for CPU-attached PCIe storage for all but the Z370 and Q370 motherboards. 

The chipset supports Wake-On-Voice and enhanced voice recognition for Alexa and Cortana, along with other voice-activated personal assistants. The new audio DSP supports using up to five independent voice-activated applications simultaneously while the PC is in a low power state. These new features only apply to the new H-, Q-, and B-series motherboards; the Z370 motherboards will not support them. Intel also announced support for increased Optane functionality.

The new chipset also supports Windows 10's modern standby, which allows for lower power states during idle and faster resumption from low power states. California's CEC has established crushing new power requirements that are coming in two tiers. Intel's modern standby feature is compliant with the tier 1 regulations that kick in on January of next year. The tier 2 regulations come into effect in July 2021. To meet the requirements, Intel had to bring support for the C10 power state (which is currently only present on mobile processors) to desktop processors. 

Intel also provides us with its internal performance data compared to five-year-old systems. This launch marks the first time that Intel has used results that represent the impact of the Spectre/Meltdown patches. We know that older processors suffer from a much larger performance hit from the patches, so while these results represent actual post-patch performance, Intel likely gains a small bump in several of the benchmarks that compare the older Ivy Bridge system to the new processors. We've included the test notes below in a click-to-expand format.

Thoughts

Intel's new chipsets are based on the upcoming Cannon Lake-based PCH and bring a newer process node and features to the desktop platform. The new chipsets feature expanded features that may leave enthusiasts with Z370 motherboards feeling left out in the cold, as Intel has no plans to bring the partially-integrated Wireless-AC support, improved audio DSP, or integrated USB 3.1 features to the now-aging platform. That's a shame, because Z370 motherboards are the only 300-series chipset that supports overclocking. We have heard rumblings about the new Z390 chipset, and recent support pages have also left clues to its existence, but Intel doesn't discuss future products, so we don't have further details.

Intel's new list of Coffee Lake processors is surprisingly light, but the Core i5-8600 and Core i3-8300 look like compelling options that should make an appearance in our labs soon.

Motherboards based on Intel’s H370, B360, and H310 chipsets for its Coffee Lake line of CPUs have leaked on some Asian retailer websites. Among them are ATX and ITX boards from Asrock and Gigabyte.

Not long after a trio of new Coffee Lake CPUs were leaked by an online retailer, motherboards based on the more budget-oriented chipsets that are set to accompany them have also leaked. We’ve known a bit about the H370 and B360 chipsets since before Coffee Lake launched, and word of the H310 has appeared in various leaks.

Considering that Z370 is practically a spec for spec refresh of Z270, we can surmise what the H370 and B360 (which isn’t named “B350” because AMD is already using it) will have for specs. They should be to the Z370 what the H270 and B250 were to Z270 for Kaby Lake. Our guess is that the H370 will have 20 PCI-e lanes (versus 24 on Z370) and eight native USB 3.0 ports (versus 10 on Z370), while the B360 will have only 12 PCI-e lanes and six native USB 3.0 ports. The B360’s predecessor, the B250, also didn’t have any built-in RAID capability.

The leaked boards don’t have any surprises. Most of the ATX H370 boards show dual PCI-e 16x slots and dual M.2 slots. The spec categories that have been listed differ for every board, though, so we doubt they can be trusted. The takeaway is that there will be no shortage of cheap ITX boards available for those hoping to build six-core HTPCs or mini-PCs. Z370 ITX motherboards haven’t exactly been cheap, after all.

Samsung announced that it has begun mass production of chips using its second-generation 10nm FinFET process, called 10LPP ( Low Power Plus)

10nm 2nd Gen

A year ago, Samsung began mass-production of chips on its first generation 10nm Low Power Early (LPE) process. The company has since made further improvements to the process technology, which can now bring an increase in performance of up to 10% or a reduction in power consumption of up to 15%, depending on the chip makers' priorities.

Other advantages for 10LPP include a faster time-to-market for chip companies such as Qualcomm. The technology is only an iteration of an existing process, so it's easier for chip makers to upgrade to the new process. Manufacturing yield has also been significantly improved, which should lower costs for chip makers, too.

The 10LPP chips are first expected to appear on the market early 2018, with wider availability expanding throughout the year.

“We will be able to better serve our customers through the migration from 10LPE to 10LPP with improved performance and higher initial yield,” said Ryan Lee, vice president of Foundry Marketing at Samsung Electronics. “Samsung with its long-living 10nm process strategy will continue to work on the evolution of 10nm technology down to 8LPP to offer customers distinct competitive advantages for a wide range of applications,” he added.

Next-Gen Process Technologies

Samsung also announced that its newest manufacturing line, called S3, will begin production of chips using 10nm process technology or the upcoming 7nm process with Extreme Ultra Violet (EUV) lithography. Samsung has one other fab in South Korea (S1) and another in Austin (S2).

Samsung will also continue to improve the 10nm process with the 10LPU (Low Power Ultimate) generation, which should bring an additional 5% increase in performance. The following 8LPP process generation should provide up to 10% reduction in die size with 10% lower power consumption.

Over the last few months, security researchers from Purism, Positive Technologies, and even Google have found ways how to disable the secretive Management Engine (ME) firmware on Intel processors. This seems to have prompted Intel to do a review of its firmware and plug most of the holes that allowed the researchers to take over the ME.

Researchers Disable ME

This year, security researchers from Purism, the maker of privacy-focused Linux-based laptops, and Positive Technologies started looking for ways to disable Intel’s ME firmware. Many privacy activists, including Purism's security researchers, worried that ME could be used as a backdoor.

The closed source proprietary model of the firmware also denies most people—except the NSA—the ability to see what it does on a computer, which means it might come with security flaws that could be exploited by sophisticated attackers.

Both of these worries prompted Google to work on disabling ME for its servers so it can be more sure that it couldn’t be exploited by attackers.

The latest major revelation around ME vulnerabilities was a recent announcement from a Positive Technologies researcher that they have achieved full takeover of ME (and therefore of the computer in question) via USB. The researcher didn’t reveal more at the time, but he’s about to present the hack at the next Chaos Computer Club conference (34C3) at the end of December.

Intel Announces Fix For The Flaw

Intel announced that it has completed a security review of its ME firmware as well as the Intel Server Platform Services (SPS) and Intel Trusted Execution Engine (TXE) with the goal of enhancing firmware resilience. The review was prompted by the latest work by “external researchers.”

Intel identified multiple security flaws in the ME firmware versions 11.0/11.5/11.6/11.7/11.10/11.20, as well as SPS Firmware version 4.0 and TXE version 3.0. As Purism told us recently, version 11 of ME is quite different from the one Intel used before, and it also runs on a separate x86 processor, as opposed to an Arc processor like before. Versions 11 and later are available to Intel 6th-gen processors (Skylake) and newer.

The full list of impacted CPUs includes:

6th, 7th & 8th Generation Intel Core Processor FamilyIntel Xeon Processor E3-1200 v5 & v6 Product FamilyIntel Xeon Processor Scalable FamilyIntel Xeon Processor W FamilyIntel Atom C3000 Processor FamilyApollo Lake Intel Atom Processor E3900 seriesApollo Lake Intel PentiumCeleron N and J series Processors

As many privacy and security activists have feared, the recent flaws could allow an attacker to gain unauthorized access to ME functionality and third-party secrets protected by the ME, the SPS, or the TXE.

To determine if the recently found vulnerabilities impact your system, Intel has created a detection tool that can be downloaded from its site. The tool is available only for Windows and Linux users.

The patch meant to fix these vulnerabilities will not be provided by Intel. You will have to check with your notebook’s OEM or your PC’s motherboard maker to see if they have released a firmware update that fixes the recent flaws.

Purism told us in an email that although they need to test the new firmware, they think they should still be able to use the undocumented mode that the NSA has also been using to continue to disable the ME in their laptops, as the recent vulnerabilities announced by Intel don't seem to relate to it.

IBM, Samsung, Global Foundries, and other equipment suppliers from the IBM Research Alliance announced the development of the first 5nm chip using a gate-all-around FET structure. The news comes two years after the alliance announced the development of the world’s first 7nm chips, which should enter production next year.

Samsung recently revealed that it plans to develop a 4nm GAAFET chip, too, but we don’t know yet when it’s going to be available.

World’s First 5nm Chip

Although Moore’s Law seems to have slowed down quite a bit in the past few years, IBM and its partners were able to push it forward using new chip manufacturing technologies such as extreme ultraviolet (EUV) lithography, nanosheet transistors, and gate-all-around FET structures.

All of these combined seem to have made it possible to develop 5nm chips that are ready for mass production. Not only that, but the 5nm chips also promise a 40% increase in performance or a 75% reduction in power consumption over the 10nm process technology currently in use by Samsung. Even though we may not see such 5nm chips on the market until 2021 or later, the improvement still looks significant.

IBM also said that a 5nm chip could be made out of 30 billion transistors, a 50% increase from the 20 billion transistors the company announced for its 7nm chip two years ago.

GAAFET And EUV

According to IBM, the EUV lithography can enable continuous adjustment for the width of the nanosheet transistor architecture, all within a single manufacturing process or chip design. This permits the fine-tuning of performance and power for specific circuits, which is not possible with today’s FinFET transistor architecture that is limited by its current-carrying fin height.

Although FinFET processes can also be scaled down to 5nm, reducing the space between the fins doesn’t increase the current flow for additional performance.

“Today’s announcement continues the public-private model collaboration with IBM that is energizing SUNY-Polytechnic’s, Albany’s, and New York State’s leadership and innovation in developing next generation technologies,” said Dr. Bahgat Sammakia, Interim President, SUNY Polytechnic Institute. “We believe that enabling the first 5nm transistor is a significant milestone for the entire semiconductor industry as we continue to push beyond the limitations of our current capabilities. SUNY Poly’s partnership with IBM and Empire State Development is a perfect example of how Industry, Government and Academia can successfully collaborate and have a broad and positive impact on society,” he added.

IBM also believes that the 5nm process will enable another big leap in performance, not just for regular consumer devices, but also for next-generation machine learning chips and supercomputers. More details about its 5nm process will be revealed at the 2017 Symposia on VLSI Technology and Circuits conference in Kyoto, Japan, which is happening this week.

At the annual Samsung Foundry Forum, Samsung announced its foundry’s roadmap for the next few years, which includes an 18nm FD-SOI generation targeting low-cost IoT chips as well as 8nm, 7nm, 6nm, 5nm, and even 4nm process generations.

18nm Fully Depleted – Silicon on Insulator (FD-SOI)

Samsung will expand its 28FDS process into a broader platform that will also offer Radio Frequency (RF) and embedded Magnetic Random Access Memory (eMRAM) options to its foundry customers.

The company will also launch the 18FDS process, which is the next-generation process that targets low-cost IoT chips.

8nm Low Power Plus (8LPP)

The 8nm process generation seems to be Samsung’s last generation before the company plans to use extreme ultraviolet (EUV) lithography. The company said that the 8LPP process combines key process innovations from its 10LPP process, as well as further improvements in performance and gate density compared to 10LPP.

7nm Low Power Plus (7LPP)

The 7LPP process will be Samsung’s first generation to use EUV lithography. According to the company, EUV lithography is what will allow Moore’s Law to continue and foundries to keep shrinking transistors down to 1nm.

6nm Low Power Plus (6LPP)

The 6LPP generation will improve on 7LPP, primarily by allowing greater area scalability and making chips more efficient.

5nm Low Power Plus (5LPP)

The 5LPP process will be Samsung’s last one to use a “FinFET” structure, as that type of architecture reaches its physical limits. This generation will also focus on area reduction and lower power consumption for chips.

4nm Low Power Plus (4LPP)

The 4LPP process generation will be Samsung’s first to use a “Gate All Around FET” (GAAFET) transistor structure, with Samsung’s own implementation dubbed “Multi Bridge Channel FET” (MBCFET). The technology uses a “Nanosheet” device to overcome the physical limitations of the FinFET architecture.

Pushing Moore’s Law Forward

Even though process generations may not accurately describe how small the transistors actually are these days, all major chip fabrication companies, including Intel, Samsung, Global Foundries, and TSMC, seem to be pushing Moore’s Law to its very limits and making steady progress with each process generation.

Samsung’s transition to EUV lithography and GAA FETs in the near future should lead to faster and more efficient chips in our devices for the foreseeable future. The competition is likely not far behind, or may even be ahead, but they may also be less willing to share information years ahead of production.

At Google I/O 2017, Google announced its next-generation machine learning chip, called the “Cloud TPU.” The new TPU no longer does only inference--now it can also train neural networks.

First Gen TPU

Google created its own TPU to jump “three generations” ahead of the competition when it came to inference performance. The chip seems to have delivered, as Google published a paper last month in which it demonstrated that the TPU could be up to 30x faster than a Kepler GPU and up to 80x faster than a Haswell CPU.

The comparison wasn’t quite fair, as those chips were a little older, but more importantly, they weren’t intended for inference.

Nvidia was quick to point out that its inference-optimized Tesla P40 GPU is already twice as fast as the TPU for sub-10ms latency applications. However, the TPU was still almost twice as fast as the P40 in peak INT8 performance (90TOPS vs 48TOPS).

The P40 also achieved its performance using more than three times as much power, so this comparison wasn’t that fair, either. The bottom line is that right now it’s not easy to compare wildly different architectures to each other when it comes to machine learning tasks.

Cloud TPU Performance

In last month’s paper, Google hinted that a next-generation TPU could be significantly faster if certain modifications were made. The Cloud TPU seems to have have received some of those improvements. It’s now much faster, and it can also do floating-point computation, which means it’s suitable for training neural networks, too.

According to Google, the chip can achieve 180 teraflops of floating-point performance, which is six times more than Nvidia’s latest Tesla V100 accelerator for FP16 half-precision computation. Even when compared against Nvidia’s “Tensor Core” performance, the Cloud TPU is still 50% faster.

Google made the Cloud TPU highly scalable and noted that 64 units can be put together to form a “pod” with a total performance of 11.5 petaflops of computation for a single machine learning task.

Strangely enough, Google hasn’t given the numbers for inference performance yet, but it may reveal them in the near future. Power consumption was not revealed either, as it was for the TPU.

Cloud TPUs For Everyone

Up until now, Google has kept its TPUs to itself, likely because it was still experimental technology and the company wanted to first see how it fared in the real world. However, the company will now make the Cloud TPUs available to all of its Google Compute Engine customers. Customers will be able to mix and match Cloud TPUs with Intel CPUs, Nvidia GPUs, and the rest of its hardware infrastructure to optimize their own machine learning solutions.

It almost goes without saying that the Cloud TPUs support the TensorFlow machine learning software library, which Google open sourced in 2015.

Google will also donate access to 1,000 Cloud TPUs to top researchers under the TensorFlow Research Cloud program to see what people do with them.

Update, 5/18/17, 7:52am PT: Fixed typo.

Samsung and Intel recently filed two amicus briefs in the FTC’s antitrust case against Qualcomm, accusing Qualcomm of violating Fair, Reasonable And Non-Discriminatory (FRAND) patent terms and legally excluding competitors from the market through the use of its own baseband processor patents. Samsung also claimed that Qualcomm stopped it from selling its chips to third-party vendors. Qualcomm has previously called this allegation “false.”

Why FRAND Patents Exist

Without FRAND patents, certain standards may not exist, simply because the other members of a standards group would not agree to adhere to a given standard without all members sharing the patents with each other in a reasonable way.

Let’s consider, for example, a situation where multiple companies within a standards body each try to create a next-generation wireless technology. If one company creates a technology called LTE-X, but doesn’t want to share its technology with the rest of the members from the standards body, then that technology would likely not end up being chosen as the standard by the standards body.

If another company creates a different technology called LTE-Z but is willing to share the patents for free or for a small price, the standards body would likely go with LTE-Z as the “next-generation wireless technology.” It would then become much harder for the company behind LTE-X to promote its technology to carriers, smartphone makers, and so on, because everyone else has already agreed on a different standard.

Now, let’s consider one company has the majority of the patents on the LTE-Z technology and at first agrees to license them out to competitors for a reasonable price. Thus, LTE-Z becomes the new standard, and everyone adopts it. However, later on, this company decides that it doesn’t want to license the technology to competitors anymore, or it does do it but for a prohibitive price, effectively stopping competitors from being able to legally use it.

This is basically what Samsung and Intel are accusing Qualcomm of doing.

Qualcomm’s “Handset Tax”

According to the FTC, Qualcomm has been imposing a “handset tax” on manufacturers that disincentivizes them from seeking alternative chip suppliers. Even if the manufacturers would use another company’s chip, they’d still have to pay Qualcomm an “onerous” sum based on a percentage of a device’s retail price, according to Samsung’s amicus brief.

Intel also argued in its own brief that it wasn’t technology prowess that helped Qualcomm stay ahead, but these types of contracts it’s been imposing on manufacturers. The contracts allegedly prevented the smartphone makers from being competitive on price when buying chips from Qualcomm’s competitors.

Qualcomm has seemingly argued before that if so many of its customers were harmed by its policies, then why hasn’t anyone sued it yet? Samsung pointed out that Broadcom did sue it over abusing FRAND patents in 2006, but the two companies eventually ended up settling, instead of concluding the trial.

Samsung also noted that Qualcomm’s customers, including itself, are also disincentivized from suing Qualcomm, because they could forever lose access to its chips, too. Samsung admitted that it has been forced to buy Qualcomm’s chips, even if it believed Qualcomm was overcharging for its FRAND patents.

Samsung asserted that Qualcomm’s policies and overcharging for FRAND patents have made it all but impossible for competing modem makers to compete in the premium handset market. This includes Intel which blamed Qualcomm for being unsuccessful in the mobile market, but also Nvidia, which had to shut down its Icera modem business a few years ago because of Qualcomm’s "unlawful abuse of dominance," as the company claimed at the time.

Recently, Apple also sued Qualcomm arguing that Qualcomm is abusing its market position and charging it five times more than what all the other licensors charge it combined. In response, Qualcomm counter-sued Apple over breach of contract and for mischaracterizing the agreements between the two companies to the FTC. Qualcomm also said Apple lowered the performance of its modem in the iPhone 7 to reduce the discrepancy in performance between its chip and Intel's modem, which was available in certain iPhone 7 models.

Qualcomm's Ongoing Antitrust Issues

Over the past few years, Qualcomm has been hit with antitrust lawsuits from Japan, China, South Korea, the European Union, and recently the United States. The company continues to claim that the accusations are flawed, even after it was already forced to pay almost a billion dollar on three different occasions (in settlement with Broadcom, as well as China, and South Korea’s cases).

Tom’s Hardware asked Qualcomm why, if these accusations are baseless, so many countries seem to be starting antitrust cases against it. The company refused to provide a direct answer to that, reiterating that its stance on these issues has not changed.

Qualcomm introduced two new Snapdragon platforms for its “High Tier” with the Snapdragon 660 and the Snapdragon 630. The chips bring higher performance compared to the previous generation, as well as improved photography, support for machine learning software, and more.

Higher Performance

Compared to the company’s previous Snapdragon 653, the new Snapdragon 660 has a Kyro 260 processor that's 20% faster and an Adreno 512 GPU that's 30% faster.

The Kryo 260 CPU is made up of two clusters of four CPU cores each. The CPUs in one cluster can go up to 2.2GHz, whereas the cores in the other cluster hit up to 1.8GHz.

The Snapdragon 630 promises a 10% increase in performance over its predecessor (Snapdragon 625) and also a 30% increase in graphics performance. However, unlike the 660 that uses Kryo cores, the 630 comes with Cortex-A53 cores, which are beginning to look a little dated, even at 2.2GHz clock speeds.

The Snapdragon 660 supports resolutions up to 2560x1440, whereas the Snapdragon 630 supports resolutions up to only 1080p. Both platforms are built on a 14nm FinFET process, support up to 8GB of RAM, and sport Vulkan (the low-level graphics API).

Camera Support Improvements

Qualcomm’s Spectra 160 image signal processor (ISP) supports improved image quality with more natural skin tones, better low-light photography, and higher battery efficiency. Additionally, the new ISP comes with improved support for dual cameras as well as features such as smooth optical zoom, bokeh effects, dual pixel autofocus, and camcorder stabilization.

Qualcomm’s new electronic image stabilization will take advantage of the gyroscope to correct the pitch, yaw, and roll. The technology will likely be similar to what Google is using in its Pixel phone in order to avoid having to use the bulkier Optical Image Stabilization (OIS).

Connectivity

The new Snapdragon 660 and 630 mobile platforms include a Qualcomm X12 modem that brings peak downlink data rates of up to 600Mbps--a first for the 600-series chips. The Snapdragon 660’s 802.11ac Wi-Fi chip offers twice as much throughput and 60% lower download power consumption compared to the Snapdragon 652.

Both of the new chips support Bluetooth 5.0, which has four times the range and twice the transmission speed of the previous Bluetooth 4.2 technology.

Quick Charge 4

Qualcomm’s latest “Quick Charge 4” fast charging technology can charge up to 50% of a smartphone’s battery within 15 minutes, or give five hours of talk time with only five minutes of charge. Quick Charge 4 is USB-PD and USB Type-C compliant.

Security

The Qualcomm Mobile Security solution provides hardware-based protection, user authentication, and device attestation on the mobile device.

Machine Learning

Qualcomm seems to want to take advantage of  the growing trend of on-device machine learning by supporting the open source TensorFlow, Caffe, and Caffe2 machine learning software frameworks. The company is also offering OEMs and developers access to its own Snapdragon Neural Processing Engine software development kit (SDK).

All of these software tools will be supported by the CPU, GPU, and Qualcomm’s Hexagon digital signal processor (DSP), as well as its HVX vector extensions.

Qualcomm said that over 1,000 Snapdragon 600 designs have already launched or are in the pipeline. The Snapdragon 660 is already shipping, whereas the 630 will ship towards the end of this month.

“With the introduction of the Snapdragon 660 and 630 Mobile Platforms, we are thrilled that features such as improved image quality and fast LTE speeds will now be available in a wide array of devices without sacrificing performance or quality,” said Kedar Kondap, vice president, product management, Qualcomm Technologies, Inc. “This ensures that a greater number of consumers will be able to take advantage of higher quality user experiences in camera, audio and visual processing, connectivity, improved CPU and GPU performance, fast charging, security and machine learning,” he added.

ARM announced the first Image Signal Processor (ISP) for the automotive market with Ultra Wide Dynamic Range (Ultra WDR) capabilities, allowing autonomous driving systems to work in even the most difficult daylight lighting conditions.

ARM To Focus On Computer Vision

Last year, ARM purchased Apical, a company that was licensing imaging and embedded computer vision intellectual property. ARM seems to have realized that computer vision is an important part of the future of embedded chips, considering how many types of products could benefit from it: self-driving cars, drones, robots, surveillance cameras, and any other product that has a camera and needs to analyze data.

"Computer vision is in the early stages of development and the world of devices powered by this exciting technology can only grow from here," said Simon Segars, ARM’s CEO, when the company acquired Apical. "Apical is at the forefront of embedded computer vision technology, building on its leadership in imaging products that already enable intelligent devices to deliver amazing new user experiences. The ARM partnership is solving the technical challenges of next generation products such as driverless cars and sophisticated security systems. These solutions rely on the creation of dedicated image computing solutions and Apical's technologies will play a crucial role in their delivery," he added.

ARM wants to focus on the self-driving car market in particular. According to Strategy Analytics, a market research firm, it’s expected that mid-range cars, such as the Volkswagen Golf, will have at least three cameras, and and luxury cars should be approaching ten cameras by 2023.

Not all cameras will be external; some will be internal in order to provide certain features, such as monitoring to see if the driver falls asleep while the car is in cruise control. This could benefit cars that haven’t yet reached Level 4 or higher for autonomous driving, and they still require a human to be alert in case something goes wrong.

Cameras on a car could also offer better ways to detect pedestrians or provide night vision for the driver, thus enhancing regular cars that don’t yet have self-driving systems.

Mali-C71 Ultra Wide Dynamic Range ISP

The first product to come from the Apical acquisition is the Mali-C71, a high-performance ISP that’s capable of 24 stops of dynamic range. In photography, a stop is a doubling of light exposure for an image, so the bigger the range of stops, the brighter you can make an image or part of an image. That means that an advanced driver assistance system (ADAS), for instance, can more easily distinguish what’s in the shadows in the daylight.

The Mali-C71 can also reduce light exposure to more easily observe something that has too much light on it, as seen in the example below.

ARM’s ISP can support both human displays computer vision simultaneously. This means humans (that's you!) can watch what the cameras see while the computer vision system does its work in the background to analyze every pixel at the same time.

The ISP has a processing performance of 1.2 Gigapixels per second and can support up to four different cameras at the same time, each with up to 4K resolution. Manufacturers that want to put more than four cameras on a car can opt for multiple ISPs.

ARM said that the Mali-C71 ISP has been designed with the highest levels of safety in mind and is in accordance with the Automotive Safety Integrity Level D (ASIL D). ASIL D certification is required for components where the risk of fatal injury is highest, which means the highest level of assurance is also necessary.

According to the company, the ISP has 300 fault detection circuits, a built-in continuous self-test, Cyclic Redundancy Checking (CRC) on data paths, and every pixel is tagged for reliability. The software for the ISP is also being developed with ASIL compliance in mind.

Competition

ARM will be entering a market where the competition is already quite strong. Nvidia has made significant progress in the automotive market over the past few years, and Intel recently acquired Mobileye, the company that used to make Tesla’s "Autopilot" systems.

The market is still young, however, and ARM will likely also be an important player by licensing its IP to other chip makers. Then, it could gain market share in the same way it has in smartphones, networking equipment, and other embedded chip markets.

Samsung announced that the second-generation 10nm FinFET process, called “Low Power Plus,” has been qualified and is ready for production.

Small, But Important Improvements

The LPP generation is an iteration of the existing 10nm Low Power Early (LPE) process, which has been used by chips such as the Qualcomm’s Snapdragon 835 and Centriq 2400, as well as Samsung’s own Exynos 8895.

The new process generation brings improvements of either a 10% increase in performance, or a 15% reduction in power consumption. The improvement isn’t a large one, but considering how much Moore’s Law has slowed down lately, a 10% increase in performance year-over-year isn’t too bad, either.

“With our successful 10LPE production experience, we have commenced production of the 10LPP to maintain our leadership in the advanced-node foundry market,” said Ryan Lee, Vice President of Foundry Marketing at Samsung Electronics. “10LPP will be one of our key process offerings for high performance mobile, computing and network applications, and Samsung will continue to offer the most advanced logic process technology,” he added.

Future Generations

After the 10LPP node, Samsung will adopt another 10nm process generation called 10LPU. The company previously announced it at the same time as it announced the fourth-generation 14nm process, 14LPU. However, despite the similar names, the 10LPU generation will be die-shrink and cost-saving process, similar to the 14LPC node, whereas 14LPU will be a high-performance process technology. It’s unclear whether another 10nm generation will follow after this, in parallel with the company’s 7nm generation.

Speaking of 7nm, Samsung still plans on using extreme ultraviolet (EUV) lithography for its 7nm process. That would be a first in the industry and quite the breakthrough, because it would allow Samsung and other chip makers to manufacture 7nm and smaller chips more easily. The EUV lithography technology has been studied since the '90s, but chip makers agree it should finally arrive in the next few years.

Nvidia announced that Baidu, the biggest search engine in China, now uses the company’s deep learning software platform as well as the latest Pascal-based Tesla P40 GPUs for both training and inference.

Baidu Modernizes Cloud Platform

Baidu, like other companies interested in machine learning, has been using Nvidia’s GPUs for many years. However, the company seems to be looking to get the latest Pascal-based GPUs to take advantage of their increased performance and efficiency.

“Our partnership with NVIDIA has long provided Baidu with a competitive advantage,” said Shiming Yin, vice president and general manager of Baidu Cloud Computing. "Baidu Cloud Service powered by NVIDIA's deep learning software and Pascal GPUs will help our customers accelerate their deep learning training and inference, resulting in faster time to market for a new generation of intelligent products and applications,” he added.

As a search engine and cloud services company, Baidu seems to have realized that machine learning is critical for the company’s success. Therefore, it's continuing to invest big in staying up to date with hardware platforms, as well as in machine learning research to improve its services.

Tesla P40

Baidu bought some Tesla P40 “inference GPUs,” but Nvidia has said before that they can also be used for training neural networks. The P40 is Nvidia’s highest-performing inference chip right now, due to its support for 8-bit integer computation.

Nvidia recently pitted the Tesla P40 against Google’s own inference chip, the Tensor Processing Unit (TPU). The GPU did well against the TPU in at least one metric: inference with sub-10ms latency.

Considering that one of the reasons why the “AI on the edge” trend is growing now is because machine learning over the internet has too high latency, this can be quite an important metric. However, as discussed previously, it’s not the only metric that matters, as performance/Watt can often be the most important metric in data centers.

However, Nvidia has many customers with many different needs, and for applications such as natural language processing, traffic analysis, intelligent customer service, personalized recommendations, and understanding video. This is also why even Google continues to use GPUs where they make the most sense. Nvidia also numbers customers from other major cloud services companies such as Amazon, Microsoft, IBM, and Alibaba, who all use machine learning to enhance their services in different ways.