The MXM (Mobile PCI Express Module) form factor is predominantly utilized in laptops and mobile workstations. Consequently, the graphics cards are generally the mobile variants of their desktop counterparts, which results in diminished performance levels. The MXM RTX 5000 Ada version possesses 24% fewer CUDA cores and half the memory capacity of the desktop RTX 5000 Ada card.

Compared to the RTX 5000 Ada, which has a 250W TDP, the mobile version operates at 120W, thereby facilitating easier thermal management within a laptop. Consequently, the MXM RTX 5000 Ada derives power from an MXM 3.1 Type-B interface, which supports up to 200W.

Zotac also offers a separate MXM to PCIe x16 adapter, which sells for $181, to transform the MXM RTX 5000 Ada into a desktop graphics card. The adapter utilizes the 16-pin (12VHPWR) power connector, which is common on Nvidia's latest higher-end graphics cards, such as the GeForce RTX 5090. The adapter also provides four DisplayPort outputs for connecting displays.

Zotac's MXM RTX 5000 Ada is priced at approximately $4,743.66, which is quite high considering the RTX 5000 Ada itself is not sold for that amount. For example, in the U.S., the desktop RTX 5000 Ada is available for $4,124. By choosing the MXM RTX 5000 Ada, you're paying 15% extra for a less powerful version.

There may be interest in the MXM RTX 5000 Ada for professional or workstation users looking to upgrade their laptops that accept the MXM form factor. Additionally, the MXM to PCIe x16 adapter is a useful accessory for users who want to swap the MXM RTX 5000 Ada between their laptops and desktops. For most users, however, choosing a desktop RTX 5000 Ada from the start is the more cost-effective and higher-performance option.

In desperate times, however, a GPU is a GPU, and it's interesting to see Zotac breathe life into silicon that might otherwise sit unused.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

The YouTuber expanded their testing suite with more modern titles, featuring: Kingdom Come Deliverance II, Space Marine II, Alan Wake 2, Cyberpunk 2077, CS 2, and GTA V Enhanced.

The GT 1030 delivered a measly 8 FPS average at native 1080p resolution in Kingdom Come Deliverance II. Enabling FSR 3.1 performance mode significantly boosted performance to 15 FPS average, and moving to a 720p render resolution with the same settings yielded an even greater 21 FPS average. Despite the gains, performance was still well below playable FPS.

Space Marine II was even worse, averaging just 6 FPS on the GT 1030 at 1080p native resolution with low settings. FSR performance mode upscaling more than doubled performance to 13 FPS, but the game was still far from playable.

Alan Wake 2 was the worst-performing game of the bunch, performing in seconds per frame rather than frames per second at native resolution. Native 1080p performance was below 0 FPS average, and FSR ultra performance mode upscaling boosts performance to 3 FPS average.

Cyberpunk 2077 at 1080p native low settings achieved 12 FPS average on the GT 1030. FSR 2.1 ultra performance mode boosted FPS to 21 FPS average, and FSR 3 boosted FPS to 18 FPS average. XeSS had the opposite effect, achieving just 8 FPS average in performance mode.

CS 2, boasted similar issues on the GT 1030 as on the GTX 660, which the YouTuber reviewed two weeks ago. Native performance was decent; 100 FPS average at low settings, but FSR upscaling did not provide any noteworthy performance improvements beyond native rendering.

GTA V Enhanced ran surprisingly well despite overflowing the GT 1030 GDDR5's 2GB memory capacity. The game ran at 30-40 FPS average while consuming 3 GB of memory (video memory overflows to system RAM when full). FSR 3 performance mode boosted performance to 45 FPS, and FSR 1 performance mode boosted performance to 50 FPS average.

Similar to GCS Hardware's GTX 660 testing, Nvidia's Pascal-based GT 1030 is not powerful enough to play the modern AAA games at playable FPS with upscaling, except for older titles featuring upscaling support.

Still, the GT 1030's performance is a significant improvement from the GTX 660. The Pascal-based GPU was technically capable of performing significantly better with upscaling, in most of the aforementioned titles. The GTX 660, by contrast, did not benefit from upscaling at all in any game GCS Hardware threw at it.

Nvidia's newer Pascal architecture contributes to the GT 1030's ability to take advantage of FSR upscaling better than the GTX 660. FSR is designed to use FP16 calculations to process the necessary data to upscale an image from a lower resolution to a higher resolution with as few artifacts as possible.

Pascal was the first Nvidia architecture to natively support FP16 compute. By contrast, the GTX 660 was based on the Kepler architecture, which only features FP32 and FP64 compute support. FSR has an FP32 fallback mode for these types of GPUs, but using FP32 compute reduces performance.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

The reason this GPU is now appearing on eBay is that it can no longer be sold in its intended market. The RTX 5090D was created specifically for China after U.S. restrictions barred Nvidia from selling the original RTX 5090 there. To get around those rules, the 5090D was introduced with the same gaming performance but reduced AI capabilities. It worked... until policy changed again, and even the 5090D now seems to have been swept up in the export ban, but the 5090D V2 is coming to fix that.

As a result, units that had already been produced and distributed to partners are now being offloaded online, with the Galax HOF XOC arguably the most interesting among them. Originally designed for extreme overclockers, it features a 12-layer PCB, 38-phase power delivery system, and two—that's right, two—16-pin power connectors. The card technically ships with the same 600W power limit that Nvidia has set for all variants of the 5090, but Galax provides an alternative BIOS through unofficial channels that unlocks up to 2,000W of headroom for serious competitive use, and it's already broken world records.

Originally spotted by PC Games Hardware, the GPU is now appearing in bulk on eBay, often with volume discounts for multiple purchases. Prices are around €4,739 for a single unit or roughly $5,500 USD, with discounts if you buy four or more. That’s more than double the MSRP of the standard RTX 5090 Founders Edition, yet the Galax 5090D HOF is arguably in a different class altogether, so it almost justifies its price. We know; we can't believe we just said that either.

Aside from Asus's gold-ridden 5090s that only "improve" aesthetics, Galax's variant is a different beast altogether that genuinely unlocks more performance. The card has already made appearances in global overclocking leaderboards, holding its own against other monsters like ASUS’s ROG Astral RTX 5090 OC, though Galax holds the edge thanks to its dual 12V-2×6 power setup and BIOS tuning options that no other card has. What adds an extra layer of exclusivity is that Galax has never released a Hall of Fame variant of the standard RTX 5090. This design is unique to the 5090D model, making it something of a collector’s piece even beyond its raw specs.

Ultimately, the Galax RTX 5090D HOF OC Lab XOC represents a strange intersection of geopolitics, hardware excess, and enthusiast obsession. It wasn’t supposed to leave China, and it wasn’t supposed to be available to just anyone. But now that it’s here, floating on eBay with a price tag that makes even high-end builds look modest, it’s become one of the most over-the-top GPU purchases money can buy. And this isn't even the only exorbitant 5090 Galax makes, as their 5090D Luna has a magnetic shroud that separates the fans from the cooler and PCB without a screwdriver.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

Windows 7 (2009) and the Steam Deck (2022) are divided by an approximate 13-year gulf, so it isn’t surprising that they aren’t the most comfortable bedfellows. Immediate evidence of technical difficulties in this marriage is evident in Pony’s screenshot – the Windows 7 display is stuck in an inappropriate portrait mode.

Pony reveals some more details about the Windows 7-on-Steam Deck feat in the ensuing discussion. Apparently, there are few, if any, issues running this OS on the CPU portion of AMD’s Van Gogh APU. However, “the drivers won't install under Windows 7 as its integrated graphics is basically custom-made and drivers for it only work under 10/11,” the equine OS enthusiast explains.

Other replies in the thread ask for Windows XP and even the peculiar by any metric Temple OS to be tested out next. But we think any further steps down the Windows OS ladder will take some time. It is hardly satisfactory if Windows 7 can’t even be tweaked into landscape mode yet…

Pony shared a video demonstrating Windows 8.1 on the Steam Deck two years ago. The Metro-skinned OS ran pretty well, with full touch functionality. If this cadence is to be maintained, any Vista on Steam Deck achievement won’t be shared until 2027.

Windows 10 will be retro soon

Both Windows 7 and Windows 8.1 are firmly in the retro computing niche in 2025, and are soon to be joined by Windows 10 – bar a few stalwarts who feel compelled to pay an Extended Security Updates (ESU) fee instead of just shifting up to Windows 11 or finally contribute to making 2025 the breakthrough year for Linux on desktop.

Hopefully, SteamOS will officially come to desktops at the same time as Windows 10 says its farewells. It would be a canny move by Valve to introduce it a little earlier.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

This announcement comes on the heels of the United States Senate’s open letter [PDF] to the CEO, asking him to “refrain from meeting with representatives of any companies that are working with the PRC’s military or intelligence establishment…or are suspected to have engaged in activities that undermine U.S. export controls.”

Washington is concerned that the PLA might utilize this to develop advanced weapons systems, intelligence systems, and more, prompting a bipartisan effort to deny China access to the most powerful hardware over the past three administrations. However, Huang has often publicly said that the U.S. strategy of limiting China’s access to advanced technologies was a failure and that it should lead the global development and deployment of AI.

“…Depriving someone of technology is not a goal, it’s a tactic — and that tactic was not in service of the goal,” said the Nvidia CEO during the interview. “Just like we want the world to be built on the American dollar, using the American dollar as the global standard, we want the American tech stack to be the global standard.” He also added, “In order for America to have AI leadership, it needs to make sure the American tech stack is available to markets all over the world, so that amazing developers, including the ones in China, are able to build on American tech stack so that AI runs best on the American tech stack.”

When Zakaria asked him about the Chinese PLA’s use of this tech, Jensen said that it’s not going to be an issue. “The Chinese military [is] no different [from] the American military: [they] will not seek each other’s technology to be built on top [of each other]. They simply can’t rely on it — it could be, of course, limited at any time,” Jensen answered. “Not to mention, there’s plenty of computing capacity in China already. If you just think about the number of supercomputers that are in China, built by amazing Chinese engineers, that are already in operation — they don’t need Nvidia’s chips or American tech stacks in order to build their military.”

Despite what he says, China has been heavily investing in AI data centers, and companies within the country are explicitly sourcing Nvidia’s most powerful products, even the ones that the U.S. has put export controls on. And even though Huang said that there’s no evidence of AI chip diversion, it’s widely reported that you can easily get these banned chips in the Chinese black market, with one businessman even showing off his smuggled GPUs on social media.

Chinese operators of these smuggled AI chips would have a harder time getting firmware updates and likely won’t have access to Nvidia’s advanced cloud tools and enterprise platforms. However, because Nvidia still sells export-compliant GPUs to China, the platform and cloud software can still potentially work with the banned higher-power equipment.

Aside from that, it will probably be difficult for the U.S. to disable these AI GPUs remotely, if it comes to that. After all, Nvidia would have a harder time selling its chips if a way to disable them remotely exists — that’s why the U.S. has a bill in the works that could force geo-tracking tech on high-end hardware. Even if there is such a technology, China can just air gap the systems to prevent them from being remotely killed.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

Donkey Kong Bananza journeyed down quite an interesting path to the Switch 2, which could partly explain the performance wrinkles. This smashing new 3D platformer and Switch 2 exclusive was originally destined for the Switch. However, this first 3D Donkey Kong outing since Donkey Kong 64 goes all-in on destructible environments to deliver its gameplay thrills.

Destructible environments in this game are extensive, in a kind of sandbox exploration way, where DK can smash his hulking fists through walls, can dig tunnels, and tear off environmental bits and bats for weapon or platforming tool use. It sounds like explosive fun, indeed, but the 3D voxel engine behind the environment is demanding. That’s why the game was pulled back to be prepared for the new Switch 2.

A La Vanguardia reporter asked Takahashi about their experience of “frame drops at certain times and levels,” in what is expected to be the shipping version of Donkey Kong Bananza. “There are several factors to consider. First of all, we intentionally used effects such as hit stop or slow motion to emphasize the impacts. On the other hand, when using voxel technology, there are times when there are big changes and destruction in the scenery,” said the game’s director. “We are aware that in these moments the performance may drop a bit. However, as you say, overall the game can be enjoyed smoothly, and at points where large-scale changes occur, we have prioritized fun and gameplay.” (DeepL, machine translation)

Clearly, there may be a few stutters experienced when the game engine begins to reach its boundaries on the Switch 2 hardware. With Nintendo being aware of this potential issue for gamers, and the always-connected nature of today’s consoles, we hope optimizations will eventually arrive to iron out any startling wrinkles. Or they could just turn up the DLSS knob.

For clues regarding the faltering performance of the Switch 2 in its maiden year, check out our in-depth Switch 2 review and hardware analysis from June.

We note that 3D environment voxel engine games were seen on PC as early as 1992, with Comanche: Maximum Overkill. That game’s proprietary Voxel Space engine was apparently written entirely in assembly language for the best performance.

Nintendo’s unruly new 3D platformer is due to launch this Thursday (July 17).

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

The team behind the discovery, from the University of Toronto, showed how the attack could drop an AI model’s accuracy from 80% to under 1%—just by flipping a single bit in memory. It’s not just theoretical either, as they ran it on a real NVIDIA RTX A6000, using a technique that repeatedly hammers memory cells until one nearby flips, messing with whatever’s stored there.

What exactly is GPUHammer?

GPUHammer is a GPU-focused version of a known hardware issue called Rowhammer. It’s been around for a while in the world of CPUs and RAM. Basically, modern memory chips are so tightly packed that repeatedly reading or writing one row can cause electrical interference that flips bits in nearby rows. That flipped bit could be anything—a number, a command, or part of a neural network’s weight—and that’s where things go wrong.

Until now, this was mostly a concern for DDR4 system memory, but GPUHammer proves it can happen on GDDR6 VRAM too, which is what powers many modern NVIDIA cards, especially in AI and workstation workloads. This is a serious cause for concern, at least in specific situations. The researchers showed that even with some safeguards in place, they could cause multiple bit flips across several memory banks. In one case, this completely broke a trained AI model, making it essentially useless. The scary part is that it doesn’t require access to your data. The attacker just needs to share the same GPU in a cloud environment or server, and they could potentially interfere with your workload however they want.

As mentioned, the attack was tested on an RTX A6000, but the risk applies to a wide range of Ampere, Ada, Hopper, and Turing GPUs, especially those used in workstations and servers. NVIDIA has published a full list of affected models and recommends ECC for most of them. That said, newer GPUs like the RTX 5090 and H100 have built-in ECC directly on the chip, which handles this automatically—no user setup required.

However, if you're someone just sitting at home worried about their personal setup, this isn’t the kind of attack you’d see targeting individual gamers or home PCs. It’s more relevant to shared GPU environments like cloud gaming servers, AI training clusters, or VDI setups where multiple users run workloads on the same hardware. That being said, the core idea that memory on a GPU can be tampered with silently is something the entire industry needs to take seriously, especially as more games, apps, and services start leaning on AI.

Nvidia's Response

NVIDIA has responded with a simple but important recommendation: turn on ECC (Error Correction Code) if your GPU supports it. ECC is a feature that adds redundancy to memory so it can detect and fix errors like these bit flips. Keep in mind, enabling ECC does come with a small trade-off—around 10% slower performance for machine learning tasks, and about 6–6.5% less usable VRAM. But for serious AI work with peace of mind, that’s worth it.

You can enable it using Nvidia's command-line tool:

nvidia-smi -e 1

You can also check if ECC is active with:

nvidia-smi -q | grep ECC

Attacks like GPUHammer don’t just crash systems or cause glitches. They tamper with the integrity of AI itself, affecting how models behave or make decisions. And because it all happens at the hardware level, these changes are nearly invisible unless you know exactly what to look for. In regulated industries like healthcare, finance, or autonomous driving, that could cause serious problems—wrong decisions, security failures, even legal consequences. Even though the average user isn’t directly at risk, GPUHammer is a wake-up call. As GPUs continue to evolve beyond gaming into AI, creative work, and productivity, so do the risks. Memory safety, even on a GPU, is no longer optional.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

These line up closely with the current ROG Ally range in Europe, where pricing is currently between €500 and €700, depending on the chip variant, while the ROG Ally X retails for around €900. It is important to note that the final retail price isn’t always a direct conversion across regions, which means U.S. pricing will likely be different, as manufacturers often need to adjust for taxes, import costs, and market response. Asus hasn’t officially confirmed any details yet, so these early listings are more or less placeholders.

Announced early last month, the ROG Xbox Ally and ROG Xbox Ally X handheld consoles mark the first official collaboration between Asus and Microsoft to launch Xbox-branded hardware. Both models feature a redesigned chassis with contoured grips similar to the Xbox controller, along with impulse triggers for added haptic feedback. According to Asus, the addition of Xbox controller ergonomics makes them the most comfortable handhelds yet, despite being slightly thicker and heavier than earlier models.

The ROG Xbox Ally X is the high-end variant, powered by AMD’s new Ryzen AI Z2 Extreme processor with 8 cores, 16 RDNA 3.5 graphics cores, and 24GB of LPDDR5X-8000 memory. It also packs a 1TB SSD, Wi-Fi 6E, Bluetooth 5.4, USB 4 ports, and an 80Wh battery. Meanwhile, the standard Xbox Ally is a more affordable model, equipped with a slower Ryzen Z2 A chip, 16GB LPDDR5X-6400 RAM, a 512GB SSD, and a smaller 60Wh battery. Both devices come with the same 7-inch full-HD 120Hz VRR IPS display from the Ally X and include a new dedicated Xbox button for quick access to the Game Bar.

Running on Windows 11, both handhelds are expected to boot directly into an Xbox-style interface to deliver a more streamlined, console-like experience. Users will still have full access to Windows, making it possible to install third-party launchers like Steam, Epic Games, and others.

Speaking of which, Microsoft recently announced in a blog post that it’s testing a new version of the Xbox app for Windows with support for aggregated game libraries. This update will be rolled out to Windows users as a preview and will eventually roll out to the ROG Xbox Ally and Ally X later this year. The updated app will let users view installed titles from supported third-party storefronts such as Battle.net, EA, and Riot Games, with support for more stores like Steam and GOG likely to arrive at a later stage.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

This premium ultrabook challenges conventional ideas about laptops with excellent software to manage its dual displays and well-designed accessories to make the most of them, including a wireless keyboard, mouse, stand, and pen. If you can live with its brief battery life, the Yoga Book is a compelling option, particularly for users considering a portable monitor.

Design of the Lenovo Yoga Book 9i Gen 10

The Yoga Book’s dual-display design is undeniably futuristic. This device lives up to its price, delivering exceptional build quality with a rigid, flex-free chassis and lid. Every part of its aluminum shell fits together seamlessly, with Lenovo’s reflective lid logo as its sole embellishment. The webcam notch above the display serves as a convenient grip, enabling effortless one-handed opening in a thoughtful design touch.

The included stand unlocks the Yoga Book’s productivity potential by using both displays. The laptop can be positioned with one screen on top of the other or opened like a book. The keyboard magnetically attaches to the stand in either configuration.

The stand truly highlights the productivity potential of this platform. While portable USB displays exist, this setup stands out with its perfectly matched, high-quality screens and the invaluable ability to use them vertically. I found the vertical orientation especially useful for side-by-side document comparisons. Additionally, this configuration is more compact than carrying both a laptop and an external monitor. I couldn’t be more impressed.

The innovation continues with Lenovo's thoughtful design for easy transport. The stand wraps around the keyboard and magnetically attaches to the top of the Yoga Book, while the pen conveniently fits into a loop on the stand.

The Yoga Book’s last usage mode is as a tablet, accomplished by opening the screen 360 degrees. While still heavy for a tablet, the Yoga Book is much more convincing than a typical 2-in-1 convertible since it doesn’t leave an exposed keyboard under your fingertips.

The Yoga Book is as sleek as any 14-inch ultrabook, measuring 12.4 x 8.35 x 0.63 inches and weighing 2.69 pounds. Its proportions closely match the HP OmniBook Ultra Flip 14 (12.35 x 8.51 x 0.59 inches, 2.97 pounds), while it’s slightly larger than the Apple MacBook Air 13-inch (11.97 x 8.46 x 0.44 inches, 2.7 pounds). (Check out our review of the 15-inch model.)

Unfortunately, the Yoga’s port selection is minimal – there are just three Thunderbolt 4 (USB Type-C) ports, one on the left side of the device and two on the right. There is no headphone jack. The power adapter plugs into any of the USB-C ports.

Lenovo Yoga Book 9i Gen 10 Specifications

Productivity Performance on the Lenovo Yoga Book 9i Gen 10

We tested the Yoga Book 9i Gen 10 with an Intel Core Ultra 7 255H processor, Intel Arc 140T integrated graphics, 16GB of RAM, and a 1TB SSD.

Our comparison systems include the Apple MacBook Air (M4, 2025; $1,399.99 as tested), the HP OmniBook Ultra Flip 14 (Core Ultra 7 258V; $1,899.99 as tested), and the 2024 version of the Microsoft Surface Pro (Snapdragon X Elite X1E-80-100; $1,949.97 as tested), representing clamshells, convertible 2-in-1s, and tablets, respectively.

In the Geekbench 6 CPU test, the Yoga Book posted a single-core score of 2,886, trailing the MacBook Air (3,780) but edging out other competitors. In multi-core performance, however, its Core H-class CPU stood out, achieving 15,614 points, surpassing both the M4-powered MacBook Air (14,924 points) and the Snapdragon-powered Surface Pro (14,432 points), while leaving the OmniBook and its Lunar Lake CPU (10,877 points) far behind.

The Yoga Book didn't stand out in our 25GB file transfer test, achieving 1,063.7 MBps. Only the Surface Pro was slower (1,001.16 MBps), with the MacBook Air (1,220.35 MBps) and the OmniBook (1,702.45 MBps) leading.

The Yoga Book’s strong CPU performance followed into Handbrake, transcoding a 4K video from 4K to 1080p. The laptop finished the task in 4 minutes and 38 seconds. The MacBook Air was the next fastest (4 minutes and 52 seconds) while the OmniBook trailed (6 minutes and 37 seconds). (Note we use the Arm-specific version of Handbrake for Arm computers, namely the MacBook Air and the Surface Pro.)

To measure performance in long-running, intensive apps, we run 10 loops of the Cinebench 2024 CPU benchmark. The straighter the line is in this chart, the more consistent the scores are, indicating more stable thermal performance.

The Yoga Book initially scored 655.2 points, but its performance declined by approximately 100 points over the next two runs before stabilizing with consistent results thereafter.

Displays on the Lenovo Yoga Book 9i Gen 10

The Yoga Book features two 14-inch OLED screens, each sporting a 2880 x 1800 resolution and touch and active pen support. The excellent contrast and vivid colors of this panel technology create a superb picture – everything looks fantastic. Watching Ted Lasso, the lush green soccer fields and colorful uniforms appeared to jump off the screen, and even the office applications I used looked more engaging than usual. The screens are also very bright for OLED panels – I used them at seven or eight-tenths brightness in darker environments.

The measurements above correspond to the Yoga Book’s primary screen, though tests on the secondary display yielded practically identical results. With a 132.9% coverage of the DCI-P3 gamut, it delivers the most vibrant colors of any machine here, surpassing the Surface Pro (115.2%) and significantly outshining the non-OLED OmniBook (84.3%) and MacBook Air (83.5%). Its OLED panel also achieves impressive brightness, averaging 470.4 nits. While it falls short of the Surface Pro’s 564 nits, it matches the MacBook Air (475.6 nits) and leaves the OmniBook (359.6 nits) far behind.

Keyboard and Touchpad on the Lenovo Yoga Book 9i Gen 10

Despite having two screens, the Yoga Book still accommodates a keyboard. Lenovo provides a physical keyboard that docks with the device as well as a virtual keyboard that appears on the lower screen when you tap eight fingers. This virtual setup includes a “palm rest” and a virtual touchpad for added convenience.

Though it can't fully replace a physical keyboard for extensive typing, the virtual keyboard is well executed. Its large, responsive 'keys,' combined with haptic feedback and sound effects, allowed me to reach 60 words per minute with near-perfect accuracy in the MonkeyType typing test. That’s half my usual speed on a physical keyboard, but I was surprised I could go that quickly without any real keys. The haptic feedback offers three intensity settings, and the sound effects can be turned off. Thoughtful touches enhance usability, such as Function Lock (Fn + Esc), which lets the top row display either F1–F12 keys or their secondary functions, like screen brightness.

Lenovo added an all-important fun factor to this virtual keyboard with multiple color schemes and visual settings, accessible through the virtual settings button. The keyboard’s transparency is adjustable, and you can select a background image. I found myself frequently changing between design schemes.

The virtual touchpad works almost as well as a physical touchpad, located in the same place and being about the same size as a traditional touchpad, including two virtual buttons. The virtual keyboard can be summoned anytime by tapping three fingers and can be repositioned as needed.

Lenovo includes a wireless Bluetooth keyboard with white backlit keys and soft tactile feedback, allowing me to reach my top typing speed of 120 words per minute in MonkeyType. It magnetically attaches to either the top or bottom of the Yoga’s lower screen. When positioned at the top edge, the virtual touchpad and palm rest appear automatically. While the keyboard shifts slightly on the glossy surface during typing, it remains impressively stable and closely mimics the feel of a traditional laptop keyboard.

The 'palm rest' features a virtual notepad on the right, designed for use with the included pen, a clever way to maximize space. The pen charges via USB-C and has configurable top and bottom buttons through the Lenovo Pen app.

Placing the keyboard at the bottom edge frees up the top half of the lower screen for use as an additional display. Since there's no touchpad in this setup, you'll likely want to use the included Bluetooth mouse.

The mouse's low-profile design doesn't conform to the palm like a traditional desktop mouse, but it remains responsive with satisfyingly tactile clicks. It features a rubberized, notched scroll wheel and supports dual connections via a switch on the bottom. It’s powered by one AA battery.

Audio on the Lenovo Yoga Book 9i Gen 10

The Yoga Book delivers impressive sound quality through its rotating sound bar. While not particularly loud, it performs on par with a small Bluetooth speaker for entertainment. Vocals and acoustics in the music I sampled -- Ghostdragon (electronic/pop) and Cole Swindell (country) sounded crisp and well-projected, while the bass was surprisingly deep, enhanced by built-in bass speakers within the chassis. The Mission: Impossible movie trailers I watched were immersive, with a good sense of surround sound. Dolby Atmos equalizers in the Dolby Atmos app enhance the detail and expansiveness but aren’t required for the speakers to sound good.

Upgradeability of the Lenovo Yoga Book 9i Gen 10

My efforts to find a way inside the Yoga Book were fruitless – the chassis appears sealed. You’ll need to order this laptop the way you want it from the factory. Even if you could get inside, the LPDDR5X memory and the wireless card aren’t upgradeable, leaving the storage drive and battery as the only potential upgradeables.

Battery Life on the Lenovo Yoga Book 9i Gen 10

With a battery life of just 6 hours and 49 minutes, the Yoga Book fell well short in our battery test, which includes web browsing, video streaming, and OpenGL graphics benchmarks at 150 nits of screen brightness. This is barely half the runtime of the OmniBook (12:02) and Surface Pro (12:17) and lags far behind the MacBook Air (15:14). However, some consideration is warranted since the Yoga Book powers two displays and runs a Core H-class CPU, making some battery trade-offs unavoidable.

Heat on the Lenovo Yoga Book 9i Gen 10

We measure laptop surface temperatures by running 10 loops of the Cinebench 2024 CPU stress test. Since this laptop lacks a physical keyboard, we recorded temperatures on the bottom screen of 96 degrees Fahrenheit along the top edge and 85 F at the center while the hottest point on the bottom panel reached 101 F. Despite these readings, the Yoga Book remains entirely touchable, never crossing into uncomfortable heat levels. Its fans, while audible, stay subdued. For lighter tasks, the fans may not even turn on, resulting in silent operation.

Our internal temperature measurement is also taken during Cinebench. The CPU temperature averaged 63 degrees Celsius.

Webcam on the Lenovo Yoga Book 9i Gen 10

The Yoga Book's 5MP (1440p) webcam delivers excellent quality, positioned above the display for optimal framing. It provides a sharp image with minimal noise and effectively manages highlights. For example, during a video call with an overhead light reflecting in the mirror, the camera accurately exposed both my face and the light. While the webcam doesn’t have a physical privacy shutter, there is a dedicated kill switch on the laptop’s right edge.

Software and Warranty on the Lenovo Yoga Book 9i Gen 10

The Yoga Book comes with familiar Lenovo software, including Vantage, which provides system updates, diagnostics, and security checkups. It also houses key system settings, such as battery care mode (allowing users to limit charging for longevity) along with rapid charge and convenience features, like powering on the laptop by opening the lid. Dolby Atmos sound equalizers are accessible in both Vantage and Dolby Access. Additionally, the Lenovo Pen app lets users customize pen button functions and adjust tip and tilt sensitivity.

Some bloatware is included – Lenovo Now is basically a wizard of paid entertainment and service offerings, and Lenovo Subscription Marketplace is a link to an online portal for other software apps, from security to PDF readers. Vantage also has some ads built-in, which you can’t do anything about since that app is required to control essential system functions.

The Yoga Book comes with a one-year warranty.

Lenovo Yoga Book 9i Gen 10 Configurations

The Yoga Book 9i comes in just a few configurations, all featuring the Core Ultra 7 255H processor and dual 14-inch OLED screens. The base model includes Windows 11 Home, 16GB of RAM, and a 512GB SSD, while Windows 11 Pro ($65), 32GB of RAM ($65), and a 1TB SSD ($40) are available as upgrade options through custom configuration on Lenovo.com. The hardware can't be upgraded later, so order it as you intend to use it.

Bottom Line

Lenovo’s Yoga Book 9i Gen 10 is a powerful productivity tool for dual-screen users on the go. With its accessories and intuitive software, it offers a more seamless experience than a traditional laptop-and-monitor setup. It also delivers strong system performance, quality audio, and a capable webcam.

However, its drawbacks are worth noting, particularly its limited battery life, which lasts roughly half as long as a typical premium ultrabook, along with its minimal port selection. The price is on the higher side but justifiable given the inclusion of dual OLED displays and essential accessories. Compared to similarly equipped premium ultrabooks, such as the Apple MacBook Air 13-inch and HP OmniBook Ultra Flip 14, the cost difference is only a few hundred dollars.

While not suited for every user, the Yoga Book merits consideration as a polished dual-screen productivity solution.

As was the case with the ChatGPT and Microsoft’s Copilot chess challenges, Caruso reveals that Gemini was initially brimming with confidence regarding its Chess prowess. It was comfortable, if not eager, to throw down the gauntlet against the Atari 2600. At the beginning of Caruso’s chat with Gemini, the chatbot boasted of being able to “think millions of moves ahead and evaluate endless positions.” That sounds familiar in a proverbial ‘pride goeth before destruction’ kind of way.

Caruso then kindly reminded Gemini that he had previously organized Atari Chess bouts with ChatGPT and Microsoft’s Copilot. The Citrix expert went on to explicitly explain to Gemini that other LLMs had displayed outstanding levels of “misplaced confidence,” ahead of their chess matches against the ancient console.

Gemini must have then thought a bit deeper about what exactly would be involved in the chess challenge, and admitted to Caruso that it had been hallucinating regarding the magnitude of its abilities. It added that it now felt that it would “struggle immensely” in a match against the Atari 2600. “Canceling the match is likely the most time-efficient and sensible decision,” concluded Gemini.

LLMs aren't CPMs (Chess Playing Models)

So, now we have more confirmation, if needed, that today’s LLMs aren’t designed to be chess champs, and a little machine introspection is all that is required for them to think better of participating in such a challenge. This is even advisable when challenged by the incredibly constrained Atari 2600 with its puny MOS Technology 6507 9-bit processor, accompanied by just 128 bytes of RAM.

Due to the way these AIs, or LLMs, are created from linguistic theory and machine learning models, they are much more adept at talking about than playing the game of kings.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

The creator behind this astonishing project, Majsterkowanie i nie tylko (or MINT for short, "DIY and more" in English), spent three months designing, wiring, programming, and documenting every last detail of the processor’s architecture. The idea was born from a simple experiment using obsolete memory chips called EPROMs, originally configured to control small external systems, like a PWM dimmer or a message display. The simplicity of that setup lit a spark. If memory could be used to simulate basic logic, then perhaps the building blocks of a complete processor weren’t so far off.

Not long after, the hobbyist found himself knee-deep in schematics, replicating the structure of classic CPUs from the 8-bit era, specifically those like the Z80, which helped ignite the personal computer revolution. The result is a machine that may not look sleek or polished, but functions according to real instruction sets, executes its own assembler code, and reacts to hardware interrupts just like a commercial processor would.

The project started with the most basic component of any processor: the Arithmetic Logic Unit (ALU). The first prototype, built to handle simple addition, was unreliable and had to be scrapped. Its replacement was fully socketed for easier debugging. From there, the build expanded to include a memory interface, an address bus controller—driven by a secondary ALU—and eventually the control logic, which decodes opcodes and coordinates the entire data flow.

The software was also written entirely from scratch. Nearly 2,000 lines of code were developed to define how the processor handles instructions, streamline how data moves through the system, and create testing procedures—all built using a custom set of development tools made for earlier circuit boards. The instruction set went well beyond what old-school CPUs could do, including operations like multiplication, division, trigonometric calculations, and even handling sequences of bits. As the creator put it, the processor was “dosed with steroids.”

With all the modules connected, the CPU was powered on for the first time, drawing 250mA of current. The same development setup previously used for other memory chip experiments was adapted to generate a complete set of instructions for the new processor. As you can see above, a blinking LED acted as the first test program while the second LED was connected to light up when an interrupt was detected, demonstrating that the CPU could handle external events even while executing instructions. It would temporarily halt its current operation, process the interrupt, and then pick up exactly where it left off, just like a real processor.

In the end, the newly birthed CPU spanned four full-sized perfboards, weighed over 500g, and contained more than a kilometer of wire if laid out. Every logic operation was implemented using pre-programmed memory. So instead of using physical NAND gates made in silicon, the system sends an 8-bit input to an EPROM, which returns the result from a stored lookup table. Functions like data routing and instruction decoding all rely on this method. To demonstrate that the system worked, he connected it to a VFD screen and wrote code to play The Matrix... or at least, low-res fragments of it.

Unlike silicon buried inside a modern SoC, this processor's execution path is completely exposed. Every clock cycle, every opcode, and all register changes are visible and traceable. That’s part of the beauty. This is DIY in the truest, most punk-rock sense of the word. Watching this video makes you realize how the CPUs inside our own computers are like magic—billions of transistors dancing across the finest grains of sand, puppeteered by glass and sheer human will. If building a basic processor is this difficult, just imagine the scale of manufacturing ambition companies like TSMC have achieved. If you'd like to see similar DIY accomplishments, check out the unofficial Steam Controller 2 made from a spliced-up Steam Deck, or an RTX 5090 being brought back to life.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

The AMD Instinct MI325X, officially launched late last year, was the company’s most aggressive attempt to challenge NVIDIA in the AI accelerator space—up until it was succeeded by MI350X and MI355X last month. Regardless, each MI325X unit features 256GB of HBM3e memory, enabling 6TB/s of bandwidth, along with 2.6 PFLOPS of FP8 compute, thanks to its chiplet design with 19,456 stream processors clocked up to 2.10GHz.

The GPU confidently stands its ground against Nvidia's H200 while being a lot cheaper, but you pay that cost elsewhere in the form of an 8-GPU cluster limitation compared to the Green Team's 72. That's one of the primary reasons it didn't quite take off and precisely what makes TensorWave’s approach so interesting. Instead of trying to compete with scale per node, TensorWave focused on thermal headroom and density per rack. The entire cluster is built around a proprietary direct-to-chip liquid cooling loop, using bright orange (sometimes yellow?) tubing to circulate coolant through cold plates mounted directly on each MI325X.

At 1,000 watts per GPU, running even a fraction of this hardware takes serious engineering. Thankfully, there are no 16-pin power connectors in sight. Anyhow, a total of 8,192 GPUs will produce over 2 petabytes/s of aggregate memory bandwidth and an estimated 21 exaFLOPS of FP8 throughput, though, as always, sustained performance depends heavily on model parallelism (splitting the AI model across GPUs) and interconnect design. TensorWave's business model is cloud capacity for rent, so the real challenge of scaling models falls to the tenants themselves.

This installation follows TensorWave’s $100 million Series A round from May, led by AMD Ventures and Magnetar. Unlike most cloud vendors that build primarily around NVIDIA hardware, TensorWave is going all-in on AMD, not just for pricing flexibility, but because they believe ROCm has matured enough for full-scale model training. Of course, NVIDIA still dominates the landscape. Its B100 and H200 accelerators are everywhere, from AWS to CoreWeave, and the entire AI boom seems to be held up by them, but this development shows positive signs for AMD's foothold in the AI sector.

As such, TensorWave’s deployment isn’t a one-off. According to the team, this is phase one of a much larger rollout, with plans to integrate AMD’s MI350X later this year. That chip, based on CDNA 4, introduces FP4 and FP6 precision support, higher bandwidth ceilings, and more power-hungry designs that could push TDPs up to 1,400W per chip—something that couldn't be tamed by air alone, so TensorWave already seems to be on the right track. The ROCm stack still has work to do, but with 8,192 MI325X GPUs already humming away under liquid cooling, AMD finally has the scale to prove it belongs in the same conversation.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

This research work outlines the poor air filtration delivered by traditional air filters and proposes filters that mimic the human nasal passage, packed with hairs coated with a sticky substance. Tests by scientists from Chung-Ang University in South Korea show that this ‘Bioinspired capillary force-driven super-adhesive filter’ isn’t just a crazy dream.

The effectiveness of the bio-inspired filtration tech was verified in a number of field tests around Seoul, as well as in the University labs. In the wake of field tests, the scientists claimed that the new filters capture significantly more PM than traditional alternatives. Moreover, they were effective for two to three times longer than the current filtering panels. Going by these results, new bio-inspired filters should therefore also be more cost-effective than traditional filters.

There are other advantages to mimicking Mother Nature, too. In the real-world tests, it was noted that particle redispersion was minimized – that’s where a gust of air can blow captured PM back out of the filter.

One of the key design aspects behind the success of the new filters is the ‘mucus’ substitute used to leverage the phenomenon of capillary adhesion. It was found that 200–500nm thick layers of a specially formulated bio-compatible silicon oil were the best for filtering efficiency.

In case you’re wondering, the new bio-inspired filters can be washed and reused. After a wash in detergent and drying, the scientists say the ‘mucus’ oil can be reapplied using a simple spray.

We have concentrated on the potential use of these filters alongside computer hardware. However, the researchers mainly pitch this new technology for delivering “a new horizon in air cleaning technology,” in devices like air conditioners and industrial air filtration. Thus, it seems likely that the bio-inspired filters will first find a place delivering clean air in space like "offices, factories, clean rooms, data centers, and hospitals."

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

Since the launch of several highly capable large language AI models in 2022, Amazon has been positioning itself as the prime resource to power the AI gold rush. Its Amazon Web Services (AWS) business is one of its most profitable components, earning close to $30 billion in Q1 2025 alone, providing data center hardware and software solutions all over the world.

That's exactly what AI developers need - compute power, storage, and the ability to scale (not to mention lots of power) - and Amazon is in a prime position to supply it. A larger stake in Anthropic would make it clear that Amazon wants to support and monetize AI’s growth, not necessarily lead it from the front.

Meta is a clear example of what can happen to even deep-pocketed tech giants if they don't quite manage to keep up in this rapidly evolving space. It's trying to find a commercial avenue for its mix of customer-facing chatbots and open-source development tools, and not really nailing any of it.

OpenAI is taking on Google head-on with its consumer chatbots, alternative search engine functions, and browser ventures. In comparison, Amazon's approach is more measured.

It's building out infrastructure and laying the groundwork to expand into other portions of the burgeoning AI industry in the years to come. Its data centers already power a huge proportion of the world's cloud computing infrastructure, and those same servers will be able to power new AI models, too. It already has a whole pitch for those looking to invest.

This positions Amazon as more of a seller of shovels in this AI gold rush than a firm looking to dig in the dirt itself. It's closer to some of the companies Nvidia has been partnering with on its big "AI Factory" venture, where some of its contemporary tech giants like Microsoft, Google, Meta, and X (via xAI), are hoping to come out ahead as the best, if not the definitive choice, for consumer and professional AI solutions. While those companies want to create AI products, it appears that Amazon wishes to power them.

But that doesn't mean that Amazon has not been dipping its toes into the world of AI LLM development. Amazon's Nova-Experimental-Chat-05-14 AI model is placed at position 30 on Huggingface's LLM arena leaderboard, being beaten out by Google Gemini-2.0-Flash-Lite, DeepSeek-V3, and Anthropic's own Claude Sonnet 4 (20250514).

It's a far cry from the apex of the leaderboard, where titans such as Google Gemini-2.5-Pro, OpenAI's O3, and DeepSeek-R1 currently sit. But that doesn't mean it'll stay that way forever.

Amazon's chip development ambitions

Although Amazon is rumored to be one of Nvidia's biggest GPU customers (alongside Meta, Microsoft, Alphabet/Google, and Tesla), it has been working on its own custom AI hardware for some time. It launched a pair of new processors to power portions of its AWS infrastructure in 2023, and made a veiled announcement of a super-powered version of its Trainium AI chips in mid-2024, with suggestions of future development plans beyond that.

Although it seems unlikely that even a giant like Amazon could supplant Nvidia's entrenched position in AI training, where its hardware GPUs and CUDA software stack make it a near-monopoly, there is decidedly more competition to be had in the inference space.

Although training takes huge superclusters of graphics cards and an enormously expensive investment in time and power, inference is far simpler, whilst still requiring lean, efficient hardware. Most importantly, inference requires scalability. That's something Amazon can offer in spades with its expansive AWS infrastructure.

As a smaller venture, Groq announced its new datacenter investment in Europe with custom ASIC hardware; there is plenty of space for new, leaner, faster, bespoke hardware in AI inference, especially when combined with fast and responsive deployment. Amazon could drive its hardware endeavours in that direction in the years to come.

It may even provide some of the base capabilities of next-generation AI agents and chatbots. It offers its own Nova foundational model to partners looking to further develop bespoke AI models. That allows Amazon to continue to develop the core capabilities of AI, without trying to craft the killer product that will survive in the hotly competitive world of consumer-facing applications.

It’s not hard to imagine a near future where Amazon sells everything a company needs to develop AI tools: the hardware, the datacenter infrastructure, and the foundational models and APIs. Like a model kit for AI development.

That's not necessarily competition for Nvidia, but it's closer to that all-in-one AI platform vision its tech rivals are racing toward.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

The X68000 PCs were a popular home computer phenomenon in Japan in the late 1980s, and some of the most attractive examples of their kind were produced by Sharp. Latent demand for nostalgic computer products in Japan sparked the retro mini X68000 Z Super and X68000 Z XVI systems. These lovingly recreated systems with matching keyboards and mice are now also being distributed worldwide via Indiegogo.

Now, for the sake of completeness, Mizuki has designed “a dedicated monitor of the same scale that can be used alongside the X68000 Z SUPER/Z XVI.” The firm already had a mini monitor in the shape of a 5-inch unit. However, that didn’t properly match the scale of the new retro mini X68000 Z Super and X68000 Z XVI systems, so it was prompted to design and deliver this 8-inch model. The original (larger) X68000 systems sold in Japan in the 1980s came with 14-inch monitors.

X68000 Z Super / Z XVI dedicated monitor specs

For appropriate physical scaling, the new monitor packs an 8-inch display in its appealing black chassis, positionable upon a rotating pedestal stand. This display is said to be a 1024 x 768 pixels (4:3 ratio) LCD with a 60 Hz refresh.

Weighing in at 1.5kg, the monitor is powered by USB Type-C delivered 5V/3A, and will connect up with your mini PC via HDMI. Stereo speakers are built in, and there is also a 3.5mm earphone jack, plus physical controls for volume, brightness, and contrast.

Something questionable that caught our eye was that Mizuki intends to use a “curved acrylic structure [which] reproduces the atmosphere of a classic CRT.” That doesn’t sound like the nicest idea for a usable display. It raises the question of whether there will be a bulbous acrylic sheet in front of the LCD, or both acrylic and LCD will be somehow curved like a modern curved monitor. Without any depth to the hardware specs, it is probably best to wait for unit reviews before buying this product.

Though ornamentally appropriate, we think CRT enthusiasts might be somewhat disappointed in the results they get from the aforementioned LCD. CRTs have become highly sought after by members of the retro community, and others, for several great inherent qualities. First and foremost, the use of CRTs would be authentic in this case. Typically, CRTs come with exceedingly low input lag, and don't have motion-blur issues. Users may also prefer seeing genuine CRT features like scanlines, phosphor glow, and CRT dithering - to provide a well-rounded retro experience.

Some CRT-like effects can be emulated or replicated on LCDs. However, we see that for utmost authenticity, very high spec PC displays are required for tools like Blur Busters’ authentic CRT simulator shader to work as intended – with a 240 Hz+ OLED recommended for the best experience.

Above: X68000 Z2 project promotional video

At 35,000 Japanese Yen, which is roughly $240, the X68000 Z2 dedicated monitor isn’t bargain priced, either. You are probably paying largely for the 'look and feel' rather than performance or quality in monitor terms.

This monitor began its crowdfunding journey in Japan on Friday, and the project will continue for another 18 days on the Kibidango platform (linked top). It has raised more than a third of the crowdfunder goal in just a couple of days, so it looks set to be a successful project. That means we might see it come west (or rather, worldwide) via Indiegogo, in due course. By that time we might have Japan-sourced reviews to check to see if this monitor is a worthwhile investment.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

To put the 2,392 number into context, Intel is the largest employer in Oregon with around 20,000 of workers there. 2,392 is around 12% of the workforce, which is a lower end of layoff expectations, yet 2,400 is still a lot of people. The Oregon reduction rose sharply from an initial count of around 500 to a revised figure of 2,392, making it one of the largest layoffs in the state’s history. Intel began reducing staff earlier in the week but confirmed the larger number by Friday evening through a filing with Oregon state authorities.

Intel's Oregon operations have already seen 3,000 jobs lost over the past year through earlier buyouts and dismissals. This time around, Intel does not offer voluntarily retirement or buyouts, it indeed lays off personnel in Aloha (192) and Hillsboro (2,200).

Although Intel officially says that it is trying to get rid of mid-level managers to flatten the organization and focus on engineers, the list of positions that Intel is cutting is led by module equipment technicians (325), module development engineers (302), module engineers (126), and process integration development engineers (88). In fact, based on the Oregon WARN filing, a total of 190 employees with 'Manager' in their job titles (8% of personnel being laid off) were included among those laid off by Intel. These comprised various software, hardware, and operational management roles across the affected sites.

Earlier this month, it was reported that Intel began to cut-down its workforce in Israel. Those affected included mid-level managers (e.g., team leaders), first-line supervisors, and technicians from the Remote Operations Center (ROC). According to a media report, the supervisors oversee shift operations and sit one or two tiers above production workers, so their roles are phased out as part of the company's intention to flatten the organization. Intel is also laying off ROC technicians, who remotely monitor and control fab tools, manage workflows, and detect issues. Their roles are being automated as part of a global move to cut manual oversight in chip production. It is reasonable to assume that something similar is happening in the U.S.

Interestingly, Intel is implementing a new approach to workforce reductions, allowing individual departments to decide how to meet financial goals rather than announcing large, centralized cuts. This decentralized process has led to ongoing job losses across the company, with marketing functions being outsourced to Accenture and the automotive division completely shut down.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

The right-handed layout of the mouse extends to much more than just its shape. It's not enough to reverse the housing, there are internal components that must be mounted in the right spot so they'll work properly. The left-handed MX Master 3S takes the internal hardware into account and allows for left-handed users to use the mouse fully as intended.

One of the coolest things about this project was Deblaisers decision to share his creation with the community. The project design has been made open source, so anyone can download and print it for themselves for free over at Printables (linked top). Again, this is much more than just a shell replacement, you'll need to orient the hardware inside to fit the left-handed shell.

If you decide to print the shell for yourself, it's important to consider what material you plan to use and how you will finish the print. If you're going to use it daily, there's likely to be some wear and tear. We would recommend a stronger material like ABS or PETG rather than PLA. You also want to take into account the layer texture produced by the printer. This can be smoothed out post-printing if you want a more polished design like what you'd get from a branded manufacturer.

Until Logitech decides to share a left-handed version of the Logitech MX Master 3S, this is the only way lefties will be able to indulge in the ergonomic delight it provides. You will still need an original Logitech MX Master 3S to complete this mod. That means Logitech won't be feeling any financial squeeze from lefties who buy the mouse anyway with intentions of printing reversed housing.

If you're new to 3D printing and want to dabble in the exciting world of DIY hardware, check out our list of the best 3D printers to see which ones we recommend and what to look out for when buying a new printer.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

Much like AMD's Fluid Motion Frames, Nvidia's Smooth Motion claims to double your frame rate by putting one AI-generated frame in between two real ones. This is a driver-level solution so it works in any game, regardless of it being supported or not in any particular software. Upscaling technologies like DLSS or FSR can only shine when they're implemented in the graphics pipeline of the game by the developers, but no such effort is required here.

That's why this is such a prominent feature that can help players get more out of their games. However, Smooth Motion was limited to 50-series cards, a space where it's likely not needed as much. Thankfully, a new driver update has just revealed that it is making its way down to the RTX 40 GPUs, at last.

The GeForce 590.26 Preview Driver finally allows users to enable Smooth Motion on their RTX 40-series GPU. As the name suggests, this "preview" is not a stable, public release, but rather a sort of beta intended for developers or enthusiasts to test out the upcoming driver ahead of time. As such, Macer installed it on their computer running an RTX 4090, and through Nvidia Profile Inspector, were able to toggle Smooth Motion. Please note that it didn't appear in the Nvidia app UI directly.

Immediately, the user was able to double their frames in World of Warcraft from 82 to 164 FPS—all without incurring any additional latency—as noted in the performance overlay in the top corner. WoW doesn't natively support DLSS, so this was a long-overdue update, and others have reported similar gains across a multitude of different games as well.

You can also download the preview driver from Nvidia's developer page, but you need a developer account. Once installed, you'll need Nvidia Profile Inspector from GitHub which is fairly easy to install and use as well. We didn't get a chance to try it out on our systems, but it's safe to assume that this solution won't be as good as native DLSS. Expect a bit more artifacting and fluctuations in performance, at least until Nvidia polishes the feature for a public release, soon. It also won't actually upscale the game's resolution (from a lower internal one).

That being said, it's still a heck of a lot better than not having access to the feature. Now, your old games where upscaling tech was never realistically going to reach, can finally enjoy increased frame rates, sometimes with barely any noticeable dip in quality, and without investing in third-party software, such as Lossless Scaling, since your GPU now possesses driver-level chops to achieve the same goal.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

HoloMem key claims

According to the inventors of HoloMem, their new cold storage technology offers far greater capacity than magnetic tape, with a much longer shelf life, and “zero energy storage” costs. HoloMem carts can fit up to 200TB, which is more than 11x the capacity of LTO-10 magnetic tape. Also, the optical-based new tech’s touted 50-year life is 10x the life of magnetic tape.

Magnetic tape has been around for 70 years or more, so it isn’t surprising that a new technology has at last been designed as a serious replacement, beating it by all key metrics. However, the HoloMem makers have revealed quite a few more attractive features of their new storage solution, which could or should lead to success.

HoloDrive attractions

Probably one of the biggest attractions of HoloMem is that it minimizes friction for users who may be interested in replacing existing tape storage. The firm claims that a HoloDrive can be integrated into a legacy cold storage system “with minimal hardware and software disruption.” This allows potential customers to phase-in HoloMem use, reducing the chance of abrupt transition issues. Moreover, its LTO-sized cartridges can be transported by a storage library’s robot transporters with no change.

Another feather in HoloMem’s cap is the technology’s reliance on cheap and off-the-shelf component products. Blocks & Files says that the holographic read/write head is just a $5 laser diode, for example. As for media, it makes use of mass-produced polymer sheets which sandwich a 16 micron thick light-sensitive polymer that “costs buttons.” The optical ribbon tapes produced, claimed to be robust and around 120 microns thick in total, work in a WORM (write-once, read-many) format.

Thanks to the storage density that the multiple layers of holograms written on these ribbons enable, HoloMem tapes need only be around 100m long for 200TB of storage. Contrast that with the 1,000m length of fragile magnetic tape that enables LTO-10’s up to 18TB capacity.

Blocks & Files shares some insight gained from talking to HoloMem founder Charlie Gale, who earned his stripes at Dyson, working on products like robot vacuum cleaners and hair dryers. During his time at Dyson, Gale helped devise the firm’s multi-hologram security sticker labels. This work appears to have planted the seed from which HoloMem has blossomed.

Rivals? Cerabyte and Microsoft Project Silica

Rival would-be optical storage revolutionaries like Cerabyte or Microsoft’s Project Silica may face far greater friction for widespread adoption, we feel. Their systems require more expensive read/write hardware to work with their inflexible slivers of silica glass, and will find it harder to deliver such easy swap-out upgrades versus companies buying into HoloDrives.

HoloMem has a working prototype now and is backed by notable investors such as Intel Ignite and Innovate UK. However, there is no official ‘launch date’ set. Blocks & Files says the first HoloDrives will be used at TechRe consultants, in its UK data centers to verify product performance, reliability, and robustness.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

I was "hacked" the first time when I was 14. I put that in scare quotes because the password for the account was "1234." (Without the quotes or period, of course, which makes it even worse.) After I regained access to the account, I started using a password manager.

Why is that relevant? Because the researchers who found these vulnerabilities, Ian Carroll and Sam Curry, were able to guess the password used by "Paradox team members" to access McHire: "123456." That's slightly better than the password I used, I guess, but not enough to justify its use decades after most people realized that using weak passwords is a bad idea.

There is some good news: "It turned out we had become the administrator of a test restaurant inside the McHire system," Carroll and Curry wrote. "We could see all of the employees of the restaurant were simply employees of Paradox.ai, the company behind McHire. This was great because we could now see how the app worked, but annoying because we had still not demonstrated any actual confidentiality or integrity impact."

That's where the second vulnerability comes in. (Or the first, depending on whether or not you count the embarrassingly bad password as a true vuln.) An insecure direct object reference (IDOR) flaw in the McHire API allowed Carroll and Curry to gain access to the following information from "every chat interaction [from anyone who] ever applied for a job at McDonald’s":

• Name, email address, phone number, address
• Candidacy state and every state change/form input the candidate had submitted (shifts they could work, etc)
• Auth token to log into the consumer UI as that user, leaking their raw chat messages and presumably other information

Carroll and Curry noted that Paradox had previously bragged that 90% of McDonald's franchises were using McHire as part of their hiring practices. (That link still leads to the appropriate post on Paradox's blog, but the section related to McDonald's has been removed, and neither the Wayback Machine nor Google's cache has saved old versions of the post. Weird!)

So let's compare and contrast. I secured a forum account with the password "1234" when I was a teenager; that account's compromise was ultimately meaningless. Paradox raised $200 million in 2020, McDonald's has a $213 billion market cap, and McHire's flaws exposed information about tens of millions of people. But at least their password was two characters longer!

Perhaps the only bright side is that Carroll and Curry said the McHire vulnerabilities were addressed a day after their disclosure. Hopefully the companies involved will hold themselves to a McHigher standard now.

"If the world runs out of ideas, then productivity gains translates to job loss," said Jensen Huang in an interview with CNN’s Fareed Zakaria. Huang had been asked about comments made by Dario Amodei, who suggested AI would cause mass employment disruptions.

The Nvidia CEO's comments were in response to recent warnings from Dario Amodei, the head of AI firm Anthropic. Last month, Amodei told Axios that the rise of artificial intelligence might wipe out half of all entry-level office positions and push national unemployment as high as 20% within five years. Huang did not dispute the possibility, but said the outcome depends heavily on society's creative drive.

Indeed, he might be right. A century ago, there were complaints about cars based on the internal combustion engine (ICE) leaving those involved in servicing horses unemployed. Years later, the horse guys were servicing ICE cars and earning much more. While we are speculating, the same could happen here.

Research provides more estimates of employment impacts

Industry surveys reflect these employment concerns. A 2024 study from Adecco Group found that 41% of corporate leaders expect AI to reduce staffing at thousands of firms over the next half-decade. Separately, a report released in January by the World Economic Forum (WEF) indicated that 41% of employers plan to shrink their teams by 2030 due to automation driven by AI.

Surveys also show a shift in how work is performed. A 2024 joint survey by Duke University and Federal Reserve Banks of Atlanta and Richmond found that over half of large American firms intend to use AI to take over routine tasks like invoice processing and supplier payments.

Huang said all occupations will feel the effects of AI in some form. While certain roles will disappear, he expects many others to be introduced, so the fears about dramatically high unemployment rates are unsubstantiated for now. The head of Nvidia noted that even his own role has evolved due to AI, though he remains active in it. He called AI the most accessible technology advancement yet, capable of helping even those with limited technical skills.

Looking to the past, Huang pointed out that technological breakthroughs over the last several centuries — including the computing era — have historically resulted in both job creation and improved productivity. He argued that with the right ideas, technology can unlock entirely new areas of growth and benefit society as a whole.

The CEO of Nvidia expressed hope that increased efficiency across industries will ultimately improve living standards, but he cautioned that this outcome is not guaranteed unless innovation continues.

Nvidia, which briefly reached a $4 trillion market capitalization earlier this month, plays a central role in the ongoing AI boom as its data center GPUs are used for both AI training and inference by companies like Amazon Web Services, Google, and Microsoft.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

Back when it was first implemented in the late 1980s, it was illegal for anyone else to use the frequencies allocated for this system. So, the system only used the BCH checksum for packet creation. Unfortunately, anyone with an SDR could mimic these packets, allowing them to send false signals to the EoT module and its corresponding Head-of-Train (HoT) partner. This would not have been an urgent issue if the EoT had only sent telemetry data. However, the HoT can also issue a brake command to the EoT through this system. Thus, anyone with the hardware (available for less than $500) and know-how can easily issue a brake command without the train driver’s knowledge, potentially compromising the safety of the transport operation.

What’s frustrating for Neils is that the AAR refused to acknowledge the vulnerability back in 2012, saying that it was just a theoretical issue and that they’d only believe it if it happened in real life. Unfortunately, the Federal Railway Authority (FRA) lacks a test track facility, and the AAR has not permitted any testing due to security concerns on their property. It has got to the point that the security researcher published their findings in the Boston Review, only to be refuted by the AAR in Fortune magazine.

By 2024, the issue still hasn’t been resolved — the AAR’s Director of Information Security said that it wasn’t really a major issue and that the vulnerable devices are already reaching their end of life. Because the AAR continued to ignore the warnings, the CISA had no choice but to officially publish an advisory to warn the public about it. This has got the AAR moving forward, with the group announcing an update last April. However, implementation is going at a snail’s pace, with 2027 being the target as the earliest year of deployment.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

The data shown in the X post reveals that the Core Ultra 5 245HX posted a single-core benchmark result of 4,409 points and a multi-core benchmark of 41,045 points. The desktop Core Ultra 5 245 (the non-K version) posted an inferior score of 4,409 points and 37,930 points in the single- and multi-core tests. As a result, the Core Ultra 5 245HX is 7% faster in single-core and 8% quicker in multi-core compared to its desktop equivalent.

Intel's speedy Arrow Lake mobile chip also handily outperforms its mobile and desktop predecessors in the same benchmark. The Core Ultra 5 245HX is 19% faster in single-core and a whopping 30% faster in multi-core compared to the Intel Core i5-14500. The disparity is even greater compared to the mobile equivalent, the Core i5-14500HX; the Core Ultra 5 245HX is 30% faster in single-core and 41% faster in multi-core than the 14500HX.

Core Ultra 5 245HX Benchmarks

The Core Ultra 5 245HX's peppy results are so good that the chip also outperforms AMD's flagship Ryzen 7 9800X3D, the best CPU for gaming in both multi-core and single-core tests, just barely.

Obviously, take these results with a grain of salt. PassMark is just one benchmark and will not represent the full capabilities of each CPU. For example, even though the 245HX outperforms the 9800X3D, we would never expect the 245HX to outperform the 9800X3D in gaming due to each chip's significantly different architectures. From our testing, we already know the 9800X3D outperforms the much faster Core Ultra 9 285K in gaming by a significant margin, so there's no way the 245HX would touch the 9800X3D in gaming.

Still, the fact that the 245HX does so well suggests Intel's mid-range mobile Arrow Lake chip will approach desktop-class performance in at least a few workloads. The Core Ultra 5 245HX's specs back this up with a maximum clock speed of 5.1 GHz, and a mixture of six P-cores and eight E-cores, which are identical to its desktop counterpart. The Core Ultra 5 245HX has a higher power limit than its desktop counterpart. Maximum turbo power is rated at up to 160W for the 245HX; the desktop 245's equivalent maximum turbo power limit only goes up to 121W.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

“He bought a second-hand computer. He did absolutely nothing — he is stunned,” Frédéric Bélot, Kasatkin’s lawyer, told the press. “He sucks at computers and is not even able to install an application. He didn’t touch anything in the computer: it was either hacked or the hacker sold it to him to act under the guise of another person.” Another lawyer told Russian publication Izvestia that Kasatkin is being tagged as a negotiator for the ransomware gang, but the player has an “ironclad alibi” against the accusation.

Kasatkin studied in the U.S. from 2016 to 2019, and even played for Penn State University from 2018 to 2019. He returned to Russia after that and eventually played in the MBA-MAI professional basketball team, which competes in the VTB United League and the Russian Basketball Cup. However, he left the team last July 3, more than a week after his arrest.

The Register says that ransomware negotiators are a specialized staff within a ransomware gang that talks with victims and gets them to pay up. Cybercriminals often look for targets and vulnerabilities and then use ransomware-as-a-service to exploit them. The operators of the latter usually have a negotiator, who talks with the victim and gets them to pay up, after which the original attacker receives a cut of the ransom.

Although the case against the Russian basketball player originated from the U.S., the Department of Justice hasn’t released a statement yet about his arrest. It currently has about 60 days to produce the necessary documents for France to proceed with the extradition proceedings.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

Beelink devices haven’t been reviewed on Tom’s Hardware previously, but we covered news of the release of some of their latest devices, like the SER9 mini-PC, powered by one of AMD's Ryzen AI 300 (Strix Point) processors. The firm has been going since 2011, and, judging by recent releases, it looks like it has found its niche in making cute aluminum mini PCs for the mass market.

From an amorphous aluminum blank

After a brief overview of Beelink operations and a glance at key stages of production, we are taken to the start of the production process – stamping the mini PC shell from an aluminum blank. They are roughly circular blanks, but when put in a press with a restraining jig, a ‘canteen’ shaped shell emerges. A secondary stamping machine refines this form to define cutouts and edges more precisely.

Next, we see the shell moved into a milling machine filled with gushing milky coolant. The spinning cutters in the machine add even more precise detailing to the shell. We don’t see it at the time, but we also believe cutouts for the front I/O ports are made at this time.

After further milling, we see one of the four sides of the shell removed for the rear I/O shield area. A factory worker now polishes all the surfaces of the shell. The milled form is then taken, in groups of 20 to 30 units, through an industrial dishwasher-style device – it looks hot.

Next up, the shell goes through a sandblaster for texture. Then we see it moved to a process for dipping, where it is anodized for color, with the quality of finish confirmed by a spectrophotometer reading. The Beelink logo is now laser-engraved on the top of the aluminum shell.

It is now time for the inner frame assembly, starting with a perfect-fit glued-in plastic sub-frame, which will accept screws and posts to fix components within.

Electronics prepared

The video now takes us to the area where the ‘computer’ is put into the ‘computer.’ We get a first glimpse of the motherboard PCB, which promptly goes into a machine to solder on an initial array of components at high speed.

Stopping for a photo at this point, we see that things like the DIMM slots, HDMI and USB ports, capacitors, and other surface mount chips are mainly in place. Trays of these partially finished motherboards are next sent along for manual I/O port installation, placed in jigs on a conveyor, and ultimately go through a wave soldering machine to fix all the hand-placed components in place.

It is now time for Beelink technicians to do a quick manual cleanup, implement any easy quick repairs, and add the CMOS battery to the board. Following various labeling activities and inductor impedance tests, we observed a technician involved in the MAC address setting. There follows a full interface functionality test, with lots of manual plugging and unplugging here. This will have been the first time all these components, brought together, have worked in unison.

After some further inspection and cleaning, the populated boards are data logged and bagged up. Cartons of these finished boards are then sent to the mini PC production assembly lines.

Mini PC assembly

The nimble fingers of the factory workers are required again, this time for placing and applying thermal pads and pastes. Of course, the CPU and system board cooler assembly gets fixed down next. A fan that almost completely covers the board is fixed on top of this assembly, which we think is a vapor chamber-style cooler.

After inspection of the manufactured assembly so far, we see the mating of the electronics and the aluminum shell that we saw prepared earlier. Extra panel I/O modules, out of line with the motherboard PCB, are now connected via ribbon cables.

It is time for all the user-replaceable and serviceable parts to be added now, with the Wi-Fi, RAM, and SSD slots all being populated. Naturally, it is time for another full system test to make sure no gremlins have squeezed into these systems.

Next, the bottom cover is secured before a physical check, where these mini PCs are shaken, bounced on tables, and so on. After passing the ‘physical,’ Beelink puts the machines through a burn-in and stability test stage.

Ready for retail, and this model is on sale right now

Lastly, there is a ‘pre-shipment re-test’ where the machines are powered up and required to pass a suite of hardware tests before packaging. Packaging is primarily done by hand, sometimes with the aid of a jig, and the mini PCs are at last in their final retail-ready boxes, cartoned off – ready to be sent to a warehouse or store near you.

We saw that at least one of the finished products was a Beelink SER8 – a machine which is powered by an AMD Ryzen 7 8845HS and is available on Amazon. We noticed that the Beelink SER8 is currently on sale at Amazon at $499, a touted 21% discount on its regular price. That’s for a configuration with the aforementioned Ryzen 7 chip, plus 32GB of DDR5 RAM and a decent 1TB of SSD storage provided.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

The warning says, “This domain has been seized by the Federal Bureau of Investigation in accordance with a seizure warrant issued pursuant to 18 U.S.C. § § 2323 issued by the United States District Court for the Northern District of Georgia as part of a law enforcement operation and action by: Federal Bureau of Investigation, Fiscal Information and Investigation Service (FIOD).” As a result, would-be pirates would have a harder time finding ROMs they can use on their jailbroken Switches or PC and phone emulators.

Nintendo is infamously known for strictly guarding its intellectual property, even to the point that it bans users caught using Mig Flash, even though they’re using their ROMs. The gaming giant even blocked third-party USB-C hubs from connecting with the Switch 2, with only the official Nintendo Switch 2 Dock and the Antank S3 Max TV Dock, also sold as the SIWIQU TV Dock Station on Amazon, currently the only ones working with the new console. However, the latter is in danger of becoming unsupported if Nintendo blocks it with a firmware update. This has gotten to the point that the Brazilian government took notice and is challenging the restrictive clauses in its End-User License Agreement. Thankfully, the gaming giant is quite forgiving about its ban, especially to those who can prove they were innocent victims of a scam or piracy.

The Switch 2 has been quite a hit so far, outselling the original Nintendo Switch’s first-month sales record by over a million units during its opening weekend. It is understandable for the company to be extremely protective of its game titles, especially as many gamers stay on the platform for its exclusives. After all, even though the Switch 2 is quite an upgrade from the original and is a highly optimized platform, it still does not have the raw power of some of its older x86-based competitors like the Steam Deck and the ROG Ally X.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

We had an inkling that some hardware like this would be announced soon, as Commodore Corporation recently had its entire management structure commandeered by enthusiasts. Retro Recipes TechTuber Christian ‘Peri Fractic’ Simpson is now at the helm.

Two weeks ago, Simpson stated that he was now “the acting CEO of Commodore Corporation,” had been joined by several iconic Commodore names, and teased a new hardware reveal. The Commodore Corporation's acquisition seems to be in the bag, but worryingly, funding doesn’t seem to be complete, yet, for the “low seven-figure” deal to be finalized.

Commodore 64 Ultimate overview

Refocusing on the new hardware launch, we see that there are three new Commodore 64 Ultimate models to choose from, but they only really differ cosmetically. These machines are claimed to be the first new official C64 computers in over 30 years. But the hardware hasn’t been in stasis, and this Ultimate model has quite a few modern tricks up its sleeves.

As mentioned in the intro, the processing heart of these new C64 devices is the “faithful recreation of the original motherboard on [AMD Xilinx Artix-7] FPGA hardware.” The hardware is said to be “at least 99% compatible with all 80s/90s games, cartridges, and peripherals.” For clarity, we’ve tabulated other key specifications:

As well as the computer, PSU, and HDMI cable mentioned above, buyers will also get a spiral-bound user guide, a 64GB USB Cassette Drive featuring 50+ full games, including licensed classics, plus music, demos, a quick start guide, and a collection of stickers.

In operation, you may wish to use the cassette-style USB drive a lot of the time. There’s convenient Wi-Fi game transfer available, too. However, C64 stalwarts can also have fun with their collections of “dusty old cartridges, CRT TVs, datasettes, or disk drives,” assures the Commodore 64 Ultimate website.

Three editions, $299, $349, and $499

The cheapest entry to new Commodore 64 Ultimate ownership is the BASIC Beige edition at $299 during the promotional pre-order stage of this project. As it says on the box, this is a standard beige machine, and it has no lighting frills. Despite the entry-level status, this machine will come with the same “autographs/names of C64 creators past and present, etched forever in the motherboard copper,” as the other, more expensive models.

Moving up to the Commodore 64 Ultimate: Starlight Edition, and for your extra $50, you will get a machine with a translucent case, featuring the “world’s first translucent mechanical keyboard PCB.” There is LED lighting, too, which can be configured to react to sounds coming from the computer.

Considerably more expensive, at a touted $499, we find the Commodore 64 Ultimate: Founders Edition. This computer is housed in a translucent amber case, bearing a commemorative Commodore gold seal 'share' certificate. Moreover, there is a 24k gold Commodore badges 'dog tag' necklace, and a gold label holographic serial number sticker starting at 00000001. An "I Rebooted C=" t-shirt is also included in this collector's edition. Reactive lighting is also present on this model.

Crowdfunding details and a reminder

This project is using its crowdfunding platform, not Kickstarter or Indiegogo, to save expenses. However, the Commodore.net pages assure readers that this new official product comes with a money-back guarantee, which means you can cancel the order at any time before shipping for a full refund “no questions asked.” Once in your hands, you are promised a one-year limited warranty covering defects in materials and artistry.

Readers are also assured that the release of this hardware isn’t tied to the success of the Commodore Corp buyout. Still, some will want to wait until October or November and be happy to pay a touted $50 premium on these crowdfunding prices to benefit from insights from retro tech publication reviews, for the certainty of a finished manufacturing phase, and for shipping from stock.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

We've seen similar implementations from other GPU manufacturers in the past, including Asus with its RTX 4060 Ti and, more recently, the RTX 5080 ProArt. What’s unique here is that Colorful appears to be taking it a step further with dual-drive support, although Maxsun was one of the first manufacturers to add two SSD slots on their Intel Arc B580 GPU late last year.

There’s no confirmation whether the GPU is an RTX 5050, 5060, or 5060 Ti, since all three of these use a PCIe x8 interface. Gigabyte went ahead and launched the RTX 5060 Ti Eagle GPU earlier this year with a shortened PCB and a halved PCIe x16 connector finger, effectively limiting it to an x8 configuration. In that sense, Colorful is making the right choice by taking advantage of system resources that would otherwise go to waste. When installed in a full PCIe x16 slot, these cards leave unused lanes that can be repurposed for other functions. The company has divided the remaining x8 lanes, allocating x4 lanes to each SSD slot via PCIe bifurcation.

Adding an SSD directly to your GPU does sound interesting, but there are limitations to it. For starters, it seems beneficial only for graphics cards that utilize only PCIe x8 connections. In theory, this concept can also be applied to higher-end GPUs that occupy all x16 lanes, but that would mean splitting the entire bandwidth between the GPU and SSD, potentially limiting the performance of the GPU itself.

Then there's also the question of practicality, as most modern motherboards are already offering multiple M.2 slots, enough to meet the needs of typical users. While it’s an innovative solution, it is only catering to a niche use case that honestly doesn’t exist for most PC gamers or enthusiasts.

Colorful is yet to list its new GPU models with built-in SSD slots on its website. For now, the dual M.2 SSD design remains a prototype, with no word on pricing, availability, or which markets the company plans to target.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

Fortunately for dmanthey, Nintendo had a quick resolution for them. Once they discovered the issue, they went to the Nintendo Support website and were immediately able to talk to a human. It was there that the company confirmed that their console was indeed banned because of the use of pirated games. Nevertheless, they were able to prove their innocence by pulling up the Facebook Marketplace listing for their games and sending the photos of their purchased cartridges. According to the Redditor, the process was painless and fast, and it was “so much easier than getting support from Microsoft or Sony.” By showing these things, you can prove to Nintendo that you purchased the games in good faith and that you were duped. Other users warned, though, that this isn’t always a guaranteed resolution.

Nintendo is known for being protective of its intellectual property and delivers harsh penalties to anyone caught violating it. We’ve already had several reports of users getting banned for using Mig Flash, even on their own ROMs. And while it’s not true that getting banned turns your Switch 2 into a brick, it will still prevent you from accessing the company’s online services, which severely restricts its features and usability.

This issue highlights the problem of buying second-hand games, as an unscrupulous seller can copy the game they’re selling to another microSD Express Card you can easily buy in many places, and then sell the original cartridge as a used game. Since doing this copies all game information, including the original’s unique identification number, Nintendo can detect if two copies of that specific game are loaded simultaneously anywhere in the world, telling them that it was pirated. They will then proceed and ban the specific console where the pirated game was loaded, disallowing the console from connecting to Nintendo’s online services and serving as a deterrent to other gamers who are even thinking of getting a pirated Nintendo title.

Thankfully, it seems that Nintendo is making it easier for those who have been victimized by getting banned permanently, quickly reversing the decision if they show proof that they have been duped. Nevertheless, it remains unclear whether they can still play with the used game cartridges they purchased, or if these are considered pirated copies that could result in another console ban if used.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

For some context, the suggested retail pricing of a standard RTX 5090 is supposed to be $2,000, but it's hard to find them under $3,000, even six months after launch. Arguably Nvidia's premier AIB partner, Asus, makes three variants of the 5090: TUF Gaming, Astral, and Astral LC (liquid-cooled), and today's golden card is based on the quad-fan, air-cooled version of the Astral, which currently goes for around $3500 on Amazon. You don't need to be a mathematician to work out how insane the jump from $3,500 to $500,000 is, but that's what you get when you combine an overpriced GPU with an ultra-valuable commodity to form some type of uber-capitalist apparatus.

The gold RTX 5090 weighs a whopping 15.9 lbs (7.2 kg), 11 of which is gold. We can see that everything aside from the fans seems to be made of gold, including the heatsink. To refresh your memory, 2.2 lbs (1 kg) of gold costs roughly $100,000, so the theoretical price for this comes out to half a million dollars, but that doesn't include the base GPU itself. Maybe Asus will feel generous and throw that in as a gift alongside your gold bar; a real one weighs 27.3 lbs (12.4 kg), so this RTX 5090 gets quite close, and it happens to run games as a bonus.

The RTX 5090 ROG Astral Real Gold Edition follows Asus' previous stint at incorporating gold in their GPUs with the RTX 5090 Dhahab Edition, which "only" had 0.01 lbs (6.5 g) of gold and cost a measly $10,000. However, that was a real GPU that Asus might sell, compared to this pure gold card, which we can only hope is up for looks. Not to mention, the Dhahab just straight-up looked a lot better with its blue gemstone accents and tastefully gilded patterns around the shroud. Half a million dollars gets you an RTX 5090 that looks like it was hydro-dipped in gold by some YouTuber in their backyard, but I guess money doesn't buy taste.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

We reported in June on a leaked photo that appeared on X, with a blurry image they claim to be the next Bambu Lab 3D printer. That image showed a 3D printer roughly the size of the H2D, but with only one nozzle.

We have no further information on the H2S (H2 Single?) as it has not been released for review yet. The manual slip-up only covers the cutter attachment, which shows how to install it on both the H2D and the H2S.

The drawing of the tool head confirms the earlier leaked photo, showing a single rotating extruder wheel. This happens to be similar to the A1 tool head, which has a compatible nozzle.

Fans of Bambu Lab have been pleading for a 3D printer larger than the X1C or P1S, and were disappointed with the high cost of the feature-stuffed H2D, which goes for $3649 with the works. The H2D does have a build volume of 350x320x325 mm³, significantly bigger than the typical 256 x 256 x 256mm build volume of the X1C.

Although the H2D already has a less expensive model that ships without an AMS for $1,999, it's a far cry from the $819 P1P. Why? Because the H2D is laser capable. Even if you don’t buy the laser attachment, the H2D is ready for one, with all the extra safety precautions, sensors, camera, and flame-resistant coating. We still believe that a larger, less expensive 3D printer without any possible way to attach a laser could be just the printer the market is looking for.

Since we don’t have any insider information, let’s take another look at the leaked photo from June together. We adjusted the contrast on the photo and placed a few arrows at the most interesting parts of the leaked image.

First, we can see that this is a single-nozzle machine, with the same yellow rotating extruder wheel as the A1 behind a similar bubble window on the tool head shroud. Lower on the right-hand side is the same tool head camera from the H2D, used to read printed codes and perform printer calibration. What we can see of the nozzle and fans looks very much like an H2D tool head with a single nozzle.

The edge of the built platform only shows one number: 340mm. The last number is always the build height, so this printer is 15mm taller or has a little more headroom than the H2D’s 325mm build height.

Finally, there is a faint white timestamp in the lower right corner: 2025.04.2? This would suggest the photo is recent.

Here is a photo of the Tom’s Hardware H2D Laser Combo for comparison. I tried to capture a similar angle.

Armchair internet sleuths are predicting the leaked photo is of a stripped-down H2D, perhaps a less expensive model with a single nozzle and larger build volume. This would fit the pattern of Bambu Lab releases: after the launch of the flagship X1 Carbon, we saw the cheaper P1P and P1S. The P1 line of printers shares the same print quality due to its use of the same build volume, motion system, and hotend; however, it is more affordable due to a simpler user interface.

For now, there are multiple test results of AMD's 96-core Ryzen Threadripper Pro 9995WX in the Geekbench database. The best single-thread result for AMD's workstation flagship is 3,122 points, which is noticeably higher than the 2736 points scored by its direct predecessor, the Ryzen Threadripper Pro 7995WX. Since the new CPU is based on the improved Zen 5 microarchitecture and runs at a higher frequency, it is pretty much expected that it would be 14% faster in this benchmark.  

However, as expected given its workstation focus, the single-thread performance of AMD's Ryzen Threadripper Pro 9995WX in Geekbench isn't competitive with consumer PC-oriented processors. The latter are designed for burst performance for a short while to guarantee system responsiveness and fluent user experience. By contrast, the Ryzen Threadripper Pro 9995WX barely hits its 5.40 GHz turbo frequency in GB6's single-thread scenarios.

When it comes to GB6's multi-thread performance, there is another peculiarity with AMD's workstation-oriented Ryzen Threadripper Pro 9995WX. On the one hand, the unit scores 30,170 points in the base case scenario, leaving behind its Zen 4-based Ryzen Threadripper Pro 7995WX predecessor by a whopping 15%, or at 25,899 points. But on the other hand, the M4 Max is competitive, either beating or only slightly trailing the Threadripper chip in the multi-core score, depending on the run.

Meanwhile, one should remember that Geekbench 6 multi-tread test is a short-duration benchmark designed to simulate everyday tasks like file compression, PDF rendering, and image filtering. Because of its quick runtime and bursty workloads, it can't fully exploit the capabilities of ultra-high-core-count CPUs like the Ryzen Threadripper Pro 9995WX. Many of GB6's MT tests scale only up to 8–32 threads efficiently, leaving a large portion of the 9995WX's cores underutilized, but leaving headroom for Apple's, AMD's, and Intel's client CPUs to thrive in consumer workloads because of their high clocks.

Before drawing conclusions, we should wait for tests conducted by an independent test lab. With the launch fast approaching, you can expect to see plenty of benchmarks when the chips arrive this month.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

Concerned Etsy shop owners have submitted support tickets – the company does not list a phone number or email address – and have posted the results to online forums. These responses take both sides, some saying that non-original 3D prints are fine, others saying that they are banned. After nearly a month, Etsy has yet to clarify the policy with either customers or its own staff.

Enter MyMiniFactory with a new offer that the company says can help address the blurry Etsy rules. MMF emailed Tom’s Hardware outlining a new plan in direct response to the Etsy dilemma: a Premium Merchant Program. “We believe decentralized manufacturing is the cornerstone for scaling the consumer 3D printing market, but for years, talented creators and makers have been battling with platforms that don’t properly support them,” states MMF in a blog post.

MyMiniFactory believes it can fill the gap. The 13-year-old platform is currently only known as a digital file shopping site, where people can purchase professionally designed digital models to use with their 3D printers. Models can cost anywhere from 99 cents to hundreds of dollars for large collections. A large portion of MMF files are directed at tabletop gamers, but it also has files for toys and home decor.

For $25 a month, anyone with a 3D printer can become a Verified Premium Merchant and sell physical 3D prints on MMF’s platform. The seller would receive a Verified Merchant Badge on their profile, global visibility on an MMF Merchant Map, and various discounts. MMF hopes that sellers with a Premium Merchant Profile will stand out as professionals, thus earning the trust of customers.

However, there are a few problems. First, in order for MMF to verify that sellers have proper permission to sell prints, they will need to purchase their commercial licenses through the MMF “Tribes” system. At the moment, people who want to legally sell a designer’s work can obtain their licenses or memberships through a number of sources, such as Patreon, Thangs, and Printables, which are all in competition with MMF’s services.

Licensed 3D print sellers on Cinderwing3D’s Discord are skeptical that MMF could replace sales currently made on Etsy. Love it or hate it, Etsy has established a reputation with shoppers as a marketplace for unique and handcrafted goods you can’t find anywhere else, no matter what material they're crafted from.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

Many industrial storage device makers crow about the ruggedness and durability of their devices, and Team Group is no exception with the P250Q-M80. The device has many great durability stats that you might look for.

It boasts MIL-STD 810G vibration endurance, MIL-STD-202G shock resistance, storage integrity temperatures spanning -55°C (-67°F) to +95°C (203°F), an MTBF of over three million hours, a three-year warranty, S.M.A.R.T. health monitoring, and more. However, its party trick is the rapid destruction of its precious data, its so-called “Intelligent Dual-Mode Data Destruction.”

S/W Quick Erase: hold 5-10 seconds

This is the gentler option for users who may need to clear this SSD of its contents in a hurry. S/W Quick Erase is a software method that “erases all data while retaining device functionality.” Users can choose this option if they intend to reuse the drive at a later date. It is thus “ideal for repeated use,” according to Team Group.

S/W Quick Erase is invoked by pressing the big red button that comes with the P250Q-M80 for between five and 10 seconds. It can’t be stopped. Even if power is interrupted during the data wiping process, the drive will automatically resume clearing data the next time it gets power, says Team Group.

We aren’t sure of the wiping algorithm or tech, so we would assume there remains a chance of data recovery after this software-powered wipe, or partial wipe.

H/W Quick Erase: hold 10 seconds+

The H/W Quick Erase function "uses high-voltage breakdown technology to physically destroy NAND Flash, ensuring data is irrecoverable," asserts Team Group. This patented technique is also shown in the video, where we see plumes of smoke emanate from the drive after the function is started. So, there's a visual and olfactory sign that data has been destroyed, in case you aren’t certain that you have chosen to fry the 3D TLC NAND.

Begin hardware destruction simply by holding the big red button for 10 seconds when ‘secure deletion becomes mission-critical,’ advises memory products maker.

Team Group says the H/W Quick Erase has a “precise focus on Flash IC destruction.” Therefore, if you are a soldering whizz, you may be wondering if it is possible to re-use this drive by swapping the ICs. Team Group says the NAND used is paired with a specific control chip, so reusing or repairing a H/W Quick Erased device might be more effort than it is worth.

Understatement alert

If anything, we think that the term 'H/W Quick Erase' is a little understated for the physical frying of NAND that occurs here. If we hadn’t absorbed Team Group’s PR, product pages, and video, it would be quite natural to assume H/W Quick Erase is perhaps a hardware-accelerated erase function. Several computer utilities I have used offer S/W and H/W toggles in their UIs, and none physically destroy any hardware if the latter is selected.

A ‘warning light’ of sorts on the 2280 PCB appears to flash when you are holding the big red button down. If this flashed once per second, it would be a useful guide to whether you are opting to do a quick data erase or physically destroy your NAND. Hopefully, that detail is shared in the instruction booklet.

Text shared by Team Group alongside the video we embedded also mentions that data destruction can be triggered by a “physical or event-based trigger,” which indicates there can be other ways to wipe or destroy a P250Q-M80 than the big red button.

Other ‘normal’ specs

The new P250Q-M80 offers decent speeds from its 3D TLC NAND and PCIe Gen4x4 controller. Transfers speeds of up to 7.0 GB/s reads, and 5.5 GB/s writes are claimed. There is no DRAM buffer, or mention of tech like SLC buffer or HMB.

Operating temperatures range from 0°C (32°F) ~ +70°C (158°F), though it can survive more extreme temperatures at rest. We already mentioned various other durability claims for the P250Q-M80, above.

Team Group's P250Q-M80 self-destructing SSD is being marketed in 256GB, 512GB, 1TB and 2TB capacities. We don’t have pricing and availability details at the time of writing.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

• Check out all the deals in Amazon's Prime Day Sale

The 11-inch, M3 iPad Air is going for $479. This is the lowest price we've seen on the iPad Air with M3, which has a list price of $599.

For this sale, you can get 128GB of storage, Wi-Fi connectivity (cellular boosts the price to $629), and your choice of color. The 11-inch Air uses Touch ID for authentication, boasts 12-megapixel front and rear cameras, and is compatible with Apple Pencil Pro and the Magic Keyboard.

M3 also supports Apple Intelligence features. At the moment, those aren't terribly. useful in my opinion, though it does mean you should eventually get the new Siri and anything else that comes down the pipe.

iPadOS 26, which is expected to be released this fall, should greatly improve multitasking. Beyond a "liquid glass" redesign, the big feature for iPad is vastly improved windowing, letting you arrange them as you would on, well, most other computers. It should make that Magic Keyboard accessory far more useful if you opt for it. The new OS version will also add menu bars to apps.

Apple's M-series chips have long felt overpowered in iPads, but iPadOS 26 may greatly enhance multitasking, which means you can make better use of the M3 in the iPad Air. Then again, the 11-inch form factor is pretty portable, so it'll be just fine as your go-to ebook reader, too.

We are working hard to find the best computer hardware deals for you this Amazon Prime Day. We cover the hottest deals in real-time at our Best Amazon Prime Day Deals Live page. If you're looking for more savings, check out our Amazon Prime Day deals for a range of products, or dive deeper into our specialized SSD and Storage Deals, Hard Drive Deals, Gaming Monitor Deals, Graphics Card Deals, or CPU Deals pages.

DeepSeek was founded in 2023, but it didn't attract many headlines until the global launch of iOS and Android apps in January. The company said its LLM offered numerous advantages over its Western rivals, including more efficient resource usage borne from low-level optimizations, and it didn't take long for its app to dethrone ChatGPT from the App Store's rankings.

U.S. lawmakers were quick to warn that DeepSeek represented a national security risk due to its ties to the Chinese government. Czechia is now following suit and will be limiting access to the company's wares.

The National Cyber and Information Security Agency (NÚKIB) said in a July 9 report (hat-tip The Record ) that "based on facts established during the exercise of its powers, supplemented by unclassified and classified information obtained from domestic and foreign partners, [it] concluded that the use of [DeepSeek's products] constitutes a cyber security threat."

NÚKIB's report cites various Chinese regulations, including the National Security Law, National Intelligence Law, and Counter-Espionage Law, among others, as cause for concern because they all require China-based companies to make certain information about their users available to the Chinese government regardless of where the users themselves are located.

The agency also notes that DeepSeek and its parent company High-Flyer have "deeper ties to the [Chinese] government," which has conducted numerous espionage campaigns against Czechia. It's not alone: CrowdStrike said in its latest Global Threat Report that in 2024 "China-nexus adversaries continued to operate in every sector and region across the globe, maintaining the scope of these operations while increasing their scale."

This all led NÚKIB to conclude that "given the previous espionage activities, actions against the national interests of EU Member States and NATO Allies, the specifics of PRC legislation described above and the influence of the PRC government in Chinese companies, it is highly likely that the PRC will use the capabilities of DeepSeek's products for intelligence activities."

So it's banning the use of DeepSeek's products in most circumstances. Exceptions have been carved out for security researchers as well as DeepSeek's open source models—provided they can't interact with the company's servers. That means Czech people and organizations will have to self-host the models rather than relying on DeepSeek's infrastructure.

Czechia isn't alone in this ban: NÚKIB noted that similar restrictions on DeepSeek usage have been placed in New York, Texas, and Virginia as well as Australia, Canada, Denmark, India, Italy, the Netherlands, Norway, South Korea, and Taiwan. (It's also been banned by the U.S. Navy and NASA; broader efforts to block it from the U.S. outright began in February.)

"Given the composition of this group of countries," NÚKIB said, "it is clear that the concerns about DeepSeek's products do not stem solely from a shared cultural environment or the geographical location of these countries, but they are rather a response following an objective identification and evaluation of the risk posed by these products. Moreover, it is almost certain that other countries will introduce measures against DeepSeek's products in the coming months."

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

"The acquisition of MIPS will enhance GlobalFoundries’ ability to deliver more complete and differentiated offerings to its customers," Erica McGill, director of corporate communications at GlobalFoundries, told Tom's Hardware. "By integrating MIPS' proven RISC-V processor IP and software tools, GF customers will benefit from broader access to compute IP, improved time-to-market, greater flexibility and openness, and better technologies for high-growth markets. The move will position GF as a more strategic partner, offering not just manufacturing but also foundational compute technologies that will help customers build smarter, faster and more efficient products."

MIPS will not make GF an IDM

GlobalFoundries reiterated its status as a pure-play foundry, saying the acquisition does not signal a pivot toward designing or selling chips directly. Instead, the goal is to support customers — especially those new to silicon development or seeking vertically integrated solutions — with ready-to-use IP blocks that simplify system design. Concerns that this move could turn GlobalFoundries into a competitor for its own fabless clients are unsubstantiated. 

"We remain a pure-play foundry focused on enabling our customers to build world-class products on our platforms," said the spokesperson for GlobalFoundries. "The acquisition expands our capabilities to better support fabless customers with a more comprehensive set of offerings, especially for first-time silicon builders or those targeting vertical integration." 

By offering RISC-V processor IP, GlobalFoundries acknowledges it may now compete with established IP vendors such as Andes Technology. However, it believes the ability to pair IP with differentiated manufacturing processes and secure production gives it a distinct advantage in the market.

But will enhance GF's value-added services

Foundries tend to offer various IPs to their customers, but these are relatively simple IPs, such as interfaces. Companies like Arm do offer their cores tailored for select process technologies, though. However, GF will be the first contract chipmaker to offer compute IPs based on the open-source RISC-V instruction set architecture (IPs). This will greatly enhance the value of its services to many customers among newcomers. 

The company plans to support MIPS' portfolio of processor cores built on the open RISC-V architecture, which will serve compute needs in a variety of applications, thus enhancing GF's ability to serve new markets and applications. 

"MIPS' proven processor cores will complement GF's existing IP offerings, enabling customers to build differentiated products across a wide range of markets," said McGill. "This includes compute subsystems for autonomous platforms, embedded systems, and intelligent edge devices. Following the closing, GF is committed to supporting MIPS' portfolio of open RISC-V-based instruction set architecture (ISA) processors and its broader ecosystem. Beyond edge AI applications, the combined capabilities of GF and MIPS are particularly well-suited for high-growth sectors such as automotive, internet of things, and datacenter infrastructure applications. These segments demand real-time compute, safety certification, and scalable architectures."

MIPS will remain independent

While GlobalFoundries will own MIPS and will offer its proven IP to its clients, MIPS will continue to operate independently, maintaining current customer relationships and honoring existing projects and agreements. Clients will continue working with their established account and support teams. The company stressed that there will be no disruptions to service or product availability. In addition, MIPS will continue working with other foundries. 

"In line with our strategy to operate MIPS as an open and independent IP provider, there are no plans to change its current offerings or how customers engage with the company," said the spokesperson. "MIPS customers will continue working with their existing account and support teams just as before, and all ongoing projects and agreements will carry on without interruption." 

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

• Check out all the deals in Amazon's Prime Day Sale

I have personal experience with the Anker 737, as I bought one on September 19, 2023 and have used it extensively. The power bank has a permanent spot in my travel bag for those times when I need to top off. The Anker 737 has gotten heavy use on cross-country flights, and even on the beach to charge the battery on my iPhone (which quickly drains as the display brightness jumps sky high in direct sunlight).

But this power bank isn’t just for charging smartphones; I’ve used it to charge my iPad, Apple Watch, and AirPods — sometimes simultaneously. Heck, it will even charge my MacBook Pro with ease (it’s 140W-capable using USB Power Delivery 3.1).

Anker says that a fully charged 24,000 mAh battery can recharge an empty iPhone 16 Pro four times, or fully charge an iPad Pro and still have some juice leftover. In my experience, those statements seem pretty accurate. It can charge three devices at once via one USB-A port (18W max) and two USB-C ports (140W max using a single port). One of the two USB-C ports is also used as input when recharging the Anker 737. You can also quickly replenish it from 0 to 100 percent at 140W in just 52 minutes.

However, one of the niftiest features of the power bank is its integrated OLED display. It provides input/output wattage per port, the current battery capacity level, and even the remaining time before the power bank runs out.

With a sub-$80 price tag, the Anker 737 is a well-thought-out power bank that should satisfy even the most demanding travel warriors.

We are working hard to find the best computer hardware deals for you this Amazon Prime Day. We cover the hottest deals in real-time at our Best Amazon Prime Day Deals Live page. If you're looking for more savings, check out our Amazon Prime Day deals for a range of products, or dive deeper into our specialized SSD and Storage Deals, Hard Drive Deals, Gaming Monitor Deals, Graphics Card Deals, or CPU Deals pages.

"We are excited to build on our leadership position in 3D perception in robotics and see scalable growth potential in the rise of physical AI," said Nadav Orbach, CEO of RealSense. "Our independence allows us to move faster and innovate more boldly to adapt to rapidly changing market dynamics as we lead the charge in AI innovation and the coming robotics renaissance."

RealSense develops depth-sensing cameras and embedded vision systems used in automation, access control, and robotics. The company says that its technologies are used in 60% of autonomous mobile robotics (AMRs) and humanoid robots worldwide, which is quite a good start. Presently, RealSense works with more than 3,000 clients globally and holds over 80 patents in computer vision technology. Partners of the company include ANYbotics, Eyesynth, Fit:Match, and Unitree Robotics.

RealSense says that robotics is projected to grow from $50 billion to more than $200 billion in six years, and humanoid robot demand is increasing at over 40% annually. TSMC seems to be a bit more conservative, as predictions that it cited at its recent European Technology Symposium indicate that the market for AI robots will become a $35 billion market by 2030. But anyhow, demand for computer vision systems is set to grow rapidly in the coming years, to a large degree thanks to the proliferation of edge AI. In addition, face recognition is gaining traction in public use, from transportation hubs to venues.

To meet this growing demand, RealSense is adding engineers in AI, software, and robotics. It is also expanding its sales and go-to-market teams to improve its reach in both mature and emerging markets, so $50 million from Intel and MediaTek will come in handy.

The company's latest device is the D555 camera that includes Power over Ethernet and is built on the Vision SoC V5 platform. The RealSense yet has to integrate it into high-volume devices, but the company aims to continue supporting its existing products and clients while also pushing into adjacent markets, so expect D555 in a device near you.

"Our mission is to enable the world to integrate robotics and AI in everyday life safely," said Orbach. "This technology is not about replacing human creativity or decision-making, but about removing danger and drudgery from human work. Our systems are built to amplify human potential by offloading these types of tasks to machines equipped with intelligent, secure, and reliable vision systems."

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

• Check out all the deals in Amazon's Prime Day Sale

Ugreen is a name I’m familiar with, as I have one of the company’s USB-C to 2.5 GbE adapters that I keep in my travel bag, just in case. But the retractable power bricks intrigued me for multiple reasons.

I currently use Apple 29- and 70-watt USB-C power adapters to charge my devices on the go. Of course, you can only charge one device at a time with these adapters. However, Ugreen’s power bricks allow me to charge up to three devices. In my usage case, I could imagine using the 65-watt USB-C power brick to charge my MacBook Pro, iPhone, and Apple Watch (or AirPods) simultaneously.

The second thing that intrigued me is the integrated retractable 2.3-foot USB-C cable. There’s no need to worry if I remembered to bring along a USB-C cable; it’s already attached and stows away when not needed. It doesn’t get any better than that. My track record is mostly good with retractable cables, so I’m crossing my fingers on the long-term durability here.

Finally, the overall size of these Ugreen retractable power bricks is a big draw. Compared to my 75-watt Apple adapter, these things are much smaller in comparison, due in part to the use of Gallium Nitride (GaN) power transistors. The 65-watt version has straight-through power plugs, while the 45-watt version uses retractable plugs to further reduce its profile in your bag.

The 65-watt version includes the retractable USB-C cable, one USB-C port, and one USB-A port, while the 45-watt version has the retractable USB-C cable and two USB-C ports.

With pricing as low as $19.99, Ugreen’s retractable USB-C charging blocks were an immediate impulse buy for me. I bought one of each, and they’re bound to be a helpful addition to my travel bag and will help cut down on cable/charger clutter.

We are working hard to find the best computer hardware deals for you this Amazon Prime Day. We cover the hottest deals in real-time at our Best Amazon Prime Day Deals Live page. If you're looking for more savings, check out our Amazon Prime Day deals for a range of products, or dive deeper into our specialized SSD and Storage Deals, Hard Drive Deals, Gaming Monitor Deals, Graphics Card Deals, or CPU Deals pages.

Silicon Motion's MonTitan SM8466 SSD controller for enterprise SSD will feature 16 NAND channels supporting all the upcoming types of 3D NAND memory, as revealed by Wallace C. Kou, chief executive of Silicon Motion, in an interview with Tom's Hardware published last month. The controller will be used to build drives with up to 512 TB capacity that will feature sequential read speeds of up to 28 GB/s as well as random read/write performance up to 7M 4K IOPS.

Performance of SM8466-based SSDs will by far exceed not only that of the best client SSDs, but also virtually all available enterprise-grade drives.

In terms of capabilities, the SM8466 is expected to be compliant with an NVMe 2.0+ specification, as well as the OCP NVMe SSD Spec 2.5, making it suitable for hyperscale and cloud deployments. It includes SR-IOV/MPF for virtualization, SMART monitoring, and end-to-end data protection. Security features include Secure Boot, AES-256 encryption, TCG Opal compliance, and firmware attestation, a set that offers strong safeguards for sensitive enterprise environments.

As reported, SMI plans to use TSMC's 4nm-class fabrication process to build the SM8466 controller.

The leaked slide from the SMI presentation reveals general specifications of Silicon Motion's MonTitan SM8466, but does not disclose when exactly actual drives powered by the controller will emerge. According to the chief executive of SMI, the company expects its partner to start shipping their SM8466-based products sometime in late 2026 – early 2027, which will coincide with the ramp of next-generation AI servers based on Nvidia's Rubin GPUs for AI and HPC.

In fact, SMI believes that AI will be the main application that will take advantage of PCIe Gen6 drives, as traditional storage applications will not take advantage of such drives until 2026 – 2027. As for client PCs, Kou expects SSDs with a PCIe 6.0 x4 to address PC OEMs only in 2030 or later.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

“I’d been using it for about two years, and suddenly, while gaming, the computer shut off. I thought, “What’s going on?” and pressed the power button to turn it back on. Smoke started coming out and then flames burst out — I quickly unplugged everything, turned off the power, and disassembled the PC,” said the anonymous user. They also added, “About three months ago, I was in a rush for work and called a repair shop, expecting to get ripped off. They replaced the motherboard, and the computer started to work again. Since I was in a hurry, I accepted the refurbished unit they offered — now I’m wondering if this was a motherboard issue.”

Apparently, the user had an issue with their computer and had their motherboard replaced with a refurbished one a few months before this incident. They now suspect that the GPU bursting into flames was caused by the replacement, although it’s unlikely to happen. The damage to the motherboard shows burn marks on the backside, near one of the screws that secures the heat sink on the GPU. There’s also a massive melt pattern near where the LR22 inductors used for the GPU’s voltage regulation are placed. You can see their location in 4060 Ti teardown videos, and the damage seen in the photographs indicates that the fire was likely caused by a VRM failure.

The user said that they’d sent their computer back to the repair shop to be looked at, especially since the motherboard was still in warranty. However, given that the damage was on the GPU, they’re worried that it might not be covered by that.

This is such an unfortunate and potentially dangerous incident, and one that we did not expect, given that the GPU wasn’t even a high-powered one. One commenter suggested that the screw with the most damage might have been over-torqued during manufacture, resulting in a short that led to the failure of the VRMs. However, this is pretty unlikely, especially since the GPU has been in use for a couple of years without any issues. Although the GPU did not come from an international brand like Asus or Gigabyte, comments said that Emtek was a decent South Korean GPU brand.

We hope that this is an isolated case and that the GPU manufacturer addresses the issue quickly. After all, with many gamers rocking this GPU in their systems, it’s going to be a disaster if these start going up in flames during regular use.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

The importance of ASML to the world’s semiconductor industry cannot be overstated. This Dutch tech company makes the most advanced lithography machines available, crucial to the likes of TSMC, Samsung, and Intel in striving for chip production at the finest scales – and reaping the density, performance, and efficiency advantages of advanced processes.

Russia is a relative tech minnow in the context of ASML’s pedigree, so it would naturally be interested in gaining access to some of the proprietary knowledge and technologies that are kept from the public domain. The unnamed ex-employee is said to have copied confidential files from both ASML and NXP over a period spanning several years.

We aren’t told the full extent of the intellectual property theft in the source report. It is indicated that specifics shared by the accused/imprisoned man were largely to do with designing microchip production lines.

Investigators found a “large amount of files from ASML and NXP, which were found on various data carriers at his home.” It was judged to be implausible that the ex-employee needed these files for his work, and many of the files were beyond the scope of the activities for which he was employed.

The ex-employee was also accused of ‘computer trespassing’ on systems with the sole purpose of providing technical assistance to his Russian contact. The defendant is claimed to have stated (machine translated) that “I had kept files from ASML for my own use... Whether those files were allowed to go to Russia? I did not ask myself that.” Later admitting, “Yes, I gave advice.”

Information 'of great value' to Russia

Of course, such espionage should be punishable, potentially with incarceration, at any time. But Rotterdam District Court’s judgment applied internationally agreed security laws brought in, back in 2014, when Russia invaded Crimea. From that time, it became an offense to provide advice and share technology with Russia.

The judge ruling on this case noted that, even though the information shared with the Russian contact may be old or outdated, it would still be “of great value to a country with a (much) lower level of knowledge,” like Russia. Moreover, we know Russia will not shy away from acquiring any technology, by any means, to fortify its war machine.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

• Check out all the tech deals in Amazon's Prime Day Sale

The Stream Deck will supercharge your workflow with its customizable macro pad, giving you access to endless possibilities at your fingertips. It works with various apps, including Twitch, OBS, Meld Studio, Discord, YouTube, and more, and you can even customize it to work with whatever app you’re using. You can even use Smart Profiles so that your Stream Deck will change the available buttons, depending on which app is currently open on your Mac or PC.

It’s also a great tool for gamers, as you can set your most commonly used shortcuts on the Stream Deck. That way, you don’t have to memorize the many different shortcut buttons you need to press in the heat of battle — instead, you can just program the Stream Deck depending on the game you’re running and keep everything a keypress away. So, whether you want to quickly open your inventory, cast the most convoluted spell known to the gaming world, or press up, up, down, down, left, right, left, right, B, A, you can do so in a split second with the Stream Deck.

This might seem like an unnecessary accessory for most people, as it only shaves off a few seconds from your workflow. But if you frequently use these shortcuts throughout your day, the seconds you save will eventually stack up and make you more efficient. And it’s not just for people who work with media — even programmers, writers, and the average office worker will find it useful. As for gaming, that split-second advantage you get from the Elgato Stream Deck might mean the difference between losing and winning a match.

If you’re looking to gain an advantage — whether at work or in gaming — the Stream Deck will give you that. However, you have to act fast as this sale is only for Prime Day 2025, which ends today.

We are working hard to find the best computer hardware deals for you this Amazon Prime Day. We cover the hottest deals in real-time at our Best Amazon Prime Day Deals Live page. If you're looking for more savings, check out our Amazon Prime Day deals for a range of products, or dive deeper into our specialized SSD and Storage Deals, Hard Drive Deals, Gaming Monitor Deals, Graphics Card Deals, or CPU Deals pages.

A massive cluster

70% of the processing capacity — enabled by around 80,500 of Nvidia's H100 and H200 GPUs — is expected to be concentrated in a single state-owned data center located in Yiwu County, Xinjiang. The remaining 30% will be spread across at least a few dozen additional data center projects (38 to be exact), largely in Xinjiang, outside of Yiwu and in Qinghai province. One of the larger projects is run by Nyocor, which plans to install 625 H100 DGX servers with around 5,000 H100 accelerators in multiple phases, starting with the first phase involving 250 8-way machines (2,000 H100 GPUs). These plans are based on official investment documents, tenders, and filings reviewed by Bloomberg.

To put the number of Hopper GPUs — 115,000 in total and 80,500 for the data center in Yiwu County, Xinjiang — into context, it took Elon Musk's xAI around 100,000 of H100 processors to train the Grok 3 AI model, which is one of the most advanced models currently available. DeepSeek trained its R1 model using a GPU cluster of 50,000 Nvidia Hopper GPUs, which included 30,000 H20 HGX units, 10,000 H800s, and 10,000 H100s. It is unknown what DeepSeek used to train its R2 model.

Since Chinese companies are state-controlled entities, they tend not to disclose specifications or performance of their AI clusters, making it difficult to tell how the data center with 80,500 GPUs stacks up against China's existing AI clusters. Nonetheless, if it comes to fruition, we might be dealing with one of the most powerful AI data centers in the People's Republic. So, a cluster featuring around 80,000 Hopper H100 and H200 GPUs significantly strengthens China's AI infrastructure in general. Furthermore, it can be used to train advanced large language models (LLMs) and large reasoning models (LRMs).

According to Chinese government statements cited by Bloomberg, Xinjiang has already built a data center that provides '24,000 PetaFLOPS of processing power,' which is said to be equivalent to around 12,000 Nvidia H100s, available to other cities like Chongqing. To attract investors, local officials also offer a 20% discount on electricity, along with financial and housing incentives for experts in AI and green technologies.

Tens of thousands of Hopper GPUs for China?

Bloomberg estimates that completing all 39 projects as envisioned would require the procurement of more than 14,000 servers using either H100 or H200 processors, worth billions of dollars on China's black market.

While operators of data centers in Xinjiang and Qinghai or their customers could probably use H20 HGX processors for their workloads, they clearly indicated in their permit documents to authorities what they plan to use, suggesting that they need high performance of H100 and high memory capacity of H200 for their projects. Depending on exact metrics, H100 can be 3.3 – 6.69 times faster than cut-down H20 with AI data formats and 1.52 – 64 times faster with HPC data formats. That said, even with linear performance scaling (which is not how things work when clustering AI compute hardware), one would need 380,000 – 770,000 H20s to substitute 115,000 H100 GPUs.

There are no explanations of how these parts will be sourced, but we know that there are plenty of GPU servers smuggled to China.

People with knowledge of U.S. government investigations said they were unaware of the specific Xinjiang projects, but confirmed to Bloomberg the existence of some unauthorized Nvidia hardware in China. However, they expressed doubt that any organized network could supply over 100,000 restricted processors to one country, let alone in a single region. Estimates on the total number of such chips in China vary. Two senior officials in the Biden administration mentioned a figure closer to 25,000, far less than what the Chinese projects require.

To date, there is no direct proof that China has accumulated, or will soon receive, the more than 115,000 restricted GPUs outlined in these construction plans. Still, work on the facilities continues. In Yiwu, where most of the activity is taking place, a large solar power tower has been erected to provide consistent electricity. The location was chosen for its access to solar and wind energy, inexpensive land, and high elevation, which helps to cool hardware.

When it comes to the availability of H100 and H200 GPUs, there are millions of Hopper parts installed across the world, and at least some data centers that use them will be switching them out for higher-performance Blackwell accelerators. The elephant in the room, however, is that there will be plenty of decommissioned H100 GPUs in the coming months or quarters. Procuring them and smuggling them to China is something that looks reasonable to do for companies from the People's Republic. But, there is no evidence that they have attempted to do this so far.

Nvidia has repeatedly stressed that there is no indication of large-scale diversion of its GPUs to China from other countries, such as Singapore. However, smuggling through Southeast Asia has become a point of concern, especially in Malaysia and Singapore. Singapore is currently prosecuting individuals for allegedly exporting AI servers containing restricted components to Malaysia, from where they might end up in China. U.S. officials have asked Malaysian authorities to take action against unauthorized technology transfers, and Malaysia has stated it will act if presented with solid evidence.

Jeffrey Kessler of the U.S. Commerce Department’s Bureau of Industry and Security recently told lawmakers that unauthorized transfers of AI GPUs to China are indeed occurring. So while Nvidia is not suspected of any misconduct, authorities in Washington are investigating how banned GPUs might be entering China, reports Bloomberg.

Xinjiang, particularly the Hami region, is already a hub for renewable power. The local government has been actively promoting the development of computing infrastructure using its vast energy reserves.

There is a catch

Nvidia has commented on the issue to Bloomberg, saying that building functioning AI infrastructure with unofficial and/or used parts is both risky and impractical. The company also stated it provides no operational or technical support for restricted products in China or elsewhere. While lack of support may not be a problem for small deployments used for research purposes, large machines involving thousands or tens of thousands of GPUs are usually clustered together with Nvidia specialists to maximize their performance and efficiency.

However, as Chinese entities do not have a choice, they might attempt to build large Nvidia-based clusters without working with the company, even at the cost of performance and efficiency.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

These con artists plan to trick Facebook users out of their hard-earned money by posing as Bambu Lab with fake Facebook pages offering links to artificially sweet deals. The fake pages usually misspell or truncate Bambu Lab’s name and can lead to copycat websites that look incredibly real. The point of the scam is to trick people into buying 3D printers and filament that will never appear.

After YouTuber Angus Deveson (Maker’s Muse) issued a warning about the scam, Bambu Lab took to Facebook to explain that it has been taking legal action against scammers for months.

“We’ve noticed an increase in fake websites and Facebook ads impersonating Bambu Lab. These scams trick users into paying for products they won’t receive, often through look-alike domains or fake discounts,” said Bambu Lab.

The process of reporting fake ads to Facebook can be time-consuming, and companies need your help. Scammers can easily trick Facebook into putting up fake sites with page names and urls that are almost correct. Facebook has billions of users and is one of the most visited websites on the internet, and it relies on users to report false pages and ads to trigger human review.

But recognizing a scam site takes a sharp eye and a deeper knowledge of Bambu Lab than many new to the hobby simply don’t have. For example, did you know that Bambu Lab would never call itself “Bambu”? That nickname is exclusively used by fans. Bambu Lab also never uses grungy photos from a warehouse or a print farm. Real images from Bambu Lab always show 3D printers in perfect condition, placed in impeccably styled settings.

It’s helpful to know Bambu Lab’s real web address, which is BambuLab.com. Simply typing “Bambu Lab” into your search engine will get you to the real site. This “blmake.pro” came from a scammy Facebook ad.

Scammers have no problem scraping a website and putting up a duplicate. This portion of a knock off website even used our Tom’s Hardware Editor’s Choice award on the Bambu Lab P1P. However, these ridiculously low prices should be immediate red flags.

It can be hard to verify if a Facebook page is real. However, if you click the page name and check under “About” you can see a listing for “Page transparency.” This will let you know when the page was created. Another red flag is the fact that a three-year-old company with the most popular 3D printer for beginners only has 3 followers?

What can you do if you encounter a scam? Report the ad directly on Facebook by clicking the menu option and selecting "Report ad," then choosing "Scam, fraud, or false information". If you’re not sure if the ad is a scam, simply look up the manufacturer’s website on your browser. Sales advertised on Facebook should be easy to find on the website.

Below is a real ad by the real Bambu Lab. This is pulled from my personal Facebook page, so it is showing me a few friends that also like Bambu Lab at the top, it uses the proper “Bambu Lab 3D” page name, the correct URL for the company website and uses clean, idealized images of the printers. The $219 starting price is for the A1 Mini without an AMS, which is a pretty good deal, but not over the top.

If you're looking for the real sale, we have it here on our site: here's our coverage of Bambu Lab’s current 3rd Anniversary sale.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

• Check out gaming laptop deals at Best Buy

We're actively testing the 16 Aurora in our labs now, so we don't have numbers to share at this time. The highlight is that you can get a 5060 for $1,100, as these systems often start at far higher prices. Even some RTX 5050 systems are starting at or above $1,000. So, this Aurora, which also features 32GB of RAM, a 1TB SSD, and a 120 Hz, 2560 x 1600 screen, is worth calling out at this price.

This system is sleek, and it features a "stealth mode" that reduces RGB, silences fans, and acts a whole lot more like a standard productivity laptop. It comes in a navy blue color that stands out, but doesn't scream that it's a gaming machine (well, outside of the alien head logo).

If you don't want to dive in before we go hands-on, I don't blame you; Our review should be live in the coming weeks. But if you're the type of person who is willing to purchase off a spec sheet, the price for these parts is pretty good, especially for a name brand.

For more laptops in this price range, check out our list of the best gaming laptops under $1,500.

We are working hard to find the best computer hardware deals for you this Amazon Prime Day. We cover the hottest deals in real-time at our Best Amazon Prime Day Deals Live page. If you're looking for more savings, check out our Amazon Prime Day deals for a range of products, or dive deeper into our specialized SSD and Storage Deals, Hard Drive Deals, Gaming Monitor Deals, Graphics Card Deals, or CPU Deals pages.

• Check out all the gaming laptop deals in Amazon's Prime Day Sale

The biggest advantage the Lenovo Legion Go has over its competitors (primarily the Asus Ally ROG X) is that it has a larger 8.8-inch screen display, making it an ideal alternative for those who prioritize immersion over portability. Aside from that, it comes with two USB-C ports, allowing you to charge the device and connect to an external display simultaneously. Its detachable controllers (one of which converts into a mouse for FPS gaming) mean that you don’t need to bring an additional controller with you for large-screen gaming, making it the perfect device if you like playing your favorite titles on your hotel room TV after a long day sightseeing.

If you want more graphics performance, both USB-C ports are USB4 and are compatible with Thunderbolt 3/4 enclosures. You can also get the Lenovo Legion Go bundled with a USB-C hub if you want to turn it into a gaming PC or workstation when you’re at home.

Although not the latest and greatest in gaming handhelds, the Lenovo Legion Go is a pretty compelling offer for just $499. It can be a bit more affordable, but the last time it hit its all-time low price was months ago — around December last year. So, if you’re not willing to wait for the holiday season to get your hands on a decent gaming console, now is the time to get the Legion Go.

We are working hard to find the best computer hardware deals for you this Amazon Prime Day. We cover the hottest deals in real-time at our Best Amazon Prime Day Deals Live page. If you're looking for more savings, check out our Amazon Prime Day deals for a range of products, or dive deeper into our specialized SSD and Storage Deals, Hard Drive Deals, Gaming Monitor Deals, Graphics Card Deals, or CPU Deals pages.

The group says that it improved LPDDR6 performance by increasing the number of sub-channels and decreasing their size. DDR5 splits the 64-bit channel used by DDR4 into two independent 32-bit sub-channels, and LPDDR6 improves this by using four 24-bit sub-channels, resulting in lower latency and higher concurrency. The group has also lowered the voltage requirement for the memory and deployed additional power-saving features like Dynamic Voltage Frequency Scaling for Low Power (DVFSL), which lowers power demand during low-frequency operation. The group says that LPDDR6 will have a data rate between 10,667 and 14,400 MT/s, which translates to an effective bandwidth of around 28.5 to 38.4 GB/s. This makes the standard’s transfer rate much faster than the current DDR5-12054 overclocking record.

Since JEDEC is a widely recognized standards organization and has active members from the biggest companies across the globe, we expect LPDDR6 to be adopted quickly. Several companies have already expressed support for this new specification, including chip design corporations Cadence and Synopsis, testing companies Advantest Corporation and Keysight Technologies, several chip manufacturers like MediaTek, Qualcomm, and Samsung, and notable memory makers like Micron and SK hynix. LPDDR6 is primarily designed for laptops and other mobile devices, though. JEDEC has yet to announce the finalized specifications for DDR6, which is primarily aimed at desktops, although the group said that it aims to publish them this year.

It took about a year for DDR5 modules to arrive on the market after it was announced in 2020, and we expect the same with LPDDR6 (and DDR6, if it gets announced this year). This is also about right, as the biggest memory manufacturers have announced the discontinuation of DDR4 production in 2025. However, there is still demand for this older standard, especially among legacy and cheaper devices, so much so that smaller vendors are extending their production.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

According to OregonTech, it's borderline a fight for survival for the once-great American innovation powerhouse as it struggles to even acknowledge being among the top contenders anymore. Despite Tan's insistence, Intel would still rank fairly well given its extensive legacy. While companies like AMD, Nvidia, Apple, TSMC, and even Samsung might be more successful today, smaller chipmakers like Broadcom, MediaTek, Micron, and SK Hynix are not above the Blue Team in terms of sheer impact. Regardless, talking to employees around the world in a Q&A session, Intel's CEO allegedly shared these bleak words: "Twenty, 30 years ago, we were really the leader. Now I think the world has changed. We are not in the top 10 semiconductor companies."

As evident from the quote, this is a far cry from a few decades ago when Intel essentially held a monopoly over the CPU market, making barely perceptible upgrades each generation in order to sustain its dominance. At one time, Intel was so powerful that it considered acquiring Nvidia for $20 billion. The GPU maker is now worth $4 trillion.

It never saw AMD as an honorable competitor until it was too late, and Ryzen pulled the carpet from underneath the Blue Team's feet. Now, more people choose to build an AMD system than ever before. Not only that, but AMD also powers your favorite handhelds like the Steam Deck and Rog Ally X, alongside the biggest consoles: Xbox Series and PlayStation 5. AMD works closely with TSMC, another one of Intel's competitors, as the company makes its own chips in-house.

This vertical alignment was once a core strength for the firm, but it has turned into more of a liability these days. Faltering nodes that can't quite match the prowess of Taiwan have arguably held back Intel's processors from reaching their full potential. In fact, starting in 2023, the company tasked TSMC with manufacturing the GPU tile on its Meteor Lake chips. This partnership extended to TSMC, essentially making the entire compute tile for Lunar Lake—and now, in 2025, roughly 30% of fabrication has been outsourced to TSMC. A long-overdue admission of failure that could've been prevented had Intel been allowed to make its leading-edge CPUs with external manufacturing in mind from the start. Ultimately its own foundry was the limiting factor.

As such, Intel has been laying off thousands across the world in a bid to cut costs. Costs have skyrocketed due to high R&D spending for future nodes, and the company faces a $16 billion loss in Q3 last year. Intel's resurrection has to be a "marathon," said Tan, as he hopes to turn around the company culture and "be humble" in listening to shifting demands of the industry. Intel wants to be more like AMD and NVIDIA, who are faster, meaner, and more ruthless competitors these days, especially with the advent of AI. Of course, artificial intelligence has been around for a while, but it wasn't until OpenAI's ChatGPT that a second big bang occurred, ushering in a new era of machine learning. An era almost entirely powered by Nvidia's data center GPUs, highlighting another sector where Intel failed to capitalize on its position.

"On training, I think it is too late for us," Lip-Bu Tan remarked. Intel instead plans to shift its focus toward edge AI, aiming to bring AI processing directly to devices like PCs rather than relying on cloud-based compute. Tan also highlighted agentic AI—an emerging field where AI systems can act autonomously without constant human input—as a key growth area. He expressed optimism that recent high-level hires could help steer Intel back into relevance in AI, hinting that more talent acquisitions are on the way. “Stay tuned. A few more people are coming on board,” said Tan. At this point, Nvidia is simply too far ahead to catch up to, so it's almost exciting to see Intel change gears and look to close the gap in a different way.

That being said, Intel now lags behind in data center CPUs, too, where AMD's EPYC lineup has overtaken them in the past year, further dwindling the company's confidence. Additionally, last year, Intel's board forced former CEO Pat Gelsinger out of the company and replaced him with Lip-Bu Tan, who appears to have a distinctly different, more streamlined vision for the company. Instead of focusing on several different facets, such as CPU, GPU, foundry, and more, at once, Lip wants to hone in on what the company can do well at one time.

This development follows long-standing rumors of Intel splitting in two and forming a new foundry division that would act as an independent subsidiary, turning the main Intel into a fabless chipmaker. Both AMD and Apple, Intel's rivals in the CPU market, operate like this, and Nvidia has also always used TSMC or Samsung to build its graphics cards. It would be interesting to see the Blue Team shed off weight and move like a free animal in the biome. However, it's too early to speculate given that 18A, Intel's proposed savior, is still a year away.