A new GitHub repository dubbed ram-altman features a script to monitor your PC’s RAM usage and report back to the eponymous OpenAI boss. Sam Altman infamously described people having spare, underutilized DRAM in computers at home as “quite terrible” in a recent podcast. Thus, developer Iain Kay was inspired to write a bash script that emails the OpenAI boss once a day to report on any RAM that is underutilized. Clearly this is not a serious tool, or rather “The memory accounting is. The correspondence isn't,” says Kay.

ram-altman leans on the MemAvailable query, which returns the kernel's own estimate of how much memory programs can claim without swapping, explains the developer of this script. When installed, it checks every five minutes and emails Altman once a day if you are found to be wasting this precious resource.

The emails take the .CSV file logs of your RAM use throughout the day as the data source to compose a summary email. These script-generated messages rotate through eight subject lines, 10 humorous message boilerplates, and six signoffs. Sample emails can be seen on the ram-altman.lol website. Altman’s email address isn’t supplied as part of the script.

If your machine is using 90% or more of its RAM, then you won’t be reported to the OpenAI boss for wasting memory. Also, if you don’t want to disturb Altman, you can run the tests without emailing anyone with the terminal command: bash ram-altman.sh --dry-run

The developer isn’t too scathing of the OpenAI boss. Kay quotes Altman’s explanations that his remarks were taken out of context. Moreover, the dev doffs his cap to the artificial intelligence leader’s humorous display name change to ‘Ram Altman’ on X last week. That’s where the name for this script came from.

ram-altman is a free bash script available on GitHub. It runs natively on Linux. On Windows, you can use WSL to run it, but someone has ported it for PowerShell too. It won’t run on Mac, but Altman and his minions would find it inconvenient, at best, to remove your soldered-down Apple RAM.

Huawei has introduced its first desktop PCs powered by its own 10-core processor as well as proprietary HarmonyOS operating system, reports the Southern China Morning Post. The company claims that the CPU and the OS are built entirely of using 'components manufactured' in China, which matters when it comes national security and domestic supply chains inside the sanctioned country. But whether these machines are truly 'entirely domestic' is an open question.

The Huawei Qinyun HM650 and HM650g desktop PCs come with 16 GB of LPDDR5 memory (probably made by China's CXMT) and either a 256GB or 512GB SSD (probably made by Chinese flash maker YMTC) are powered by the company's Kirin X90 system-on-chip, which originally debuted in Huawei's consumer systems in 2025. The key ingedient of the system it its processor, which features 10 Huawei-designed Armv8-based Taishan CPU cores in a 4+4+2 configuration, with all cores supporting simultaneous multithreading (SMT) to execute up to 20 threads simultameusly. The chip also integrates Huawei's Maleoon graphics processing unit (Arm Mali-based) and a dedicated neural processing unit for AI workloads.

Although Huawei doesn't officially disclose the Kirin X90's manufacturing process, a teardown analysis by TechInsights confirmed that the CPU is manufactured by SMIC using its 7nm-class N+2 process technology. This makes the X90 an example of Huawei's ability to develop PC-class processors using domestically available semiconductor manufacturing capabilities, albeit on a process technology that trails leading-edge nodes from Intel, TSMC, and Samsung.

The most interesting aspect of the Qinyun HM650 series is certainly Huawei's claim that all of its new system's components, including the processor, memory, storage, power supply, and wireless connectivity are manufactured domestically. Such certification itself does not necessarily mean that the entire semiconductor supply chain behind the machine is Chinese. While the CPU, memory, and storage chips are fabricated domestically, they are still made using foreign manufacturing equipment (from the Netherlands, U.S., and Japan), materials (from Japan), intellectual property (from the UK), and electronic design automation (EDA software from the U.S. and partly from China) tools.

Meanwhile, the claim about the operating system developed in China sounds ambitious. The machines run Huawei's HarmonyOS Enterprise Edition, which certainly reduces their dependence on foreign technology suppliers. Unlike conventional PCs that rely on Microsoft's Windows operating system, Huawei's desktops use a software platform developed and maintained by a Chinese company. Furthermore, unlike earlier HarmonyOS versions that incorporated Linux components, the HarmonyOS Enterprise Edition uses Huawei's own kernel and software architecture. This indeed gives Huawei greater control over operating system development, security updates, and integration with its hardware, although compatibility with existing Windows applications could remain a limitation for enterprise customers that rely on software by multinational corporations for Windows.

Keeping all these things in mind, can we say that Huawei's Qinyun HM650 series relies entirely on components developed and made in China? Probably yes. Yet, with major caveats, including a semiconductor supply chain that relies on tools and software designed and produced elsewhere. Furthermore, incompatibility of the OS with software designed for Windows could be a limitation for some customers.

Put simply, AMD dominates the gaming CPU world right now. While the AMD Ryzen 7 9800X3D isn't the most powerful, it's still the best option for gamers, offering a good performance-to-power ratio. It's simply a powerhouse, and right now, it's down to a new record low price of $399.99 at multiple retailers. Considering the performance you can get with a 9800X3D, you'll want to grab it if you've got an AM5 build.

● Check out this deal on Amazon

● Check out this deal on Newegg

Both Newegg and Amazon have it at this record price, but you'll need to apply some coupons to get the price. At Newegg, that means using the coupon code BFF36 at checkout to knock $70 off its current list price, while at Amazon, you'll need to click or tap to apply the $49.01 coupon, shown below the price.

There aren't many options for gaming that beat the AMD Ryzen 7 9800X3D. It's a powerhouse that was, until recently, the best CPU for gaming in the world, beating every other chip due to its boosted L3 cache capacity. You're getting an eight-core, 16-thread processor, made using AMD's current-gen Zen 5 architecture, with a boost clock speed of 5.2 GHz, and a 120W TDP.

The 9800X3D is fully unlocked, so it's possible to overclock it, which was a limitation in older 3D V-cache chips from AMD. It's that extra L3 cache that makes this such a special chip, too, stacked beneath the CPU cores closer to the heat spreader on a reconfigured die.

Faster chips do technically exist, like the AMD Ryzen 7 9850X3D, which our CPU benchmark data shows is 3% faster at 1080p than the 9800X3D, although with much higher power demands. The 9800X3D offers the best performance per watt of any other gaming CPU, so you can really maximise performance without a huge increase in electrical power. You can see, too, from the Bench charts below that the 9800X3D sits very close to the top as far as CPU performance goes.

Put simply, if you pair this processor with a recent GPU, you won't see many bottlenecks. Pair it with a top-class option like the RTX 5080 or RTX 5090, and you'll be hitting high frame rates in any game you can throw at it. CPU-intensive games like Cyberpunk 2077 won't be let down by that 96MB of L3 cache included in this X3D chip.

Given how good a processor this is, and how little Intel has been able to rival it, we can't see it falling out of date for a long while yet. It's sitting at the top of the food chain and will continue to dominate games for years into the future. AMD's AM5 platform, meanwhile, remains in use, and AMD has promised to support it for at least one more generation.

If you want even better performance, though, then the 9850X3D (which, as we mentioned earlier, does manage to beat the 9800X3D by 3%) is also on sale. You can pick up the 9850X3D for $419 at Newegg right now, thanks to a $65 promo code, a significant discount on its launch price. You'll need to apply the BFF39 code at checkout at Newegg to get the discount, or apply the on-screen code if you buy it at Amazon.

The 9800X3D and 9850X3D are formidable options for gaming, the 1-2 winners that sit above all others. The record-low $399 sale price for the 9800X3D is simply exceptional, while $419 for the 9850X3D is an $80 drop compared to its launch price. Either option would be a serious upgrade for an AM5 gaming PC, or as the center point of a new build.

Your sleep data might be doing double duty. While wearable headbands promise to help users understand their sleeping patterns, some of the companies behind them are also turning the electrical signals collected from their customers' brains into training material for artificial intelligence models. Science recently conducted an investigation alongside Pulitzer Center's AI Accountability Network and found that the customer neurology companies have accumulated tens of millions of hours of electroencephalography (EEG) recordings — a rather technical way of saying brain-wave data. As per the findings, several are using those recordings to develop so-called brain foundation models, despite initially failing to spell out that use in their privacy policies.

Interaxon, the Canadian firm behind the Muse sleep and meditation headband, is one of them. Its devices monitor electrical activity in the brain, alongside other physiological measurements, while users sleep. The recordings are processed into reports that help customers track their rest and identify changes in their sleeping patterns. But there is another use for that information. Interaxon has already trained an AI model using about 50,000 hours of EEG recording and says it removes identifying information, including names, email, and addresses, before using the data. “We aim to use as much of that eligible corpus as possible,” Nadia Kumentas, its Chief Marketing Officer, said.

That was not made particularly clear to customers, though. Prior to talking to Seince, Interaxon's privacy policy did not explicitly disclose that users' brain recordings could be used to train foundation models. The company has, however, updated the same with Kumentas acknowledging, "We agree users should be explicitly informed." For customers whose recordings had already been used, the classification came late, which likely exposes the company to potential lawsuits. Privacy, it seems, remains a major issue. As per Science, more than two-thirds of the companies studied explicitly allowed corporate transfers of data without user consent. That could be significant if a small neurotechnology firm is acquired, potentially putting its accumulated recordings in the hands of a new owner. Colorado and California, however, have amended their consumer privacy laws to cover neural data, although other states lag behind.

Interaxon isn’t the only company pushing this idea. Emotiv, another wearable EEG maker, is training its Large Brain model partly on recordings from customers using its high-end devices. It also updated its privacy policy following questions by Science, disclosing the use of de-identified neural data for model training subject to opt-in consent.

Notably, the idea behind these systems is particularly different from training a large language model. Instead of feeding AI an enormous amount of text, developers supply it with recordings of electrical activity in the brain. The model then learns recurring patterns that can be adapted to recognize different brain states or assist with specific research tasks. And there are already some interesting results. AI startup Piramidal has trained its model on more than 4 million hours of EEG data in partnership with Cleveland Clinic. The company developed a seizure-detection system that correctly identified nine out of 10 seizures in a pilot study after being fine-tuned on the clinic's recordings. That, however, doesn’t mean consumer headbands can suddenly read people's thoughts. Most wearable EEG devices use just one to six electrodes, considerably fewer than 19 to 25 typically found in clinical EEG systems. Also, their recordings are generally less comprehensive.

That said, the privacy implications of wearable brain technology aren't entirely new. In February, an engineer discovered he could access other people's brainwave recordings through a smart sleep mask he had purchased. He reportedly investigated the device after running into connection problems. By examining its Android application, he discovered hardcoded credentials that appeared to be shared across devices, which he later found were EEG readings belonging to other users.

The industry has also been gathering neural recordings on a considerable scale. Last December, we covered Coundit’s effort to collect 10,000 hours of neural data from thousands of patients. The San Francisco startup built its own headset system to capture brain activity alongside speech and text, with the aim of developing AI capable of translating signals into language. Meta has also been working on BrainQwerty, a non-invasive brain-computer interface that translates brain activity into keyboard inputs. Its second-generation system improved average word accuracy to 61% from 40% in the earlier version, although the technology remains a far cry from practical everyday use.

An SSD can spend most of its day doing very little, particularly when a PC is sitting idle. That may not be the case for some AMD Radeon systems, however, with reports suggesting that AMD’s External Events Utility (EEU), which is installed alongside Radeon drivers, may be writing data to the system driver continuously, even when the computer is barely being used. The issue drew attention again on November 8, when Reddit user ‘darklinkpower’ highlighted reports of the unusual drive activity.

One Radeon owner claimed the problem had been quietly eating into their drive's reported health for more than a year, writing: "I just found out that the driver eats up 1% of my SSD health every month for 16 months." Another user reported sustained writes of approximately 5MB/s while their PC was idle. Another one claimed their SSD health had dropped by about 1% every month for 16 months. AMD has acknowledged the problem. In an October 21 response on its community forum, a company representative confirmed that they were investigating. “Yes. We're looking into it with the aim to resolve it in a future driver release." He said. Another representative, AMAD Vik, subsequently offered a more encouraging update on Reddit, apologizing and saying, “This is hopefully fixed in the next one."

Notably, the EEU is not an entirely unfamiliar culprit. Last November, a similar issue involving AMD chipset drivers arose, with a Redditor claiming he had discovered that the software repeatedly updated a log file whenever on-screen elements were moved or resized. That incident also prompted questions about whether stopping the service would affect FreeSync. The latest reports are a different proposition, though, with some users now seeing writes that barely let up, even when their PCs are doing nothing.

The numbers are worth looking at. At 5 MB/s, an SSD would receive about 18GB of host writes per hour. Over a full day, that's 432GB, and a month of uninterrupted activity could push the figures close to 13TB. That, however, does not necessarily mean being written to the NAND flash itself. SSD controllers use caching and wear-leveling, and other techniques, to manage incoming data. How much of that reported activity thus translates into actual flash wear is still unknown. The user's claim of losing 1% of drive health each month also hasn't been independently confirmed.

Moreover, there doesn't appear to be clear patterns across Radeon driver versions. Reports include systems running Adrenalin 26.19.2, as well as older releases, suggesting the problem may not be limited to the latest update. AMD's October 30 release notes for Adrenalin 26.20.2 don't mention the issue either. For now, it's difficult to say which driver versions, hardware combinations, or system configurations are responsible.

That said, there are two reported workarounds. The first involves opening the Windows Services application by pressing Win + R, entering services.msc, and locating AMD EEU. Users can then stop the service or change its startup type to Disabled. That appears to stop excessive writeups, but isn’t necessarily a clean solution. The second workaround involves disabling Resizable Bar (ReBAR), which another user reported had stopped the disk activity without disabling the EEU. This, however, isn’t ideal either, as ReBAR can improve gaming performance by allowing the PC to access the GPU’S memory more efficiently.

For now, checking whether the problem actually exists on your PC is probably the most sensible first step. Open Windows Resource Monitor by pressing Win + R, typing resmon, and selecting the Disk tab. Sort by write activity and look for sustained writes from atieclxx.exe or atiesrxx.exe. AMD has indicated that a fix is in the works, but hasn’t published a confirmed release date. Until then, Radeon owners may want to monitor their storage activity rather than immediately disabling driver services or changing BIOS settings.

Given how expensive PC gaming is becoming, picking up a pre-built PC with a discount is the best way to go. Right now, you can buy an ABS Flux II Aqua gaming PC with an Nvidia GeForce RTX 5070 for $1,999.99 at Newegg, saving you $300 compared to its list price.

● Check out this deal at Newegg

Newegg owns the ABS brand, so this isn't a no-brand option you're buying here. One oddity, though, is that the advertisement suggests this PC comes with an Intel Core Ultra 7 260K. That's not a CPU that actually exists, but given the specs, it probably refers to the Intel Core Ultra 7 265K, although we can't confirm that. If it is, though, that's a 20-core CPU and, now, given its current pricing, is a good option for an Intel pre-built, offering really solid gaming performance, as our 265K vs 285K faceoff shows.

It's the RTX 5070 that's the real star of this show, though. It's a solid mid-tier option for gaming that'll easily push through 1080p on Ultra graphics presets with high frame rates. 1440p will be an easy match here, too, and you've got potential to push up to 4K, especially with Nvidia DLSS and multi-frame generation helping to bridge the gap. There's also 32GB of DDR5 RAM in this build, along with a 1TB SSD.

Our RTX 5070 review makes clear what a good option the RTX 5060 currently is. This is the best all-rounder, with solid 1440p performance, and placing Nvidia DLSS features like multi-frame generation at your disposal to give you the best chance in even the most demanding games. Specs-wise, it has 6,144 CUDA cores and 12GB of GDDR6 VRAM. The Bench charts below show how well it ranks compared to its rivals, too.

Elsewhere, you're getting 32GB of G.Skill Flare X5 RAM, split dual-channel across two 16GB modules, with speeds of 6,000 MT/s. No DDR4 memory here, so you don't have to worry about slower speeds or longevity issues. You're also getting a 1TB MSI M461 SSD here, offering Gen 4 speeds. That'll be enough for a few big game installations alongside Windows, for instance.

All in all, the $1,999.99 sale price for this ABS Flux II Aqua gaming PC makes it a really affordable option for a 4K-capable gaming PC in the current market. You'll struggle to find a rig with a similar spec sheet for less, and with prices continuing to rise, we'd expect a build like this to sell out quickly.

When shopping for an OLED TV, most prospective owners are concerned about burn-in, a problem that's been mostly mitigated, save perhaps for scenarios that involve leaving a TV on 24/7. This has been demonstrated by RTing's famous long-running test, but the site has found a new problem that seemingly pertains to pre-2023 LG WOLED panels: edge rot, where pixels progressively go dark around the edges, similar to a mold pattern.

RTings says that 14 out of the 16 LG televisions in its burn-in test have developed this problem in varying degrees, so 87.5% of a small test sample are affected. Those sets were all made between 2020 and 2022. The site says that several staff members saw the issue pop up on their personal televisions. Some searches indicate similar reports from users everywhere, and coincidentally, I learned about this just last week as a close friend sent photos of his 2019 LG TV.

The site says that the television sets bearing Samsung QDOLED panels in its burn-in test do not exhibit these symptoms, though it notes that it only has three such units in its test run. Additionally, since the test ran between 2022 and 2025, RTings says it can't assess whether LG's newer-generation WOLED panels have this issue, and notes that the manufacturing process for said panels reportedly improved in 2024.

Needless to say, most of the television sets in the study are going to be out of warranty. LG does have five-year panel warranty coverage, but only since 2021, and only covering higher-end units like the G, Z, and M series. It's still worth looking into this, in case you own a set from those lines and within the covered timeframe.

RTings asked for statements on the matter from both LG and Samsung, and interestingly, both firms discussed moisture directly or indirectly. LG stated that the issue "may occur in a very small number of OLED products under prolonged exposure to high-temperature, high-humidity, or other harsh environmental conditions."

It's not likely that RTing's test environment is especially humid, though, and the flurry of user reports with TVs sitting in their own living rooms doesn't lend much credence to LG's argument.

Meanwhile, Samsung is more specific, saying that it's implemented its proprietary Thin Film Encapsulation (TFE), a method for blocking ingress at the panel's edge. The firm says TFE "[eliminates] the root cause of border degradation and [ensures] long-term edge-to-edge uniformity where traditional structures falter."

RTings says that according to conversations with industry experts, "contamination during manufacturing, defects in the encapsulation of some OLED displays, or degradation of that encapsulation over time" may be at the source of the problem. The site's investigation also notes that some of its test units are affected by edge delamination.

The publication also believes the pixel-death problem may be related to power-on hours, while edge delamination could be dependent on the unit's age. It's unlikely that all user reports come in from people leaving their TVs on 24/7 or close to it, though. A sample size of one means little, but the TV is used sparingly at my friend's home, only around 30 hours per week — over nearly seven years, he's only seen the pixel refresh function come up twice.

For whatever it's worth, he notes that his set's dark spots mostly reside on the side closest to a portion of the wall with very minor humidity ingress from outside the building. The set is resting on furniture, not wall-mounted, though, so the relation between humidity and the edge rot is inconclusive at best.

A hobbyist building on open-source driver for DisplayLink hardware received a cease-and-desist from Synaptics, resulting in a lengthy written argument and a threat to sue from the company. According to Phoronix, the Vino driver was “clean-room reverse-engineered” and partly developed using AI and written in Rust. However, even though the person behind the open-source driver, Mike Lothian, did not break the copyright of the code behind the driver, Synaptics argued that his work violates its patent on the DisplayLink technology.

The story began in July 2026, when Lothian reached out to Łukasz Spintzyk, a software engineer working at Synaptics, to help solve a few bugs he’s been encountering. However, Michal Lukaszek, the Strategic Engineering Partnerships Lead at the company, replied a couple of weeks later, saying that the company is concerned that Lothian is attempting to implement the DisplayLink protocol itself, which is a violation of its patents. This resulted in a cease-and-desist letter from Synaptics’ legal department asking him to take down the Vino GitHub repo, among other things.

Synaptics claims that Lothian’s work infringes on two patents: GB2484736 and GB2485613. The former, called Image generation, is about sending screen updates over a general-purpose data network like USB or Ethernet using a specific architecture and compression method, while the latter, called Display control device, details hardware that works with compressed data and only decompresses it as needed. Lothian responded, asking for complete details as to how his work infringes on its patents, but received no response, prompting him to follow up three weeks later.

This was when Synaptics decided to escalate, telling him that it would be billing Lothian “to recoup the cost and effort required to create such documentation of your infringing acts.” Andrew Bateman, the company’s Assistant General Counsel, also added, “By open sourcing our DisplayLink technology for whatever personal reasons you may have as a ‘hobbyist,’ you will be single-handedly eviscerating DisplayLink's competitive advantage in this crowded market by educating our competitors and other third parties as to how we implement our proprietary technology, which we have gone to great lengths and great expense to prevent.”

“If you'd like to notify the media or otherwise publicize our response, I encourage you to do so. We welcome any and all free publicity to let the world (in particular, our competitors and our shareholders) know that Synaptics aggressively defends and protects its DisplayLink IP from all threats of infringement.”

Lothian eventually took the GitHub repo containing the Vino driver offline, but it seems that the fight isn’t over yet. The hobbyist claims that Synaptics’ patents are invalid, saying, “In both cases the independent claims as filed contained no network or transport limitation. The claims were granted only after being amended to add delivery of the display data over a ‘general-purpose data network’ (and, for GB2485613, a ‘transport protocol’ with an ‘unpackaging’ step). That added feature was not new.”

He further argues that even though the company says that DisplayLink isn’t part of the Open Invention Network, Synaptics itself is an OIN licensee. Some say that an OIN license requires that any subsidiary acquired by the original licensee and all its patents would also fall under the company’s OIN license, although Synaptics apparently disagrees with this.

Despite the escalation, Lothian says that he remains open to discussing a sensible resolution. But unless the two parties come to an agreement, it looks like they’re getting ready for a legal face-off. You can see the complete email thread between Lothian and Synaptics in this Linux Kernel Mailing List page and the attached email thread (PDF).

Lothian said that he used a Claude subscription to help out with his cleanroom reverse-engineering efforts, revealing the headache that advanced AI tools pose for tech giants. A user has created a free Adobe Creative Suite clone in Rust using Claude, and was also reportedly working on an open-source office suite and Autodesk tools. Aside from that, we’ve also seen a lot of reverse-engineered games running on platforms they’re not designed for, forcing Anthropic to enforce some limitations on Opus 5.5. There was also concern that the widespread use of AI in chipmaking could result in the inadvertent spread of a copied design or an infringed patent across thousands of chips before anyone catches it.

A security researcher has revealed how they were forced to update the time on their car infotainment system using reverse engineering techniques. Of course, invasive tech surgery shouldn’t be needed for this kind of basic routine daylight saving time tweaking task. However, Rasmus Moorats’ 2014 Renault Latitude wouldn’t allow him to adjust the time because the original navigation map SD card was missing.

Moorats had been warned. When he bought the used car, the previous owner noted that the SD card would be required to change the time on the in-car display. “I didn't really find that believable,” recalls the security researcher. They seem to have believed it was a skill issue of the prior owner and, because the maps it contained were chronically out of date, they decided to confidently decide they didn’t need the SD card...

The time to adjust non-connected clocks to daylight saving time inevitably came, and Moorats decided to adjust the in-car time display, which had become an hour out of sync. Two settings menus were available via the in-dash system. One allowed the adjustment of Bluetooth and Phonebook settings – clearly no use here. But the second settings menu wouldn’t produce any submenu. Instead, an error message "No maps detected!" popped up every time it was prodded.

With a certain set of skills under his belt, Moorats decided to “just RE it,” where RE = reverse engineer. He found out that the head unit ran a customized Linux-based TomTom operating system on a Samsung SoC. In brief, he extracted and analyzed the firmware, built a QEMU environment for the device, reversed the scripting engine, and uncovered a number of vulnerabilities affecting the system.

The breakthrough came in a brainwave, which seems to stem from experience with kernel exploit techniques during his day job. “Instead of trying to get the userland to execute our script, why don't I get the kernel to execute our script instead?” pondered Moorats. The approach worked, rewarding Moorats with root-level access to the infotainment system.

The 2014 Renault Latitude owner subsequently discovered that the date & time settings were in /mnt/flash/settings.dat in a custom binary format. Interestingly, the system “doesn't really have a concept of time zones at all,” observed Moorats in the conclusion of his blog post. “It gets the time from GPS and stores the local offset as a number of minutes. Mine was set to 180. I changed it to 120.”

Current-gen desktop builds are becoming rare due to memory and graphics card pricing, so many enthusiasts have taken to upgrading older machines, using previous-gen platforms, or buying used hardware. For any and all of those scenarios, the G.Skill Ripjaws V DDR4-3200 32 GB kit is just perfect, and it can be yours for $209.99 at Newegg with promo code BNWF83.

• Check out the G.Skill Ripjaws V 32 GB DDR4-3200 kit at Newegg with promo code BNWF83

This dual-channel kit does exactly what it says on the tin, offering a roomy 32 GB that ought to be fine for many years to come, or until Altman finds your location. The rated speed is pretty good at 3200 MT/s, and latency is pretty good at 16 cycles. There's no onboard RGB, but that's a blessing in disguise, proven by the low asking price. The exact model is F4-3200C16D-32GVK, if you're curious.

With this kit in hand, you can conjure up a perfectly potent gaming rig using, say, a Ryzen 5800X3D CPU ($319.99 with promo code BFF43), an Asus TUF B550-Plus Wi-Fi motherboard ($119.99), and a Gigabyte Radeon RX 9070 graphics card ($729.99) and still come in under $1,500 for the baseline components.

As befits any quality memory kit, there are onboard XMP 2.0 profiles, and the maker offers a limited lifetime warranty. I've personally used the G.Skill memory for decades now, and have nothing but good things to say about the quality of the gear and the effectiveness of the company's RMA and support.

Nvidia CEO Jensen Huang has admitted to having an affair with a woman who is allegedly a Chinese spy. His apology came in a letter to Nvidia staff, which you can read in full below. Nvidia's board of directors released a statement late on Sunday to address the bombshell report that U.S. intelligence believes Huang had an affair with a Chinese spy. The board says it has no reason to believe sensitive corporate information or U.S. national security was compromised, but the company intends to examine all available information to determine what happened.

The BoD claims it is fully confident in Jensen Huang and his leadership team. Huang, who has been married to his wife, Lori Huang, for 41 years, says he fully supports the investigation.

"A committee of independent directors, assisted by outside independent counsel, will oversee the investigation," a statement by Nvidia's BoD reads. "The Board prioritizes U.S. national and economic security, which is paramount to the interests of Nvidia and its shareholders. […] Jensen fully supports the review and has committed to cooperating with it."

On Friday, the first reporting emerged claiming U.S. intelligence agencies believe Huang had a romantic relationship with Alice Huang, also known as Huang Le, a former executive at Singapore-based Megaspeed who was allegedly recruited by China's Ministry of State Security (MSS) to gather information and help Chinese companies obtain restricted AI GPUs. The company has been on the radar for smuggling restricted hardware to China since at least 2025: Megaspeed has been accused of smuggling $2 billion worth of AI chips into China via its subsidiary Speedmatrix.

Nvidia has strenuously rejected suggestions of improper dealings, saying that its CEO regularly meets thousands of customers and that these interactions do not involve sensitive corporate information or national security secrets. Chinese officials have also dismissed the espionage allegations as unfounded.

The suspected Chinese intelligence operation targeting Jensen has not been independently confirmed as a fact. The relationship is alleged to follow the typical "honeypot" strategy, wherein foreign intelligence agencies use operatives to ensnare targets into romantic relationships to steal information and influence their decision-making.

Huang has been an outspoken critic of those opposed to limiting the supply of AI GPUs to China. Huang recently expressed his opinion that the label 'China hawk' is not a badge of honor Americans should wear; it is really 'a badge of shame.' Huang also said it "doesn’t have to be all us or them. It could be us and them," drawing fire from others in the administration and other tech companies, including the CEO of Palantir.

"I owe my family and all of you an apology," Huang wrote. "I was unfaithful to my wife. My actions hurt her and our children. There is no excuse for what I did. I am deeply sorry. Throughout my life, my family has given me love and support. They deserve my honesty and loyalty. Instead, I let them down and damaged the life we built together. I take full responsibility for what I did and the hurt I caused, and I accept the criticism that comes with it. To my wife and family: I love you. I know an apology cannot undo what I have done. I hope our family will heal and that, over time, I can earn back your trust."

In June 2018, Intel chief executive Brian Krzanich resigned after an internal investigation, assisted by external legal counsel, confirmed that he had engaged in a consensual romantic relationship with an Intel employee, which violated the company's non-fraternization policy applicable to all managers. Krzanich, who had led Intel since 2013, also stepped down from the company's board, while then-chief financial officer Robert Swan took over as interim CEO. For now, it remains unclear if Jensen will meet a similar fate and lose control of the company he founded.

Jensen's full letter to staff reads as follows:

To my colleagues at NVIDIA, I owe my family and all of you an apology. I was unfaithful to my wife. My actions hurt her and our children. There is no excuse for what I did. I am deeply sorry.

Throughout my life, my family has given me love and support. They deserve my honesty and loyalty. Instead, I let them down and damaged the life we built together. I take full responsibility for what I did and the hurt I caused, and I accept the criticism that comes with it.

To my wife and family: I love you. I know an apology cannot undo what I have done. I hope our family will heal and that, over time, I can earn back your trust.

To my colleagues at NVIDIA, including those I have worked alongside for decades, and to our board, partners, and shareholders around the world: you deserve better from me. Leading NVIDIA is the privilege of my life. My responsibility is to serve you and be worthy of your trust. I have fallen short of that responsibility. I am fully committed to NVIDIA and will work to earn back your confidence.

Regarding the national security concerns raised in news reports, NVIDIA’s Board of Directors has initiated an independent investigation. I support the investigation and will cooperate fully. I love our country and our company. I am committed to protecting NVIDIA and supporting America’s national security. I am proud and grateful to be an American.

My family bears no responsibility for my choices. I do. Please give them privacy.

Jensen

The whole statement from Nvidia's Board of Directors reads as follows:

The Nvidia Board is initiating an independent investigation in light of recent reporting that founder and CEO Jensen Huang has been the target of a foreign intelligence effort.

A committee of independent directors, assisted by outside independent counsel, will oversee the investigation.

The Board prioritizes U.S. national and economic security, which is paramount to the interests of NVIDIA and its shareholders.

While the Board has no reason to believe that either national security or NVIDIA secrets were compromised, the Board's inquiry will examine all available information.

Jensen fully supports the review and has committed to cooperating with it.

The Board has confidence in Jensen and his leadership team and will provide further updates as appropriate.

Diamonds have exceptional cooling capabilities due to their excellent thermal conductivity, with the material estimated to be at least five times more conductive than copper. Because of this, Diamond Foundry, a firm focusing on lab-grown diamonds, announced the world’s first 200mm single-crystal diamond (SCD) wafer. The 200mm SCD is about the same wafer size that semiconductor fabs use for older processes, with the company boasting on its website that “Single-crystal diamond has entered the wafer-scale era.” This is part of the effort to manufacture SCD at scale so that it can be used for semiconductor cooling, especially now that increasing power of AI chips has also led to enormous thermal problems.

What makes Diamond Foundry’s 200mm SCD wafer different from other lab-grown diamonds is that it’s created as a high-quality single-crystal piece instead of a polycrystalline gem. Although the former is difficult to manufacture, it is what allows diamond to become an extreme efficient thermal conductor while simultaneously retaining its electrical insulating properties — this efficiency should help cut down on hotspots, making SCDs an excellent solution for AI chip cooling.

The company also showed in a graph that it is now thought possible to create a D300 SCD wafer before 2030, matching the size of the wafers used in leading node processes. However, it should be noted that the sample shown by the company remains unpolished, meaning it still needs to go through some processes to make it suitable for chipmakers.

Although the company was founded in 2012, it also acknowledged the work of scientists who have been chipping away at the scale problem of SCDs since the early 2000s. “This is one of those milestones that once seemed almost impossible. Hundreds of DF engineers, scientists, and technicians contributed to us getting here. It is the accumulated result of countless experiments, failures, inventions, rebuilt machines, and insights over more than a decade,” Diamond Foundry CEO Martin Roscheisen said.

“Stefan and Martin pursued this goal for 25 years. There are very few examples in technology of someone sticking to a really hard scientific challenge with that level of perseverance. They kept going when the prevailing view was that single‑crystal diamond simply could not be made at this scale. Today, the perseverance shows as the first-ever D200mm single‑crystal diamond. And best of all, it makes D300mm predictable as well.”

Diamond Foundry isn’t the only company that has identified the potential of lab-grown diamonds for semiconductor cooling. Akash Systems secured $18.2 million in CHIPS Act funding in 2024 to produce Gallium Nitride fused with synthetic diamond, which it claims will reduce temperatures by 20 degrees C. The De Beers Group, which is known for its diamond mines in Africa, is also building a copper-plated diamond composite material aimed at improving cooling efficiency for AI, high-performance computing, and GaN RF devices. Even Frore Systems, which is known for its solid-state cooler, is diving into diamond-infused watercooling cold plates for Nvidia’s next-generation AI GPUs.

Logitech's G502 X Lightspeed has been around for awhile — the mouse was introduced in 2022 alongside the fancier Logitech G502 X Plus (and the wired G502 X), but since it’s receiving a firmware update that upgrades its Hero 25K sensor to a Hero 44K sensor, this seemed like a good time to take a look at it. The G502X Lightspeed is available in black and white colorways and currently retails for $159.99, which is the same price as the G502 X Plus — however, it can (and should) normally be found for less than this.

Design and Comfort of the G502 X Lightspeed

The Logitech G502 X Lightspeed is a right-handed gaming mouse with an ergonomically contoured shape. It's the same size and shape as the G502 X Plus, with large primary mouse buttons, textured rubber side panels for grip, and a smooth matte chassis with glossy accents. It comes in black and white colorways (our review unit was black) and has no lighting aside from a small indicator light at the front of the mouse to indicate connection and battery life.

Like the G502 X Plus, the G502 X Lightspeed is definitely on the larger side. It measures 5.17 inches (131.4mm) long by 3.12 inches (79.2mm) wide, and is 1.62 inches (41.1mm) high at its tallest point. It's larger than the Razer Basilisk V3 Pro and V4 Pro (5.11 x 2.96 x 1.67 inches / 130 x 75.4 x 42.5mm) — not by a lot, but it definitely feels like a much larger mouse. It weighs about 3.6 ounces (102g), which makes it lighter than the Basilisk V3 Pro, but slightly heavier than the Basilisk V4 Pro.

The G502 X Lightspeed has 11 programmable buttons, and 13 programmable controls if you include scroll up/down. There are the left and right primary mouse buttons, as well as two side buttons along the left edge of the left mouse button. On the left side of the mouse there are the two standard thumb buttons, plus a "sniper" button that's magnetically attached and can be flipped so that it's not in your way. The sniper button can also be "deleted," and Logitech includes a delete cover in the box that makes the button unpressable (and also has the same soft, rubbery texture as the rest of the thumb rest).

The scroll wheel is a 4-way tilt wheel, so it has three programmable "buttons" (left click, right click, and center click), as well as scroll up/down. Above the scroll wheel are two buttons: One is a mechanical scroll wheel switch that switches between notched, tactile scroll and free scroll — that’s not programmable, as it's a physical switch — and the other is a profile cycle button that can be remapped.

On the bottom of the mouse you'll find UHMWPE mouse feet — at the top and bottom, around the sensor, and under the thumb rest. There's a power switch, as well as a magnetically-attached dummy puck that can be swapped out for the charging puck included in the Logitech PowerPlay 2. Under the puck is a slot for storing the mouse's wireless dongle.

In the box, the G502 X Lightspeed comes with a 6-foot (1.8m) USB-C-to-USB-A cable, a USB-A dongle, a USB-C converter, and a sniper button delete cover.

Specs

Performance of the G502 X Lightspeed

The Logitech G502 X Lightspeed ships with Logitech's Hero 25K optical sensor, which has a max resolution of 25,600 DPI and a maximum speed of 400 IPS, and can handle up to 40 G's of force. However, both the G502X Lightspeed and the G502 X Plus are receiving a firmware update that will upgrade the Hero 25K optical sensor to a Hero 44K optical sensor, with a maximum resolution of 44,000 DPI and a maximum speed of up to 678 IPS. It's not clear when this upgrade will roll out, however, and I wasn't able to upgrade my review unit in G Hub as of this writing.

Upgrades aside, the G502 X Lightspeed's sensor is very good. It tracks on just about every surface you might possibly use while gaming, and is speedy, sensitive, and very accurate. The mouse supports a polling rate of up to 1,000 Hz, which is still pretty standard for most gamers — however, if you're someone who's convinced that the extra 0.125 milliseconds of lag is the reason for your losses, this is not a mouse with an 8,000 Hz or even a 4,000 Hz polling rate.

The G502 X Lightspeed features Logitech's Lightforce hybrid optical-mechanical switches in its primary mouse buttons, which combine the feel of mechanical switches with the speed and reliability of optical ones. The G502 X Lightspeed's switches do have a nice, tactile click with a crisp sound that's definitely on the louder side. The side buttons also feel and sound good — not quite as tactile as the primary buttons, but with a very crisp click sound.

The mouse features a dual-mode tilt wheel with a mechanical switch that lets you move between a notched, tactile scroll and a smooth, speedy free-scroll. Logitech's implementation of this is different from Razer's — Logitech's is mechanical, while Razer's is software-based. This means the modes are more distinct, but I'm not really sure if that's a good thing. While I liked the free-scroll mode of the G502 X Lightspeed better than I liked the free-scroll mode on Razer's mice (such as the Basilisk V4 Pro and the Naga V3 Pro), the tactile scroll was stiff, clunky, and very loud — with a lot of rattling. The scroll wheel only has two modes, so there's no middle ground — and while I'm fine to use free-scroll for most things, there are definitely games where I need a notched scroll wheel.

Features and Software of the G502 X Lightspeed

The G502 X Lightspeed is configurable via Logitech's universal peripheral software, G Hub. You can use G Hub to change the mouse's DPI settings (DPI steps and DPI shift speed — the DPI switch that occurs when you hit the DPI shift button) and remap keys. You can also use G Hub to update the mouse's firmware, so this is presumably where the Hero 44K sensor upgrade will come through.

Wireless Experience & Battery Life of the G502 X Lightspeed

The G502 X Lightspeed connects to your PC via 2.4GHz wireless; it can also utilize a wired connection, though the included USB-C cable is rubber and won't get you that smooth, drag-free wired mouse movement you can get with flexible paracord-covered mouse cables. The mouse does not include Bluetooth.

Logitech rates the G502 X Lightspeed at 140 hours of battery life, which is about 10 hours longer than the G502 X Plus (with no lighting). This is pretty impressive, even if there are similar wireless mice that last longer (the recently-launched Razer Basilisk V4 Pro gets up to 200 hours, for example). This should give you at least two weeks before you need to recharge, even with intense use. The G502X Lightspeed is also compatible with the PowerPlay 2, Logitech's wireless charging mousepad (sold separately).

Bottom Line

The G502 X Lightspeed is a solid gaming mouse for most gamers — its contoured, ergonomic shape is comfortable, though it does feel bigger than similarly shaped mice, such as the Razer Basilisk V4 Pro. It has enough extra well-placed buttons for the average gamer (though, not enough for the average MMO player), and its 140-hour battery life is higher than average. Plus, assuming it actually gets this firmware update, it will have a competitive sensor — though it's not going to be quite as good as the actual Hero 2 sensor. But this mouse is currently priced at $159.99, which is the same price as the fancier Logitech G502 X Plus — and it's definitely supposed to be the more budget version of the two. Unless you find it at a great sale price, I'd pick up the G502 X Plus or even Razer's Basilisk V3 Pro 35K at a similar price.

The sensory overload typically experienced when visiting video game arcades in Japan was eerily absent from some venues this week, reports ITMedia News [machine translation]. It says that a Sega system outage meant there was “no end in sight” for the rows of silent black screens. However, a recent tweet from arcade network operator ALL.net indicates that both Sega and Bandai Namco machines are affected. Moreover, arcade lovers will be pleased to hear that affected arcades will “gradually resume operation on a trial basis starting today.”

ALL.Net, a contraction of Amusement Linkage Live Network, was launched by Sega in 2004 to allow arcade games to look up to the net for auto updates, data logging, implement online play, and power national leaderboards. Sega titles like Initial D The Arcade, Maimai, and CHUNITHM, rely on this network, as do Bandai Namco games like Aiko no Tatsujin, Mario Kart Arcade GP DX, and Gundam Extreme Vs.

In its official apology blog post, dated November 2, ALL.Net / Sega Fave said that it suspended services a day prior after it became aware of its servers being affected by “unauthorized access by a third party.” It reassured customers that their personal info and billing data were not on the compromised server. However, it asked for more time to investigate, run safety checks, and so on, before restoring service.

This morning ALL.Net told customers that Bandia Namco titles “will gradually resume operation on a trial basis starting today (online play and card use enabled).” Sega products remain under maintenance, and online services are unavailable. There is currently no estimate for issue resolution, but a news bulletin is promised for November 10 (Tuesday).

Classic offline arcade machines weren’t affected by this outage, of course. Moreover, rival arcade gaming legends like Taito and Konami run their own, separate, arcade networks and authentication systems.

Linux developers are preparing to make Intel’s ancient MultiMedia eXtensions an irrelevance. Intel introduced MMX in the Pentium era, but a patch expected to merge into mainline Linux 7.5 will remove the last remaining optimized software from the open-source OS. Expected to be released in 2027, the new kernel will wipe MMX-optimized code from its software RAID XOR code, explains Phoronix.

Admittedly, there wasn’t much left in Linux in terms of MMX optimizations. But few folks will be running the most modern kernels on Pentium MMX and Pentium II chips in 2026. And most other MMX-optimized code disappeared from the OS half a decade ago, when some MMX-optimized memory management code was removed. Note that, if you are a determined late-90s Pentium devotee, you can still use software RAID, but it will lack the MMX fast path.

This is a classic case of Linux code housekeeping. With this change, the kernel will lose over 800 lines of code and replace them with just 12.

Remembering MMX

Intel first introduced MMX Pentium chips in January 1997. Though I vaguely remember these processors being marketed as being enhanced by ‘MultiMedia eXtensions,’ Wikipedia insists that “MMX is officially a meaningless initialism[6] trademarked by Intel.”

In a subsequent trademark battle with longtime foe AMD, a document produced in court indicated that MMX stood for ‘Matrix Math Extensions.’ AMD relied on this to sidestep the trademark, as an initialism cannot be trademarked, according to the encyclopedic source.

Eventually, all rival x86 processors managed to integrate MMX instructions – so as not to be pushed into the shade during the multimedia boom. A surprisingly large number of non-Intel chips ended up offering MMX, too. Along with Pentium MMX chips, AMD, Centaur, Cyrix, Rise, and Transmeta eventually offered MMX CPUs.

After MMX's marketing success, AMD pioneered 3DNow! instructions, and about a year later, Intel fired back with SSE and SSE2.

Ukraine is developing a new class of combat robots designed to operate inside underground gas pipelines, where Russian troops have repeatedly attempted to bypass frontline defenses, United24Media reports. The project, disclosed by Ukraine's 2^nd National Guard Corps, Khartiia, involves modifying unmanned ground vehicles (UGVs) to perform reconnaissance and combat missions in spaces too narrow for conventional military machinery. Notably, the work is being carried out by specialists from the corps’ Lava Unmanned Systems Regiment, following Operation Veles — an underground assault that saw Ukrainian soldiers advance more than 15 kilometers through pipeline infrastructure near Kupiansk. Khartiia says the operation killed more than 100 enemy soldiers and resulted in over 30 captured, although those battlefield figures have not been independently verified.

The commander of Lavas ground robotic system battalion, identified by the callsign ‘Matematyk’, said the project addresses a category of warfare that requires a different approach from conventional battlefield robotics. “There is a very long list of tasks that need to be accomplished to fight underground and counter the enemy underground. This is a new class of technologies and challenges that we will have to confront,” he said.

The machines are being adapted from existing groundrobots,rather than developed as conventional surface vehicles. They must be compact enough to navigate narrow pipelines while retaining the ability to carry out military assignments . Most commercially available underground machinery, meanwhile, is designed for construction, drilling, and tunneling — not combat.

As of now, Khartiia is already developing several robotic systems to address those limitations, with the broader objective of sending machines into dangerous underground environments where soldiers would otherwise have to conduct reconnaissance, confront enemy positions, clear positions, or clear out themselves. “War is already underground, and it will expand. The question is how we adapt to this war,” Matematyk said.

The project follows unusual battlefield developments near Kupiansk, where Ukrainian forces used the same underground infrastructure that Russian troops have previously exploited to infiltrate behind Ukrainian defensive positions. During Operation Vels, Ukrainian soldiers moved through a gas pipeline measuring approximately 1.2 meters in diameter, according to a report by Ikrainska Pravda. The confined passage made movement and close-range firefights particularly,with Ukrainian troops having little-to-no room to maneuver.

Russia had previously used the pipeline network beneath the Oskil River to move soldiers toward Ukrainian-held territory. The Ukrainian battlefield monitoring group ‘Deepstate' reported an earlier Russian infiltration attempt in September 2025. Last month, President Volodymyr Zelenskyy said that Ukrainian forces had eliminated more than 100 Russian troops and captured over 30 during the operation. The experience also exposed the limitation of sending soldiers into such environments. Underground operations restrict visibility,complicate communications , and leave little room for maneuvering equipment.

That said, Ukraine has increased its reliance on unmanned military systems as battlefield terrain proves complicated to navigate. In July 2024, we reported on Ukrainian forces deploying AI-controlled drones capable of independently identifying and attaching targets. The country has also been experimenting with unmanned systems that operate beyond conventional aerial platforms. In August, the US Army tested an autonomous Ford F-250 breaching vehicle, which builds on the Lancer unmanned ground vehicle program and uses a counter-drone turret to protect military operations from aerial attacks.

However, autonomous ground systems still face considerable technical challenges. Russia's Shturm uncrewed assault tanks , for instance , repeatedly lost their command connection during the Tsentr-2026 military exercise. The vehicle eventually became stuck in mud, even after organizers switched to manual driving. Russia has also tested an AI powered Shahed variant using Nvidia's Jetson Orin computing platform. The system was reportedly capable of identifying targets without relying on externally supplied coordinates.

On this day in 2006, Nvidia released the legendary GeForce 8800 GTX and 8800 GTS graphics cards. Nvidia has decided to honor the flagship of the pair with its GeForce 8800 GTX - 20 Years of a Legend interactive demo, now free on Steam. However, the recommended specs to ‘play’ with this semi-interactive path-traced unboxing experience include a wallet-worrying RTX 5080. Or you could just watch the video, embedded below for your convenience.

The downloadable demo, which also requires 32GB of RAM and 30GB of free storage, was created by Dylan Browne, a senior special effects artist who frequently works with Unreal Engine, as in this case. This experience is said to deliver era-appropriate graphics, but requires the most advanced graphics tech available two decades after the 8800 GTX first hit the scene to work its graphical magic.

To run GeForce 8800 GTX - 20 Years of a Legend on a 2026 PC, you will need at least an RTX 4070 for acceptable performance, according to the Steam page. This software is stuffed with the latest graphics tech embraced by the green team. “DLSS 4.5 with Dynamic Multi Frame Generation, Ray Reconstruction, path tracing, and Mega Geometry” are all present, says Nvidia.

There’s a bit more to the interactive demo than looking around and poking a box with your mouse and then prodding it to fit the shiny new GPU into your “era-accurate PC tower.” An extra layer of fun is promised by the provision of hidden Easter eggs in this experience.

However, the demo appears to show Windows Vista on the PC in the gaming den, which launched at retail a few months after the 8800 GTX. Admittedly, this legendary GeForce card and Windows Vista OS would roam the Earth at the same time, with many folks buying the GeForce 8800 GTX and 8800 GTS to shove into their Vista PCs from 2007 onwards. Crysis launched on PCs in late 2007, a year after the G80 architecture cards debuted.

Nvidia launched the GeForce 8800 GTX (128 shader units) and 8800 GTS (96 shader units) at $599 and $449, respectively. These legendary Nvidia GPUs spearheaded the DirectX 10 era, accelerating existing PC titles and delivering next-gen visuals with new DX10 software. You can read more in our reviews from the time, like the 12-page review of the GeForce 8800 GTX and GTS from the launch week. Briefly, our reviewer thought these cards were “amazing” for their current performance and DX10 promise. They were also excited about the G80 architecture for bringing CUDA to consumers and the promise of making the GPU “something much more than a frame rendering device.” Some Luddite gamers may wish that had never happened.

Nvidia driver support for the GeForce 8800 GTX and GTS would run out of road about a decade later. By that time, DirectX 11 hardware and games were mainstream, leaving older graphics hardware designs obsolete when it came to playing the latest AAA PC titles.

Microsoft VP Scott Hanselman has revealed a new open-source GitHub project called gputaskman. The name neatly encapsulates the headline role of this tool, a GPU-focused task manager, but its scope might be broader than you expect. This new tool for Windows, Linux, and macOS can show per-process GPU usage, VRAM consumption, active runtimes, and many more stats. It also breaks out AI/LLM task-specific information.

On social media, Hanselman snappily makes clear the technical itch that gputaskman was designed to scratch. “There's a lot going on in my GPU,” wrote the prominent public face of Microsoft's developer ecosystem. “I'm making a GPU Taskman to get insights to where my GPU memory is going.”

There are several example screenshots of GPU Taskman in use by the Microsoft VP. They clearly show the kinds of data that are presented in the terminal-like display that pops up in the tool. Hanselman’s tweet shows that the most VRAM-hungry app on his RTX 4080S system is the desktop windows manager. That is selected in the list, and more details are shown for this task at the bottom of the window. Also, we see AI/LLM apps consuming VRAM, and a few Windows staples.

In a GitHub-embedded image, we see what details are broken out when the llama-server.exe app is selected in the list. The bottom of the window breakdown reveals that Qwen3-0.6B is being used, and there’s a breakdown of its weights, the KV cache, the number of parameters, and more. Very informative.

As the Microsoft developer has made portable Windows downloads of the latest v0.3.0 release of gputaskman, I couldn’t resist giving it a spin. Upon launch, I found that no tasks were using the discrete GPU in my laptop… so I fired up ye olde Unigine Valley GPU benchmark to stir the RTX 4060 from its slumber. I note the live-streaming charts showing GPU load, VRAM use, temperature, and so on are more prominent in my screenshot than in Hanselman’s examples. Perhaps he screen-shotted a previous version of the tool.

There’s a lot more information about gputaskman, its inner workings, and even its source code, on the linked GitHub repository. Hanselman is sharing his “experimental software” tool under the permissive MIT license. Finally, in social media comments, he hints that expanded GPU monitoring capabilities could be coming to Windows as standard.

Amsterdam police have arrested 150 to 200 protesters at a Microsoft data center site on suspicion of trespassing. According to a Facebook post by the authorities [machine translated], tools designed for breaking into the site were seized from the protesters, who were then allowed on their way to protest peacefully. However, it appears that some members of the group decided to enter the construction site “with violence” anyway, forcing the police to intervene and leading to their arrest. There were also some allegedly stolen items from the construction site found among the belongings of some of the protesters, plus damage and vandalism done to the data center site.

This protest is a continuation of the action that started last Friday, where protesters breached the Microsoft data center, which was being owned by Pure Data Centers. It is also the first of many organized by the “Stop Dirty Data Centers” coalition across Europe, with actions planned for Amsterdam, Dublin, Warsaw, Denmark, and several other cities across the continent. The Dutch publication NOS [machine translated] identified the exact group at the site today as Fossielvrij, who are against the use of fossil fuels in data centers such as the jet engines that Elon Musk installed at xAI’s Mississippi data center.

“Gas prices are going through the roof, and new houses are being postponed because of the overcrowded power grid. Yet tech billionaires like Elon Musk and Jeff Bezos are building energy-guzzling, hideous data centers,” the group said on its website. “We can take back the power of Big Tech by pressing the pause button now. Give us the opportunity to democratically decide what we do and do not use AI for, so that it is not at the expense of our living environment and everything that makes our lives beautiful. Then we look at how many more data centers are needed.”

The protest comes amid the AI data center build-out, which has been facing increased opposition across the United States. A Google data center site in Finland is also in trouble after it cleared out around 1,300 acres worth of forest and trees without an environmental impact assessment, displacing wildlife, and leading to a criminal probe on Tuike Finland Oy, the Google subsidiary in charge of the site.

Fossielvrij’s action was supposed to be peaceful, but it appears that another group joined the demonstration and broke through the fences around the site, leading to the police confrontation. The suspects were arrested for domestic violence and then taken to a cell complex, where it will be decided if they will be detained longer. The police also said that “All suspects will be heard afterward.”

An investigation has found that addresses for two Chinese companies approved to purchase Nvidia H200 chips led to empty offices, and the receptionists at the buildings were unfamiliar with them. According to The New York Times, these companies didn’t just have empty offices; their executives are often linked to other firms sanctioned by the United States for working with and providing advanced technology to the Chinese military.

This news has revealed concerns that the U.S. Department of Commerce is simply rubber-stamping GPU export license applications instead of scrutinizing each one for potential links to Beijing. “I have a ton of concerns that neither the Commerce Department or Nvidia is doing the due diligence needed to ensure that the licensing terms are being followed,” former China director for the White House National Security Council Liza Tobin told the publication.

Many of these approved companies include well-known entities like Baidu, BYD, Tesla Shanghai, as well as ZTE, which has previously been blacklisted from the U.S. market, Alibaba, Tencent, and ByteDance. However, the outfit’s investigation also revealed approval for two relatively unknown firms: New Matrix and Maginfra. The offices for both companies were empty shells with some office furniture, and people working nearby said they’d never heard of the company or seen anyone go to the deserted spaces.

The New Matrix website also reportedly did not contain any information about its owners or its history, although public company records reveal that it was founded by Li Bin, who, along with two other primary owners, are connected to Wise Road, a Chinese private equity firm sanctioned by the U.S. for helping Beijing acquire sensitive chip technologies. Furthermore, Mr. Li chairs Tsinghua Unigroup, a tech conglomerate partly owned by the Chinese government, and owns New H3C Semiconductor Technologies, which has also been added to the Commerce Department’s Entity List for cooperating with the Chinese military.

Maginfra, on the other hand, is suspected of being linked to Inspur, which Washington has blacklisted since 2023 for supporting the People’s Liberation Army, with both companies sharing personnel, patents, and a similar government ownership structure. These two firms aren’t listed as the final users of Nvidia’s AI chips, though. Instead, they’re supposed to assemble them into AI servers, which would then be forwarded to a final user, although it’s unclear who their customers are.

Aside from these questionable non-entities, another company that received a license to purchase H200 GPUs has some suspicious links. Zhejiang Yindun Cloud Technology’s executives are reportedly related to the founder of Range Intelligent Computing Technology Group, a data center operator that has sold compute capacity to Chinese military contractors. It’s also reportedly partly operating out of a Range data center.

President Donald Trump banned the export of Nvidia AI GPUs in April 2025, even those customized to meet the performance ceiling Washington set for AI processors. The president reversed the decision late last year in exchange for a 25% cut of the sale, but it still took more than eight months for the first H200 shipments to arrive in China. Beijing’s refusal to give Chinese tech companies the green light to buy the chips for the first few months of the year caused the delay. But once China lifted its import ban on Nvidia chips, several Chinese companies applied for licenses to buy them, and The New York Times reported that around two dozen firms have already received approval, including these questionable ones.