When the Ethereum blockchain moved away from using a technique for verifying transactions known as proof of work last September, crypto market demand for the specialized processors that performed these calculations disappeared virtually overnight. Companies that used and hosted GPUs, or graphics processing units, saw a key part of their once-booming business vanish against an increasingly difficult backdrop for crypto. But now mining infrastructure companies like Hive Blockchain and Hut 8 Mining are finding opportunities to repurpose their GPU-based equipment for another industry on the precipice of a possible boom: artificial intelligence. From a report: "If you can reapply some of that investment in the GPU mining infrastructure and convert it to new cards and workloads, it makes sense," Hut 8 Chief Executive Officer Jaime Leverton said in an interview. GPUs -- designed to accelerate graphics rendering -- require constant maintenance and physical infrastructure not all users are prepared to provide. As such, Hut 8 and a few other miners have been using the chips to power high-performance computing, or HPC, services for clients across a range of industries. But inroads with the burgeoning and much-hyped AI sector -- which requires huge amounts of computing power -- represent the kind of transformational opportunity miners had been seeking when they originally bought the processors.

The Today show's food reporter delivers a strange report on a viral AI-generated ad "for an imaginary pizza place called 'Pepperoni Hug Spot'."

Everything looks slightly ... off. Generated by AI, the audience is reminded constantly through the uncanny valley that the people aren't real — and neither is the pizza. "Cheese, pepperoni, vegetable, and more secret things," says the voiceover, which is also artificially generated... "Knock, knock, who's there? Pizza magic," the AI narrator says after a delivery driver (whose steering column is on the left side of his car) is shown delivering a pizza.

"Eat Pepperoni Hug Spot pizza. Your tummy say 'Thank you.' Your mouth say, 'Mmm,'" the ad continues while showing a trio of women eating pizza in the oddest possible fashion, complete with bizarre cheese pulls and facial contortions out of a food-based nightmare. "Pepperoni Hug Spot: Like family, but with more cheese..."

Using AI technologies Runway Gen2, Chat GPT4, Eleven Labs, Midjourney and Soundraw AI, the creator was able to produce the background music, voiceover, graphics, video and even generate the script for the ad. "I used Adobe After Effects to combine all the elements, adding title cards, transitions, and graphics," he adds... Seeing it spread, he whipped up a website that fit the uncanny vibe of the commercial and even created merch including hats and T-shirts.

"I figured I should capitalize on my 15 minutes of internet fame, right?" he jokes.
Twitter CEO Elon Musk "simply responded with an exploding head emoji."

And Pizza Hut's official Twitter account posted their reaction: "My heebies have been jeebied."

UPDATE: Saturday Pizza Hut Canada "transformed" one of its restaurants into the restaurant from the commercial, emblazoning the logo for Pepperoni Hug Spot onto its boxes, employee t-shirts, and the sign outside. There's two videos on the official Instagram feed for Pizza Hut Canada (which for the occasion changed its tagline to "Like family, but with more cheese.")

One video closes by promising the pizza does, indeed, contain "secret things."

Slashdot reader indominabledemon shared this article from TechSpot: Games today use highly-detailed textures that can quickly fill the frame buffer on many graphics cards, leading to stuttering and game crashes in recent AAA titles for many gamers... [T]he most promising development in this direction so far comes from Nvidia — neural texture compression could reduce system requirements for future AAA titles, at least when it comes to VRAM and storage.... In a research paper published this week, the company details a new algorithm for texture compression that is supposedly better than both traditional block compression (BC) methods as well as other advanced compression techniques such as AVIF and JPEG-XL.

The new algorithm is simply called neural texture compression (NTC), and as the name suggests it uses a neural network designed specifically for material textures. To make this fast enough for practical use, Nvidia researchers built several small neural networks optimized for each material... [T]extures compressed with NTC preserve a lot more detail while also being significantly smaller than even these same textures compressed with BC techniques to a quarter of the original resolution... Researchers explain the idea behind their approach is to compress all these maps along with their mipmap chain into a single file, and then have them be decompressed in real time with the same random access as traditional block texture compression...

However, NTC does have some limitations that may limit its appeal. First, as with any lossy compression, it can introduce visual degradation at low bitrates. Researchers observed mild blurring, the removal of fine details, color banding, color shifts, and features leaking between texture channels. Furthermore, game artists won't be able to optimize textures in all the same ways they do today, for instance, by lowering the resolution of certain texture maps for less important objects or NPCs. Nvidia says all maps need to be the same size before compression, which is bound to complicate workflows. This sounds even worse when you consider that the benefits of NTC don't apply at larger camera distances.

Perhaps the biggest disadvantages of NTC have to do with texture filtering. As we've seen with technologies like DLSS, there is potential for image flickering and other visual artifacts when using textures compressed through NTC. And while games can utilize anisotropic filtering to improve the appearance of textures in the distance at a minimal performance cost, the same isn't possible with Nvidia's NTC at this point.

A new set of patches have been released for the Intel Linux graphics driver that "can provide 10-15% better performance when operating in the tuned mode," reports Phoronix. From the report: The set of Intel i915 Linux kernel graphics driver patches are about exposing the Intel RPS (Requested Power State) up/down thresholds. Right now the Intel Linux kernel driver has static values set for the up/down thresholds between power states while these patches would make them dynamically configurable by user-space. Google engineer Syed Faaiz Hussain raised the issue that they experimented with the Intel RPS tuning and were able to manage up to 15% better performance. With Counter-Strike: Global Offensive with OpenGL was a 14.5% boost, CS:GO with Vulkan was 12.9% faster, and Civilization VI with OpenGL was 11% faster while Strange Brigade was unchanged. No other game numbers were provided.

But as this is about changing the threshold for how aggressively the Intel graphics hardware switches power states, the proposed patches leave it up to user-space to adjust the thresholds as they wish. Google engineers are interested in hooking this into Feral's GameMode so that the values could be automatically tuned when launching games and then returning to their former state when done gaming, in order to maximize battery life / power efficiency. The only downside with these current patches are that they work only for non-GuC based platforms... So the latest Alder/Raptor Lake notebooks as well as Intel DG2/Alchemist discrete graphics currently aren't able to make use of this tuning option.

An anonymous reader quotes a report from ZDNet, written by Steven Vaughan-Nichols: The latest Linux kernel is out with a slew of new features -- and, for once, this release has been nice and easy. [...] Speaking of Rust, everyone's favorite memory-safe language, the new kernel comes with user-mode Linux support for Rust code. Miguel Ojeda, the Linux kernel developer, who's led the efforts to bring Rust to Linux, said the additions mean we're, "getting closer to a point where the first Rust modules can be upstreamed."

Other features in the Linux 6.3 kernel include support and enablement for upcoming and yet-to-be-released Intel and AMD CPUs and graphics hardware. While these updates will primarily benefit future hardware, several changes in this release directly impact today's users' day-to-day experience. The kernel now supports AMD's automatic Indirect Branch Restricted Speculation (IBRS) feature for Spectre mitigation, providing a less performance-intensive alternative to the retpoline speculative execution.

Linux 6.3 also includes new power management drivers for ARM and RISC-V architectures. RISC-V has gained support for accelerated string functions via the Zbb bit manipulation extension, while ARM received support for scalable matrix extension 2 instructions. For filesystems, Linux 6.3 brings AES-SHA2-based encryption support for NFS, optimizations for EXT4 direct I/O performance, low-latency decompression for EROFS, and a faster Brtfs file-system driver. Bottom line: many file operations will be a bit more secure and faster.

For gamers, the new kernel provides a native Steam Deck controller interface in HID. It also includes compatibility for the Logitech G923 Xbox edition racing wheel and improvements to the 8BitDo Pro 2 wired game controllers. Who says you can't game on Linux? Single-board computers, such as BannaPi R3, BPI-M2 Pro, and Orange Pi R1 Plus, also benefit from updated drivers in this release. There's also support for more Wi-Fi adapters and chipsets. These include: Realtek RTL8188EU Wi-Fi adapter support; Qualcomm Wi-Fi 7 wireless chipset support; and Ethernet support for NVIDIA BlueField 3 DPU. For users dealing with complex networks that have both old-school and modern networks, the new kernel can also handle multi-path TCP handling mixed flows with IPv4 and IPv6. Linux 6.3 is available from kernel.org. You can learn how to compile the Linux kernel yourself here.

Video editors are flocking to DaVinci Resolve in droves, marking a major paradigm shift in the editing landscape that we haven't seen since the dreadful launch of Final Cut Pro X drove users to Adobe Premiere Pro. PetaPixel reports: Resolve has taken a convoluted path to becoming the main rival of the world's biggest non-linear editing (NLE) tool. More a conglomeration of tools than a single program, Resolve came through some acquisitions Blackmagic made when creating a broadcast and cine ecosystem. Comprised of an editing tool, a color correction tool, an audio editor, and an effects tool, Resolve is essentially multiple programs that all integrate so seamlessly that they function as a single application. The color correction tools in Resolve are particularly well regarded, and many films and shows were color graded in Resolve even if they were edited in another program. The same applies to Fairlight, the audio component of Resolve, the go-tool tool for many of Hollywood's most prominent audio engineers.

In 2011, Blackmagic decided to release Resolve as both a paid and a free version. The free version had fewer features than the full version (as it still does), but instead of being crippled, the free version works well enough for most users, with the paid version feeling like a feature upgrade. In the dozen years since Resolve became free, it has picked up an ever-growing number of users, and the YouTube emphasis on the creator market has only increased the pace of adoption. The fact that most successful YouTube channels take years to become successful means a free editing tool is valuable.

Blackmagic has never hesitated to put a feature into Resolve. The program has many options in contextual menus, user interface choices, menu items, keyboard shortcuts, and more. There is so much here that it can be overwhelming. [...] Blackmagic also releases dot-versions (like 18.1) that sometimes add enough features that it acts like a full number upgrade would if it were released by Adobe or Apple. Some of the features in Resolve 18.1, for example, unleashed the wave of recent switchers. Two significant features are buried in a list of around 20 new features in that update. The first is AI-driven Magic Mask tools that make masking people or objects a matter of drawing a line. The other prominent feature is voice isolation, another AI-based feature that removes noises from dialog tracks. Magic Mask alone is worth the price of admission. This tool makes it easy to color-correct significant portions of a shot without doing endless mask adjustments, and it also allows for instant alpha channel creation, allowing for items like text, graphics or even people to be superimposed on the same scene without needing a green screen. You can read the full article here.

It's been just a year since Intel officially announced its Bitcoin-mining Blockscale ASICs, but today the company announced the end of life of its first-gen Blockscale 1000-series chips without announcing any follow-up generations of the chips. From a report: We spoke with Intel on the matter, and the company told Tom's Hardware that "as we prioritize our investments in IDM 2.0, we have end-of-lifed the Intel Blockscale 1000 Series ASIC while we continue to support our Blockscale customers."

Intel's statement cites the company's tighter focus on its IDM 2.0 operations as the reason for ending the Blockscale ASICs, a frequent refrain in many of its statements as it has exited several businesses amid company-wide belt-tightening. We also asked Intel if it planned to exit the Bitcoin ASIC business entirely, but the company responded, "We continue to monitor market opportunities." In the original announcement that the company would enter the blockchain market, then-graphics-chief Raja Koduri noted that the company had created a Custom Compute Group within the AXG graphics unit to support the Bitcoin ASICs and "additional emerging technology." However, Intel recently restructured the AXG group, and Koduri left the company shortly thereafter.

An anonymous reader shares a report: A month ago, Adobe announced Firefly, its entry into the generative AI game. Initially, Firefly's focus was on generating commercially safe images, but the company is now pushing its technology beyond still images. As the company announced today, it will soon bring Firefly to its Creative Cloud video and audio applications. To be clear, you won't (yet) be able to use Firefly to create custom videos yet. Instead, the emphasis here is on making it easier for anyone to edit videos, color grade using just a few words, add music and sound effects and create title cards with animated fonts, graphics and logos. However, Firefly also promises to automatically turn scripts into storyboards and pre-visualizations -- and it will recommend b-roll to liven up videos. Maybe the highlight of these promised new features is being able to color grade a video by simply describing what a video should look like with just a few words (think "golden hour" or "brighten face"). Other new AI-based features include the ability to generate custom sounds and music. Firefly will also help editors create subtitles, logos and title cards by having them describe what they want them to look like.

"The fact remains that Google Chrome is the arbiter of web standards," argues FSF campaigns manager Greg Farough (while adding that Firefox, "through ethical distributions like GNU IceCat and Abrowser, can weaken that stranglehold.")

"Google's deprecation of the JPEG-XL image format in February in favor of its own patented AVIF format might not end the web in the grand scheme of things, but it does highlight, once again, the disturbing amount of control it has over the platform generally." Part of Google's official rationale for the deprecation is the following line: "There is not enough interest from the entire ecosystem to continue experimenting with JPEG-XL." Putting aside the problematic aspects of the term "ecosystem," let us remark that it's easy to gauge the response of the "entire ecosystem" when you yourself are by far the largest and most dangerous predator in said "ecosystem." In relation to Google's overwhelming power, the average web user might as well be a microbe. In supposedly gauging what the "ecosystem" wants, all Google is really doing is asking itself what Google wants...

While we can't link to Google's issue tracker directly because of another freedom issue — its use of nonfree JavaScript — we're told that the issue regarding JPEG-XL's removal is the second-most "starred" issue in the history of the Chromium project, the nominally free basis for the Google Chrome browser. Chromium users came out of the woodwork to plead with Google not to make this decision. It made it anyway, not bothering to respond to users' concerns. We're not sure what metric it's using to gauge the interest of the "entire ecosystem," but it seems users have given JPEG-XL a strong show of support. In turn, what users will be given is yet another facet of the web that Google itself controls: the AVIF format.

As the response to JPEG-XL's deprecation has shown, our rallying together and telling Google we want something isn't liable to get it to change its mind. It will keep on wanting what it wants: control; we'll keep on wanting what we want: freedom.

Only, the situation isn't hopeless. At the present moment, not even Google can stop us from creating the web communities that we want to see: pages that don't run huge chunks of malicious, nonfree code on our computers. We have the power to choose what we run or do not run in our browsers. Browsers like GNU IceCat (and extensions like LibreJS and JShelter> ) help with that. Google also can't prevent us from exploring networks beyond the web like Gemini. What our community can do is rally support behind those free browsers that choose to support JPEG-XL and similar formats, letting the big G know that even if we're smaller than it, we won't be bossed around.

Nvidia's most-advanced graphics cards are selling for more than $40,000 on eBay, as demand soars for chips needed to train and deploy artificial intelligence software. From a report: The prices for Nvidia's H100 processors were noted by 3D gaming pioneer and former Meta consulting technology chief John Carmack on Twitter. On Friday, at least eight H100s were listed on eBay at prices ranging from $39,995 to just under $46,000. Some retailers have offered it in the past for around $36,000. The H100, announced last year, is Nvidia's latest flagship AI chip, succeeding the A100, a roughly $10,000 chip that's been called the "workhorse" for AI applications. Developers are using the H100 to build so-called large language models (LLMs), which are at the heart of AI applications like OpenAI's ChatGPT. Running those systems is expensive and requires powerful computers to churn through terabytes of data for days or weeks at a time. They also rely on hefty computing power so the AI model can generate text, images or predictions. Training AI models, especially large ones like GPT, requires hundreds of high-end Nvidia GPUs working together.

Nvidia announced the GeForce RTX 4070 desktop GPU, a move that anyone who's been putting off a new midrange DIY PC build has likely been eagerly awaiting. It puts the company's impressive Ada Lovelace graphics architecture within grasp for people who don't want to spend $1,000 or more on a huge graphics card. From a report: It'll launch Thursday, April 13, starting at $599 for Nvidia's Founders Edition single-fan model. As is always the case, other manufacturers like Asus, Zotac, Gigabyte, MSI, and others are putting out factory overclocked variants, too. The Verge already has a full review up for the RTX 4070.

The RTX 4070 Founders Edition card requires a 650 W power supply, and it connects via two PCIe 8-pin cables (an adapter comes in the box). Alternatively, it can connect via a PCIe Gen 5 cable that supports 300 W or higher. The RTX 4070 won't require a humongous case, as it's a two-slot card that's quite a bit smaller than the RTX 4080. It's 9.6 inches long and 4.4 inches wide, which is just about the same size as my RTX 3070 Ti Founders Edition card. Despite being a lower-end GPU compared to Nvidia's RTX 4080 or RTX 4090, it retains the DLSS 3 marquee selling point. It's the next iteration of Nvidia's upscaling technique that drops the render resolution to make games run better, then uses the GPU's AI cores to intelligently upscale what you see.

While Chrome 112 just shipped this week and Chrome 113 only in beta, there is already a big reason to look forward to that next Chrome web browser release: Google is finally ready to ship WebGPU support. From a report: WebGPU provides the next-generation high performance 3D graphics API for the web. With next month's Chrome 113 stable release, the plan is to have WebGPU available out-of-the-box for this new web graphics API. Though in that version Google is limiting it to ChromeOS, macOS, and Windows... Yes, Google says other platforms like Linux will see their roll-out later in the year. The WebGPU API is more akin to Direct3D 12, Vulkan, and Metal compared with the existing WebGL being derived from OpenGL (ES). From Google's blog post: WebGPU is a new API for the web, which exposes modern hardware capabilities and allows rendering and computation operations on a GPU, similar to Direct3D 12, Metal, and Vulkan. Unlike the WebGL family of APIs, WebGPU offers access to more advanced GPU features and provides first-class support for general computations on the GPU. The API is designed with the web platform in mind, featuring an idiomatic JavaScript API, integration with promises, support for importing videos, and a polished developer experience with great error messages.

This initial release of WebGPU serves as a building block for future updates and enhancements. The API will offer more advanced graphics features, and developers are encouraged to send requests for additional features. The Chrome team also plans to provide deeper access to shader cores for even more machine learning optimizations and additional ergonomics in WGSL, the WebGPU Shading Language.

Facebook owner Meta intends to commercialize its proprietary generative artificial intelligence by December, joining Google in finding practical applications for the tech. From a report: The company, which began full-scale AI research in 2013, stands out along with Google in the number of studies published. "We've been investing in artificial intelligence for over a decade, and have one of the leading research institutes in the world," Andrew Bosworth, Meta's chief technology officer, told Nikkei in an exclusive interview on Wednesday in Tokyo. "We certainly have a large research organization, hundreds of people." Meta announced in February that it would establish a new organization to develop generative AI, but this is the first time it has indicated a timeline for commercialization. The technology, which can instantly create sentences and graphics, has already been commercialized by ChatGPT creator OpenAI of the U.S. But Bosworth insists Meta remains on the technology's cutting edge.

"We feel very confident that ... we are at the very forefront," he said. "Quite a few of the techniques that are in large language model development were pioneered [by] our teams. "[I] expect we'll start seeing some of them [commercialization of the tech] this year. We just created a new team, the generative AI team, a couple of months ago; they are very busy. It's probably the area that I'm spending the most time [in], as well as Mark Zuckerberg and [Chief Product Officer] Chris Cox." Bosworth believes Meta's artificial intelligence can improve an ad's effectiveness partly by telling the advertiser what tools to use in making it. He said that instead of a company using a single image in an advertising campaign, it can "ask the AI, 'Make images for my company that work for different audiences.' And it can save a lot of time and money."

The US chip-maker Nvidia has said cryptocurrencies do not "bring anything useful for society" despite the company's powerful processors selling in huge quantities to the sector. From a report: Michael Kagan, its chief technology officer, said other uses of processing power such as the artificial intelligence chatbot ChatGPT were more worthwhile than mining crypto. Nvidia never embraced the crypto community with open arms. In 2021, the company even released software that artificially constrained the ability to use its graphics cards from being used to mine the popular Ethereum cryptocurrency, in an effort to ensure supply went to its preferred customers instead, who include AI researchers and gamers. Kagan said the decision was justified because of the limited value of using processing power to mine cryptocurrencies.

The first version ChatGPT was trained on a supercomputer made up of about 10,000 Nvidia graphics cards. "All this crypto stuff, it needed parallel processing, and [Nvidia] is the best, so people just programmed it to use for this purpose. They bought a lot of stuff, and then eventually it collapsed, because it doesn't bring anything useful for society. AI does," Kagan told the Guardian. "With ChatGPT, everybody can now create his own machine, his own programme: you just tell it what to do, and it will. And if it doesn't work the way you want it to, you tell it 'I want something different.'" Crypto, by contrast, was more like high-frequency trading, an industry that had led to a lot of business for Mellanox, the company Kagan founded before it was acquired by Nvidia. "We were heavily involved in also trading: people on Wall Street were buying our stuff to save a few nanoseconds on the wire, the banks were doing crazy things like pulling the fibres under the Hudson taut to make them a little bit shorter, to save a few nanoseconds between their datacentre and the stock exchange," he said. "I never believed that [crypto] is something that will do something good for humanity. You know, people do crazy things, but they buy your stuff, you sell them stuff. But you don't redirect the company to support whatever it is."

"With an 8-core Rockchip RK3588S SoC, the Orange Pi 5 is leaps and bounds faster than the aging Raspberry Pi 4," writes Phoronix: With up to 32GB of RAM, the Orange Pi 5 is also capable of serving for a more diverse user-base and even has enough potential for assembling a budget Arm Linux developer desktop. I've been testing out the Orange Pi 5 the past few weeks and it's quite fast and nice for its low price point.

The Orange Pi 5 single board computer was announced last year and went up for pre-ordering at the end of 2022.... When it comes to the software support, among the officially available options for the Orange Pi 5 are Orange Pi OS, Ubuntu, Debian, Android, and Armbian. Other ARM Linux distributions will surely see varying levels of support while even the readily available ISO selection offered by Orange Pi is off to a great start....

Granted, the Orange Pi developer community isn't as large as that of the Raspberry Pi community or the current range of accessories and documentation, but for those more concerned about features and performance, the Orange Pi 5 is extremely interesting.
The article includes Orange Pi 5 specs:

• A 26-pin header
• HDMI 2.1, Gigabit LAN, M.2 PCIe 2.0, and USB3 connectivity
• A Mali G510 MP4 graphics processor, "which has open-source driver hope via the Panfrost driver stack."
• Four different versions with 4GB, 8GB, 16GB, or 32GB of RAM using LPDDR4 or LPDDR4X. "The Orange Pi 4GB retails for ~$88, the Orange Pi 5 8GB version retails for $108, and the Orange Pi 5 16GB version retails for $138, while as of writing the 32GB version wasn't in stock."

In 169 performance benchmarks (compared to Raspberry Pi 4 boards), "this single board computer came out to delivering 2.85x the performance of the Raspberry Pi 400 overall." And through all this the average SoC temperature was 71 degrees with a peak of 85 degrees — without any extra heatsink or cooling.

After five years attempting to make Intel into a competitor for Nvidia and AMD in the realm of discrete graphics for gamers and beyond -- with limited success -- Raja Koduri is leaving Intel to form his own generative AI startup. The Verge reports: Intel hired him away from AMD in 2017, where he was similarly in charge of the entire graphics division, and it was an exciting get at the time! Not only had Intel poached a chief architect who'd just gone on sabbatical but Intel also revealed that it did so because it wanted to build discrete graphics cards for the first time in (what would turn out to be) 20 years. Koduri had previously been poached for similarly exciting projects, too -- Apple hired him away from AMD ahead of an impressive string of graphics improvements, and then AMD brought him back again in 2013.

Intel has yet to bring real competition to the discrete graphics card space as of Koduri's departure. [...] But the company has a long GPU roadmap, so it's possible things get better and more competitive in subsequent gens. It took a lot longer than five years for Nvidia and AMD to make it that far. By the time Koduri left, he wasn't just in charge of graphics but also Intel's "accelerated computing" initiatives, including things like a crypto chip.

An anonymous reader writes: Last week, we learned that Microsoft spent hundreds of millions of dollars to buy tens of thousands of Nvidia A100 graphics chips so that partner OpenAI could train the large language models (LLMs) behind Bing's AI chatbot and ChatGPT.

Don't have access to all that capital or space for all that hardware for your own LLM project? Nvidia's DGX Cloud is an attempt to sell remote web access to the very same thing. Announced today at the company's 2023 GPU Technology Conference, the service rents virtual versions of its DGX Server boxes, each containing eight Nvidia H100 or A100 GPUs and 640GB of memory. The service includes interconnects that scale up to the neighborhood of 32,000 GPUs, storage, software, and "direct access to Nvidia AI experts who optimize your code," starting at $36,999 a month for the A100 tier.

Meanwhile, a physical DGX Server box can cost upwards of $200,000 for the same hardware if you're buying it outright, and that doesn't count the efforts companies like Microsoft say they made to build working data centers around the technology.

An anonymous reader shared this report from Phoronix: A pull request of early AMDGPU kernel graphics driver changes was submitted for DRM-Next on Friday as some of the early feature work accumulating for the Linux 6.4 kernel cycle.

Among the AMDGPU kernel driver changes this round are a number of fixes affecting items such as the UMC RAS, DCN 3.2, FreeSync, SR-IOV, various IP blocks, USB4, and more. On the feature side, mentioned subtly in the change-log are a few power-related additions... These additions are largely focused on Van Gogh APUs, which is notably for the Valve Steam Deck and benefiting its graphics moving forward.

First up, this kernel pull request introduces a new sysfs interface for adjusting/setting thermal throttling. This is wired up for Van Gogh and allows reading/updating the thermal limit temperature in millidegrees Celsius. This "APU thermal cap" interface is just wired up for Van Gogh and seems to be Steam Deck driven feature work so that SteamOS will be better able to manage the thermal handling of the APU graphics....

These power features will be exposed via sysfs while Steam OS will wrap around them intelligently and possibly some new UI settings knobs for those wanting more control over their Steam Deck's thermal/performance.

When Microsoft invested $1 billion in OpenAI in 2019, it agreed to build a massive, cutting-edge supercomputer for the artificial intelligence research startup. The only problem: Microsoft didn't have anything like what OpenAI needed and wasn't totally sure it could build something that big in its Azure cloud service without it breaking. From a report: OpenAI was trying to train an increasingly large set of artificial intelligence programs called models, which were ingesting greater volumes of data and learning more and more parameters, the variables the AI system has sussed out through training and retraining. That meant OpenAI needed access to powerful cloud computing services for long periods of time. To meet that challenge, Microsoft had to find ways to string together tens of thousands of Nvidia's A100 graphics chips -- the workhorse for training AI models -- and change how it positions servers on racks to prevent power outages. Scott Guthrie, the Microsoft executive vice president who oversees cloud and AI, wouldn't give a specific cost for the project, but said "it's probably larger" than several hundred million dollars. [...] Now Microsoft uses that same set of resources it built for OpenAI to train and run its own large artificial intelligence models, including the new Bing search bot introduced last month. It also sells the system to other customers. The software giant is already at work on the next generation of the AI supercomputer, part of an expanded deal with OpenAI in which Microsoft added $10 billion to its investment.

An anonymous reader shares a report: Nvidia released a new driver update for its GeForce graphics cards that, among other things, introduced a new Video Super Resolution upscaling technology that could make low-resolution videos look better on high-resolution screens. But the driver (version 531.18) also apparently came with a bug that caused high CPU usage on some PCs after running and then closing a game. Nvidia has released a driver hotfix (version 531.26) that acknowledges and should fix the issue, which was apparently being caused by an undisclosed bug in the "Nvidia Container," a process that exists mostly to contain other processes that come with Nvidia's drivers. It also fixes a "random bugcheck" issue that may affect some older laptops with GeForce 1000-series or MX250 and MX350 GPUs.