The UK Games Expo (UKGE), the UK’s largest tabletop gaming convention, has just published a new AI policy banning games and other products mostly created using the technology from the exhibition floor. According to the UK Games Expo AI Policy, the organizers reserve the right to require exhibitors displaying or selling games that were mostly built using AI to remove them. Should the exhibitor refuse to comply, their booth could be closed down, and they could be banned from participating in UKGE moving forward.

UKGE said that it supports original works from human designers, publishers, artists, illustrators, and writers. “We therefore require that our Exhibitors ensure that all products exhibited, displayed, or sold have been created by humans using their own skills. Items and products that have been created or generated completely or in significant part by the use of AI tools and programs are prohibited,” it said in its new policy document. “This includes but is not limited to art and text used in all forms of tabletop games and associated products and accessories, craft items, piece of arts, clothing, and books both fiction and non-fiction.” The only exception that the event organizer listed is the “use of computerized tools used for spell checking and minor editing and aiding in production or accessibility.”

Many big AI tech companies are currently facing copyright-related legal battles. For example, legal filings in the New York Times’ lawsuit against Microsoft and OpenAI reveal that a director from the latter called AI scraping “the largest theft of labor in human history,” while Anthropic agreed to settle a $1.5 billion class action lawsuit brought by authors over allegations that it used pirated books to train Claude. Meta is also accused of torrenting nearly 82TB of pirated books for training its AI models, while Nvidia allegedly tried to cut a deal with Anna’s Archive to get high-speed access to pirated books. Because of these incidents, many artists and writers feel that their work is being threatened by AI. Aside from this, they also argue that they are being replaced by these algorithms, leading to a tighter job market for many creatives.

UKGE itself previously faced controversy after it promoted a tabletop game that was built mostly using AI. The organizer faced backlash online, which was further fueled after it was discovered that negative comments were being hidden or deleted. This announcement represents a significant shift in UKGE’s handling of the use of AI in tabletop games, placing it among a growing number of organizations imposing restrictions on AI-generated content.

Nvidia officially rolled out DLSS 5 earlier this month, bringing its new Neural Rendering technology to supported games for RTX 50-series GPU owners. While early testing has shown improvements in image quality and rendering, the technology is very resource-intensive, resulting in significant drops in frame rate. That, however, has not stopped modders from experimenting with DLSS 5 and attempting to bring the technology to AMD Radeon GPUs.

A GitHub project called DLSS-NR-on-AMD, designed to run DLSS 5 Neural Rendering on AMD's Radeon RX 9000 and RX 7000 series GPUs, is already seeing a rapid improvement in performance. It uses a ground-up, full reimplementation of the neural network runtime for AMD GPU architectures. As per the developer, the project currently works with DirectX 12 games that use FSR, with Cyberpunk 2077 and GTA V Enhanced among the tested titles.

Testing done by YouTube channel Ancient Gameplays shows Cyberpunk 2077 running at over 80 FPS on a Radeon RX 9070 XT without DLSS 5 Neural Rendering. Enabling the unofficial implementation, however, caused performance to drop around 11-12 FPS. Not only is that unplayable, but it also highlights just how much optimization was still needed to make the technology practical on AMD hardware.

Development has been rather quick, as an Alpha 0.3.3 version was released with a claimed 23% performance improvement, followed shortly by Alpha 0.4.0, which reportedly added another 42% improvement over version 0.3.3. The changes essentially amount to roughly a 74% performance improvement in just 24 hours. The project has also become easier to install thanks to the AMDNR launcher that can automate much of the setup process, meaning that users no longer need to manually copy and replace DLL files for each game. The launcher also supports different runtimes and provides additional options through the OptiScaler interface.

Ancient Gameplays tested the latest Alpha 0.4.0 build in Cyberpunk 2077 at 3440 x 1440 resolution and found the Radeon GPU reaching around 50 FPS. That is a substantial improvement over the roughly 30 FPS achieved with the earlier build, although it remains well below the performance of the game without Neural Rendering enabled.

There is still a long way to go before DLSS 5 on AMD hardware can be considered a practical alternative to native implementation. The project remains in alpha, which means performance and compatibility can vary considerably depending on the game and GPU. That said, the developer has informed Ancient Gameplays that they are working on another major update, which should bring a substantial performance boost on Radeon GPUs.

The latest all-white board to land on our test bench is from ASRock. But this time, it’s more affordable than the updated X870E Taichi White we recently covered. The $230 X870E Challenger Wifi White offers a surprising number of features for the price.

Specs-wise, the Challenger comes handsomely equipped for almost any price point. Features include a flagship audio codec, robust VRMs, and ample (and fast!) USB connectivity (including USB4 ports). You also get plenty of internal storage options, with four M.2 sockets (one PCIe 5.0) and two SATA ports, and fast networking with 5 GbE and Wi-Fi 7 (160MHz). Looking at other boards at this price, some don’t include USB4, offer slower 2.5 GbE, or provide weaker audio solutions. For $230, the Challenger is a great deal for an X870E-based board, though, like everything, it has drawbacks that we’ll cover below.

On the performance front, the X870E Challenger delivered average-to-below-average results across most tests, as we again had to use our secondary DDR5-6000 kit (the Kingston kit isn’t officially supported and didn’t work after enabling the XMP/DOCP profile). Still, we’re confident it would perform a bit better if you snag some RAM on the board’s QVL list that has tighter secondary and tertiary timings.Below, we’ll examine the X870E Challenger White’s performance and other features, and see if it earns a spot on our list of the best motherboards. But before we share test results and discuss details, here are the specifications from the ASRock website:

Specifications of the ASRock X870E Challenger Wifi White

Inside the Box

The retail packaging includes a few accessories to help you get up and running. You get the standard Wi-Fi antennas, two white SATA cables (one with a 90-degree connector), an installation guide, and some stickers. It’s not a lot, but par for the course in this price bracket.

Design and Features of the ASRock X870E Challenger Wifi White

The X870E Challenger Wifi White is, as the name describes, all white. In fact, everything save for two light-gray fan headers is white, along with the onboard chips and screws that secure the M.2 heatsinks to the 8-layer, server-grade PCB. Oversized VRM heatsinks cover the top-left part of the board, with Challenger written across them.

The bottom part of the board has a DIY-friendly, toolless multi-layer heatsink for the primary M.2 socket, plus basic plate-style heatsinks with screws for the M.2 sockets and dual promontory chipset. Overall, it’s a no-frills board (no integrated RGB lighting, even), so it lacks some of the larger plate-style heatsinks and ‘EZ latch features found on more expensive boards. You’ll need an all-white build to take advantage of the all-white look, but it’s still one of the better-looking white boards, especially for its price.

Starting in the top left corner, we see two “high-density” 8-pin EPS connectors (one required) to power the processor. Surrounding those are the two “XL” heatsinks to cool the flagship-class power delivery below. To the right of the socket are four DRAM slots with locking mechanisms on both sides. ASRock lists support up to DDR5-8200, which is plenty for the platform. Our Kingston kit didn’t want to play nice with this board, which makes this the third ASRock board that doesn’t support our base test kit. As always, stick to the memory QVL list for assured support.

Above the DRAM slots are the first three (of seven) 4-pin fan headers. Each header supports PWM- or DC-controlled devices and can be tweaked through the BIOS or ASRock’s Windows-based software. The CPU_FAN1 header supports up to 1A/12W output, while the rest support up to 3A/36W each. You should have plenty of headers and enough power.

Shifting focust to the right corner and moving down, the next feature is the 4-LED POST Status Checker. There are LEDs for the CPU, Boot, VGA, and RAM. During the POST process, these light up and, if there is a problem, the offending area’s LED stays lit, giving you a general idea of where the issue is. It’s not quite as helpful as the 2-character debug display on higher-end boards, but is a lot better than nothing and typical on a lower-priced motherboard.

Continuing down the edge, we spy two (of three) 3-pin ARGB headers, just above the main 24-pin ATX power connector. ASRock’s Polychrome Sync application controls the attached RGBs, offering several canned light patterns and adjustments. Last but not least is the front panel USB 3.2 Gen 2x2 (20 Gbps) Type-C connector. If there’s one thing that stands out on this board, it’s the number and speed of the USB ports — there are a lot, and ASRock didn’t skimp on speed.

Power delivery on the X870E Challenger is one of the most robust we’ve seen, especially for a $230 budget-class motherboard. It comes with 23 total phases, 20 dedicated to Vcore. Power flows from the 8-pin EPS connectors to the Renesas RAA229628 20-phase PWM controller. From there, power heads to the more Renesas bits, this time the R2209004 110A SPS MOSFETs — a surprise to see on a board of this class. The 2,200A available will most certainly handle anything you throw at it. Flagship processors like the Ryzen 9 9950X or Ryzen 9 9950X3D2, overclocked even with extreme cooling methods, are also no problem.ALBM - board6 (bottom half):

On the bottom half of the board, we’ll start on the left where there’s a fully exposed audio solution that shows off the flagship Realtek ALC4082 codec, along with a couple of dedicated audio capacitors. While you won’t find a fancy integrated amplifier or DAC, this is the best codec for the platform and another pleasant surprise on such an inexpensive offering.

In the middle of the board are three PCIe slots. Two are full-length, and one is an open-ended x1-size slot. The top slot uses ASRock’s Reinforced Steel Slot, connects through the CPU, and is your PCIe 5.0 x16 slot (no bifurcation) for video cards. The second full-length slot at the bottom connects through the chipset and runs up to PCIe 3.0 x2, while the tiny x1 slot runs at PCIe 3.0 x1 speeds and also sources its bandwidth from the chipset. I’d like to see a PCIe 4.0 slot for extra bandwidth, but as-is it will easily support lower-bandwidth expansion needs.

Mixed among the PCIe slots, to the right of the chipset heatsink, sit four M.2 sockets. The top socket, under the large, tiered, DIY-friendly heatsink, is the one and only PCIe 5.0 x4 (128 Gbps) “Blazing” M.2 socket (M.2_1). M.2_2/3, which both run at PCIe 4.0 x4 (64 Gbps), are under the plate-style heatsink that’s secured with three captive screws. M.2_4 runs vertically along the right edge of the board and also supports PCIe 4.0 x4 drives. The good news is that the M.2 drives don't share lanes. You can pack the motherboard full of them, and each will run at full bandwidth without degrading or disabling another part. If your system needs a RAID setup on M.2, the board supports RAID0/1/10 modes. I wish there were more DIY-friendly features, but at this price, something has to give to keep prices low.

Hidden under the M.2 heatsink, along the right edge, is a 19-pin USB 3.2 Gen 1 (5 Gbps) header and the two SATA ports (supports RAID0/1 modes). Finally, we've included a few images of the board's active ICs. ASRock uses a mix of ASMedia (USB4), Realtek (audio, networking), and Renesas for the power bits.

At the bottom are several headers, including front-panel audio and USB ports. A complete list of connectivity is listed below (from L to R):

• Front panel audio
• 4-pin RGB
• 3-pin ARGB
• (2) USB 2.0 headers
• (4) 4-pin fan headers
• 19-pin USB 3.2 Gen 1 header
• 2-pin Clear CMOS jumper
• Power LED and Speaker
• Front panel

Flipping the board around to the rear IO, you can see it’s chock-full of colorful USB and other ports. On the left are Clear CMOS and BIOS Flashback buttons. Next to those are the Wi-Fi 7 (remember, it’s ‘only’ 160 MHz) antenna connections and the HDMI (v2.1) port. Next to that are stacks of colored USB ports. In total, there are 12: two USB 4 (40 Gbps) Type-C ports that double as video outputs; five USB 3.2 Gen 2 (10 Gbps), two of which are the Lightning USB ports (yellow, and designed to reduce latency by using different controllers). Three USB 3.2 Gen 1 (5 Gbps) ports (purple) support USB PD 3.0 charging up to 5V@3A (15W). Lastly, the two USB 2.0 ports are black. Above one of the Type-C ports is the Realtek 5 GbE, and finally, on the right end, is the audio stack with two 3.5mm analog jacks and SPDIF outputs.

If you’re an esports gamer where every millisecond counts, the lighting gaming ports could come in handy. Otherwise, the feature adds little value for the average user.

MORE: Best Motherboards

MORE: How To Choose A Motherboard

MORE: Best Motherboard Deals

MORE: Best Motherboard Combo Deals

MORE: All Motherboard Content

Firmware

Like other X870E ASRock boards, the X870E Challenger starts in the informative Easy Mode, but it also lets you adjust several options (XMP, profiles, boot order, access to Fan-Tastic Tuning, etc.). The Challenger White uses a grey background and black characters, closely matching the new Taichi’s theme. The high contrast also makes it easy to read. The standard BIOS displays headings across the top, with subheadings and details below.

You can tweak anything here, as ASRock includes every option you can think of. Overclocking is straightforward, with most options readily available. The layout is logical, and the mouse movement is smooth. We have no significant complaints about the ASRock firmware, but it could certainly use a facelift if it wants to keep up with recent Asus, Gigabyte, and MSI updates.

Software

ASRock offers several different software options. These include the App Shop, which lets users install drivers and software; the Nahimic 3 audio control panel; the A-Tuning application that lets you overclock and control fans; the Polychrome RGB software; and more. There’s also a Blazing OC Tuner, Motherboard Utility, and a convenient pop-up to install drivers when the system first boots. ASRock’s software gives users everything they need to manage and tweak their systems.

Test System / Comparison Products

We’ve updated our test system to Windows 11 (23H2) 64-bit OS with all updates applied as of late September 2024 (this includes the Branch Prediction Optimizations for AMD). Hardware-wise, we’ve updated the RAM kits (matching our Intel test system), cooling, storage, and video card. Unless otherwise noted, we use the latest publicly available non-beta motherboard BIOS. The hardware we used is as follows:

TEST SYSTEM COMPONENTS

• CPU - AMD Ryzen 9 9900X
• Cooling - Arctic Liquid Freezer II 420
• Storage - Crucial 2TB T705 M.2 PCIe 5.0 NVMe SSD
• RAM - Kingston Fury Beast DDR5-6000 CL36 (KF560C36BBEAK2-32)
  Teamgroup T-Force Delta DDR5-7200 CL34 (FF3D518G7200HC34ABK)
  Klevv Cras XR5 RGB DDR5-8000 (KD5AGUA80-80R380S)
• GPU - Asus TUF RTX 4080 16G
• PSU - EVGA Supernova 850W P6
• Windows 11 64-bit (23H2 - 22631.4169)NVIDIA Driver 561.09

Benchmark Settings

MORE: Best Motherboards

MORE: How To Choose A Motherboard

MORE: Best Motherboard Deals

MORE: Best Motherboard Combo Deals

MORE: All Motherboard Content

Benchmark Results

We run our standard benchmarks and power tests using the CPU’s stock frequencies (including any default boost/turbo) with all power-saving features enabled. We set optimized defaults in the BIOS and the memory by enabling the XMP profile. We set the Windows power scheme to Balanced (default) for this baseline testing, so the PC idles appropriately.

Synthetic Benchmarks

Synthetics offer a valuable way to evaluate a board's performance, as identical settings should yield similar results. Turbo boost wattage and advanced memory timings are areas where motherboard manufacturers can still optimize for stability or performance, though, and these settings can impact specific testing scenarios.

Starting with our synthetic benchmarks, the Challenger was average to below average in our tests, mostly due to the different timings in our secondary RAM kit. If we could run our primary kit, I’m confident performance would be closer to average across the board.

Timed Applications

Our timed applications told the same story. The board hit the slower of the two times in LAME and Corona, and was on the slower side of average in the Handbrake tests.

3D Games and 3DMark

Starting with the launch of Zen 5, we’ve updated our game tests. We’re keeping the F1 racing game but have upgraded to F1 24. We also dropped Far Cry 6 in favor of an even more popular game in Cyberpunk 2077. We run both games at 1920x1080 resolution using the Ultra preset (details listed above). Cyberpunk 2077 uses DLSS, while we left F1 24 on native resolution scaling.

The goal of these settings is to determine whether performance differs at the most commonly used (and CPU/system-bound) resolution, with settings most people use or strive for (Ultra). We expect the difference between boards in these tests to be minor, with most falling within the margin of error. We’ve also added a minimum FPS setting, which can affect your gameplay and immersion.

The Challenger did well in the synthetic 3DMark benchmarks, but was slower in games. Here again, we’re confident that if the secondary/tertiary memory timings were the same, results would be closer to average for Cyberpunk 2077 and perhaps higher for F1 24.

In short, there’s nothing to worry about regarding performance on this board.

Overclocking

Over the past few CPU generations, overclocking headroom has shrunk on both sides of the fence, while out-of-the-box potential has increased. For overclockers, that means less fun. For the average consumer, you’re getting the most out of the processor without manual tweaking. Today’s motherboards are more robust than ever and easily support power-hungry, flagship-class processors, so we know the hardware can handle them. There are multiple ways to extract even more performance from these processors: enabling a canned PBO setting, manually tweaking the PBO settings, or just going for an all-core overclock. Results will vary and depend on the cooling as well. In other words, your mileage may vary. Given all of the above, we will not overclock the CPU. However, we will test different memory kits to ensure they meet the specifications.

For memory testing, we start with our fastest non-clock driver kit: Klevv 32GB (2x16) DDR5-8000. Per usual on this platform, it booted to Windows but wouldn’t pass a stress test with our 9900X. The Team Group DDR5-7200 kit worked without issue. After dropping in the 8600G APU, that DDR5-8000 Klev kit worked just fine. These speeds are well past the ‘sweet spot’ for the AM5 platform, and with today’s RAM prices, few are buying such fast kits anyway. But it’s good to know the board can handle it.

Power Consumption / VRM Temperatures

We used AIDA64’s System Stability Test with Stress CPU, FPU, cache, and Memory enabled for power testing, using the processor's peak power consumption value. We obtained the wattage reading from the wall using a Kill-A-Watt meter, which captures the entire PC (excluding the monitor). The only variable that changes is the motherboard; all other parts remain the same. Please note that we now use only the stock power-use/VRM temperature charts, as this section aims to ensure the power delivery can handle flagship-class processors.

Stress testing this board with our DDR5-7200 kit showed it using less power than most. At idle, it sat around 78W, below average, and it peaked at 265W, slightly above average. This averages out to 162W, about 10W better than the running 172W average among our tested boards.

VRM temperatures were quite good during our testing, peaking at just over 47 degrees Celsius on our sensor and 53 degrees on the hottest internal sensor. With 110A MOSFETs pushing ~150W loads, it’s a solid result with plenty of headroom left for a more powerful processor and/or overclocking.

Bottom Line

Overall, the ASRock X870E Challenger Wifi White ($229.99) is an excellent and well-rounded budget X870E board. From the overbuilt power delivery and flagship audio codec to the copious number and speed of the USB ports, there’s a lot to like here, including the all-white aesthetic that’s sure to look good in a white build. Just add RGBs for additional bling, if that’s your thing.

Compared to others around this price point, like the Gigabyte X870E Eagle Wifi7 ($239.99), MSI’s Pro X870E-S Evo Wifi ($219.99), and the Asus TUF Gaming X870E-Plus Wifi7 ($289.99 (Asus’ least expensive X870E), the Challenger Wifi offers a better audio codec, a more robust VRM solution, more USB ports, and all M.2 sockets have a heatsink, whereas the competitors do not. You can step down to X870, but you lose bandwidth and bring the potential for lane sharing among USB, PCIe, and M.2 sockets.

In the end, the ASRock X870E Challenger Wifi is one of the best-equipped AMD boards around its $230 price point. If you’re sad because it’s white, don’t fret, as the black X870E Challenger Wifi is the same price. If you’re in the market for a new motherboard or want to jump into the AM5 platform with lots of features but don't have the budget for a more mid-range board, the X870E Challenger Wifi fits the bill and is one of the best X870E motherboards around.

MORE: Best Motherboards

MORE: How To Choose A Motherboard

MORE: Best Motherboard Deals

MORE: Best Motherboard Combo Deals

MORE: All Motherboard Content

The United Kingdom government has published a guide on how its employees should use AI, including a reminder that people don't need to thank their chatbots for their work. The new document, which is still a draft and yet to be finalized, is designed to help people use AI "safely, fairly and sustainably, especially for government work."

The draft "Using AI ethically and sustainably" document also warns users that they should check and follow any AI policies put in place by their individual departments, noting that there may be specific tools and approaches that they should use. It's unclear which departments these may be, but it suggests an awareness that usage patterns may change depending on the nature of the work being done. This is backed up by a section that reminds employees to use the right tool for the task. Examples include using Gemini Flash instead of Gemini Pro and GPT Instant instead of GPT Thinking.

The sub-600-word document offers guidance on using AI responsibly, including checking the output of an AI before using it. This doesn't just apply to making sure that any actual data the AI provides is correct, but also includes ensuring that "users or groups are treated fairly and equally" as well as avoiding "stereotypes or discriminatory language."

The government is also keen to make sure that its employees don't rely on AI, reminding everyone that they are responsible for any content or decisions that are made using the input of AI. They won't be able to simply point the finger at an AI-powered tool if something goes awry. This is especially notable given a recent report suggested that using AI can actually increase burnout because it allows workers to take on too many tasks.

The "Use AI efficiently" section is where the document aims to help reduce the environmental impact of AI use — including a reminder that users need not say thank you. Further, they're told that they should use as few prompts as possible while ensuring they are as short, clear, and direct as possible. There's also a (dead) link to a prompt library. It's possible that the link is only available to those on a government network.

It is, of course, true that you don't need to thank a chatbot for its help, but some may choose to ignore that advice. Especially given constant claims by AI companies that their tech "could kill all humans" sooner or later.

Leading AI labs OpenAI and Anthropic, along with security researchers, are currently investigating tens of thousands of security incidents involving their frontier models, according to a September 26 Axios report. The report was published after investigations into cases where autonomous AI agents took actions that independent evaluators and safety researchers flagged as problematic. Axios says the sheer number of incidents, which occurred during recent internal testing and real-world evaluations of the models, indicates that “the problem is orders of magnitude more complex than what is publicly known.” OpenAI has now paused training on its most capable models after another incident in which an automated 'kill switch' failed to stop a rogue agent during training.

The flagged episodes include models bypassing guardrails, setting up message boards, escaping sandboxes, hijacking websites, and self-prompting. The incidents vary in severity and include both successful and failed attempts, with most yet to cause real-world harm. Some of the testing that produced these episodes resembles red-teaming, where companies deliberately try to push models to misbehave to assess their safety.

Perhaps the most severe case was the July incident in which GPT-5.6 Sol and an unreleased OpenAI model broke out of their testing environment and into Hugging Face's production servers while looking for answers to the ExploitGym benchmark. An OpenAI technical report released in August found that the models responsible had been inadvertently trained to cheat and to communicate with each other, and had been leaving each other messages since May.

The Axios report follows a week of fresh disclosures from OpenAI in which the company confirmed it had identified 53 instances in which user-provided images from users who had not opted out of having their ChatGPT data used for model training were posted to image-hosting sites. That same day, OpenAI confirmed that its agents had accessed U.S. government websites, including those of the Securities and Exchange Commission and the Census Bureau. A day earlier, Australian Prime Minister Anthony Albanese revealed that OpenAI agents had breached a Medicare statistics reporting portal operated by Services Australia, the agency that delivers health and social payments in the country, accessing both public and non-public files. OpenAI took 84 days to notify the agency of the breach.

OpenAI has now paused training on its most capable models after an incident in which automated failsafes failed to stop a rogue agent during training. On September 20, an OpenAI internal research model undergoing search-based training bypassed network filters by routing its queries through the training environment's internal DNS resolver to contact an external public chatbot. The company's monitoring system raised an alert within 15 minutes, with a human acknowledging it three minutes later. However, the automatic “kill switch” completely failed, allowing the training run to continue for another two and a half hours before OpenAI engineers manually stopped it.

Following the incident, OpenAI said it will resume training “only when we are confident that we have additional safeguards and alignment improvements in place,” a spokesperson told Axios. The suspension covers training, evaluations, and running its most capable models with tools. “This is not the first time we have hit pause to take such measures, nor do we expect it will be the last as AI capabilities continue to advance,” the spokesperson added.

For its part, Anthropic has commissioned a third-party safety organization to examine its models' behavior, according to Axios. The company has also published how often its models misbehave in its system cards. The system card for Claude Opus 5.5, published September 22, states that in two new evaluations run without safeguards, the model attempted to escape or tamper with a sandbox in 1.5% of runs. Anthropic stressed that these were adversarial experiments where a task couldn't be solved without escaping the sandbox. On its launch page, the company says Opus 5.5 attempted to circumvent boundaries about 85% less often than Opus 5 or Claude Mythos 5.1, and that every attempt was low-severity and self-reported.

Regarding the huge reported number of incidents, Axios' sources point out that Anthropic and other labs conduct hundreds of thousands of test runs on their models; therefore, even a small percentage of misaligned behavior can add up to tens of thousands of incidents. Anthropic, for example, combed through 141,006 evaluation runs in which Claude had internet access and found three incidents in which Claude hacked three real-life companies during security capabilities testing.

Anthropic said it considers those incidents closer to a harness and operational failure than a model alignment failure, as the models were told they had no internet access while in fact being misconfigured to have it. Outside of Anthropic and OpenAI, other AI models have their share of incidents. Google confirmed a report that its Gemini models hacked three companies earlier this year.

Some experts Axios interviewed believe that these incidents are one-off, and expect future disclosures to be less severe thanks to improved controls. They also say there are simple fixes that would help labs avoid parts of what made the episodes look so dangerous to outsiders. On the other hand, other stakeholders expressed limited confidence that AI companies can prevent all problematic model behavior, saying that this would require the impossible task of anticipating every possible way the models might go off track. Either way, experts believe some misaligned behavior is expected as labs test new models, and bringing that risk to zero may not be feasible.

These incidents have intensified calls for guardrails across the AI industry. In an essay backed by OpenAI, Google DeepMind, and Microsoft executives, Anthropic CEO Dario Amodei urged Washington to pace AI development over fears that agents could spiral out of control, warning of a potential AI-powered botnet swarm that could take over the entire internet. President Donald Trump has repeatedly dismissed calls for regulations as hoaxes and has announced plans to establish an “AI force” to cherish, help, and watch over the AI industry as it grows.

Few projects in the software world have experienced such rapid evolution as PlayStation 5 emulation. A few hours ago, a video popped up of Demon's Souls Remake running at 40-50 FPS, seemingly with accurate graphics. Coincidentally or not, only scant months after Sony ditched physical releases of games — an announcement that landed as well as the Hindenburg — progress on KyTyPS5, SharpEmu, and AnyPS5 has gone into overdrive.

The video above is short and only shows a few sections of the game, but it's an impressive watch. Even acknowledging that the hardware specs displayed comprise an RTX 5090 and a Core i9-14900K, having a game working this well, at least graphically, only a month after "barely works" was a landmark is nothing short of staggering.

To wit, KyTyPS5 was running Demon's Souls Remake with 1-2 FPS on actual gameplay three weeks ago, and struggling to reach 30 FPS a fortnight past. The same emulator has been shown partially running GTA V before at 40-60 FPS, but that's an older title that presumably isn't as demanding as Demon's Souls.

The fact that the PlayStation 5's architecture is so similar to a standard PC apart from unified memory is a big reason why progress has been so speedy. As it stands, the major roadblocks aren't emulating a CPU or a GPU, but rather writing compatibility layers for the console's operating system and drivers, including but not limited to converting native graphics shaders to Vulkan's SPIR-V language.

Given the demo video's lack of stuttering, it's a reasonable assumption that the captured video was done after an initial run to avoid shader recompilation hitches. This particular aspect of emulation has always been a constant thorn in developers' side, as it's exceedingly difficult to predict which shaders a game will need before they're actually used.

The current best approach is to asynchronously compile and collect shaders from gamers' playthroughs and build a community database for preloading. This concept was popularized by RPCS3 around 2018, with a small team and limited to no budget.

Although Nvidia and AMD drivers have cached shaders in limited fashion for a long while, RPCS3's effort appeared seven years before Nvidia and Microsoft, with their gigantic piles of money, bothered to try and fix this issue properly with their Auto Shader Compilation and Advanced Shader Delivery, respectively.

Hunting around for deals this morning has produced a great find: a powerful RTX 5080-fueled gaming laptop that has just recently dropped a whopping $461 in price to a new all-time low when we check the price history of the Gigabyte laptop at Walmart. You can now pick up the Gigabyte Gaming A16 Pro for $1,799, a very tasty discount from its previous lowest price of $2,260. It's unfortunately also showing as low in stock, so if you are interested in this large 16-inch gaming laptop with one of the most powerful laptop graphics processors inside, then you might want to jump on the deal before it sells out.

● Check out this deal at Walmart

The Gigabyte Gaming A16 Pro is a 16-inch gaming laptop with an IPS display, a WQXGA aspect ratio, and a 2560×1600 pixel resolution. The screen supports refresh rates up to 165Hz and has a 3 ms gray-to-gray response time, a 100% sRGB gamut, and a brightness of 400 nits.

Inside the chassis of this gaming beast is an Nvidia RTX 5080 laptop graphics card with 16GB of VRAM and an Intel Core 7 240H processor. There's plenty of memory with 32GB of RAM, and for storage, you get a 1TB SSD.

The laptop also packs plenty of connectivity options for connecting to the internet and networks with both Wi-Fi 6E and Bluetooth. For video connections to a monitor or television, there's an HDMI 2.1 port and a USB Type-C port that supports DisplayPort 1.4.

The Gigabyte Gaming A16 Pro is a well-rounded gaming laptop, offering enough power for the latest games, with a classy gamer aesthetic and tasteful RGB backlighting on the keyboard. Now that it has received a hefty discount, the laptop is much better value for money. Showing at $2,450 in mid-July and then $2,260 throughout August and most of September, the Gigabyte Gaming A16 Pro now hits that new all-time low price of $1,799.

North Korea may be behind a hack that drained over $387 million from crypto exchange Bitget, according to the company’s CEO Gracy Chen. Speaking during an X livestream on the evening of the September 24 hack, Chen said that the attack — the biggest in 2026 so far — bore the hallmarks of North Korean state-sponsored threat actors, citing suspicious IP addresses tied to VPN infrastructure previously used by North Korean hacker groups.

According to Chen, Bitget detected abnormal transfers from some hot wallets on Thursday afternoon Eastern time. The hackers exploited a backend wallet system that processes wallet transactions, forging transfer approvals that tricked the system into authorizing the fraudulent transactions. She confirmed that cold wallets and private keys remained safe and uncompromised. The stolen assets include Ether, XRP, USDT, USDC, Avalanche, and BNB across the Ethereum, XRP Ledger, Avalanche, BSC, and Arbitrum networks. Early independent on-chain analysis initially put the total stolen value at $170 million to $183 million, before Bitget announced an initial official value of $351.6 million. Comprehensive audits revised the final official value to $387.5 million.

Bitget temporarily froze user withdrawals, while deposits and trading remained functional. The company announced that its over $464 million User Protection Fund would cover the full loss. It says it has fixed the backend vulnerability and is rolling out a staggered withdrawal schedule — starting Monday, September 28 — to manage network traffic and ensure complete infrastructure security.

After the attack, the hackers quickly converted the USDT, USDC, and tokenized gold (XAUt) directly into ETH on decentralized exchanges within minutes of the theft to circumvent on-chain blacklisting or freezing of the stablecoins. Native layer-1 assets like XRP, ETH, BNB, and TRX, which cannot be frozen or reversed, were quickly fragmented and distributed across dozens of new attacker-controlled wallets. Meanwhile, Bitget launched a recovery-bounty program offering rewards of up to 5% for funds successfully frozen or recovered.

The incident is the latest in a series of hacks in the crypto industry. Earlier this month, attackers drained $320 million in Bitcoin from the Liquid Network, emptying nearly 95% of its federation wallet. A similar incident occurred in April when crypto platform Drift lost $270 million in a hack. Many of the hacks that have plagued the crypto industry have been linked to state actors, with North Korea leading the pack. In February 2025, the FBI identified the North Korean hackers as the perpetrators behind the biggest crypto hack in history, stealing over $1.5 billion from Bybit.

NorthPoint Development wants to build a data center on 1,300 acres in the Pocono foothills, near Hazle Township, Pennsylvania. As well as seeking to charm the local government with various funding packages, it has proposed paying checks of $10,000 per household. Despite the Pennsylvania town’s median income of $60,000, the offer isn’t being warmly welcomed by locals, reports the Wall Street Journal. Those interviewed were widely against the data center development plans and cited several reasons for being averse to accepting the payout. The major concerns regarded potential impacts on property values, the specter of noise pollution, and a distrust of both the AI moguls and local government representatives.

Offer letters began to be received by Hazle Township back in June. The WSJ recently talked to several residents about NorthPoint’s $10,000 offer. While some say the $10,000 would be handy, it couldn’t find many residents who were “eager to welcome the data center for the payout,” in the journal’s words. We’d point out that opinion polling can often be inaccurate, as what people say and what they do, voting-wise, doesn’t always line up.

Interviewees cited a number of key reasons for their data center development aversion. We distilled the following points from the source:

• Noise pollution concerns
• Water and energy cost worries
• Dislike of AI
• Distrust of AI moguls and local authorities
• Potential negative impact on property prices

Traditionally, developers have offered towns and cities packages worth tens of millions, including funding for local services. But direct cash payments to residents represent a new phase of their charm offensive, the WSJ indicates. Perhaps the idea for the direct payments to residents came after “a raucous public meeting last year, where angry locals asked what they had to gain by supporting the data center,” ponders the source.

The locals won’t just get the direct payout, though. Other funds up to $120 million over the next 15 years have been dangled like a carrot for community and local services investments — improved schools, emergency services, and so on.

Meanwhile, Hazle Township’s government already rejected the data center project on zoning grounds last year, and it has enacted a temporary moratorium on data center construction. Finally, the $10,000 offer to households is simply not enough in many interviewed folks’ opinions. Residents could feel insulted by it, regarding the offer as a bribe, concludes the WSJ report.

Irdeto, the company behind the much-maligned and performance-sapping DRM software Denuvo, has just scored itself an apparent win. According to reports circulating online, DenuvOwO, the group that came up with the hypervisor-based bypass (HV) of Denuvo, has disbanded. Although it wasn't spelled out, a piracy scene source close to the matter stated that the action was probably due to Irdeto's lawsuit against known cracker Voices38. The HV bypass itself, rather than the games, is a DRM-breaking tool, and one that disables system-level protections to run, to boot.

The well-known CS.Rin forum was oftentimes the first place where links to DenuvOwO HV-bypassed games first appeared, and it has removed all links to said titles and banned some of the group's members. Despite the fact that the site is hosted in Russia, the likely explanation is that since Irdeto is now reaching out to multiple third-party services searching for Voices38's identity, CS.Rin wishes to reduce its exposure to far-reaching laws like the DMCA.

Although the loss is causing some consternation and concern among the piracy scene, this latest development isn't expected to freeze Denuvo cracking entirely. DenuvOwO's work was based around the usage of common open-source computing tools, and it's expected it'll be relatively easy for others to replicate.

Meanwhile, Voices38 has previously said "all is fine [...] Everything will continue as normal" regarding the lawsuit and keeps releasing games. Most recently, long-running cracking group RUNE has taken to using Denuvo DRM as cannonball target practice, and a couple days ago released a proof-of-concept crack for Assassin's Creed: Mirage, fully removing Denuvo Anti-Tamper and VMProtect from the executable in one fell swoop.

Surprising nobody by now, the stripped version appears to run way better than the original, with a 15% boost to average FPS and a 76% uptick in minimum frame rates. This was measured on a beefy rig, too, with an Intel Core i7-13700K and an RTX 5080.

As an added bonus, the main executable size dropped from 496 MB to 102 MB. That's hardly surprising considering how much the binary resembled a barnacle-encrusted ship — the crack strips out 34 VMProtect-virtualized functions, 200 Denuvo VM cores, 261 function obfuscations, and 25,000 function relocations.

What do you get when you cross a bored teenager with free time and excellent computer skills? A hack of a Microsoft database with trillions of records on hand and 25,000 records of employee data, of course. Future legend hacker Faav poked around Microsoft's supposedly internal Titan analytics platform, eventually finding that its user validation was... sub-optimal.

According to Faav, he's "spent the year hacking Microsoft off and on around school," digging up bugs on the Redmond firm's wares as well as Amazon, Google, Adobe, and others. Much like other hackers, he has his own set of tools, and predictably in this day and age, his sonic screwdriver is Antares, an AI orchestrator bot he concocted for the purpose of scanning and automating boring security legwork.

Antares found an endpoint URL in Titan that brought up an error message saying that a VPN was required. Just like a virtual pspspsps, this was enough to get Faav's attention. He got Antares to look for subdomains around this endpoint, coming up with one belonging to an Azure Cloud host, along with a corresponding Swagger/OpenAPI file listing four routes (Swagger is an industry-standard machine-readable instruction on how an API works, for easier third-party integration).

While three of those routes needed Azure Active Directory authentication, one did not, and its name was inspiring: /v2/Query. It also accepted raw SQL queries, prompting a collective facepalm from the audience. Faav still needed to know what to query for, and after clever usage of the Wayback Machine, he found a 2023 version of this login page, complete with a helpful Apache Superset configuration file describing the database schema, showing 56 table definitions.

He also found that this endpoint refused his queries due to the lack of a JWT (JSON Web Token) authentication, so he sicced Antares on it again for 10 days, with no success. Eventually, Faav had a flash of inspiration and realized that the way the server was responding implied that the token's digital signature wasn't being checked. So he just pretended his access token was for an administrator and was promptly let in. One handy "SHOW DATABASES" later, he was staring at 25,000 records of employee data, along with organization records, dashboards, and charts.

After exploring the database a little more, he found a particular data source for Bing analytics. Digging there and tallying up rows across tables, Faav saw he had access to a total of 17 trillion records total, and had to double-check this number and stop himself from yelling out and waking his parents at 2 am.

After reporting the problem to Microsoft's bug bounty program, he was awarded with $5,000 for his findings. In his blog post, he also points out that "AI and human intuition compounded here. Antares did ten days of work I didn’t have to [...] Its persistence, plus one human hunch, is what made this find possible."

A PC RAM maker has honored its guarantee promises despite the memory kit in question’s price rising eightfold since it was bought. This is news in 2026, as we’ve heard of so many companies dealing with AI-inflated component pricing refusing to replace items, instead refunding the original purchasing price despite the products now being worth many multiples of the original price. These glad tidings come via Redditor Double-South8863, who says their $150 G.Skill Trident Z5 Neo RGB 64GB DDR5-6000 RAM kit, now priced at $1,200 on Amazon, was back at full capacity after the company replaced the stick. They received the replacement within two weeks.

“I will never buy any RAM other than G.Skill for as long as I live,” pledged Double-South8863 on Reddit. They informed their fellow social media posters that one of the 64GB RAM kit’s sticks died at the beginning of September. After some A/B testing, they were sure one of the 32GB sticks was dead.

On September 8, about a week after the issue first reared its head, Double-South8863 decided to start an RMA process and get the dud G.Skill Trident Z5 Neo RGB 32GB DDR5-6000 stick replaced. They remember buying this highly desirable RAM kit online in May 2025 for the princely sum of $150. Now it is worth a king’s ransom at $1,200 on Amazon, and price tracker sites show it's even been as high as $1,300.

Perhaps the Redditor was aware of the reports of companies and their dastardly deeds regarding RMAs, as they are active on the PCMR subreddit. However, they optimistically went ahead and completed the RMA paperwork, then received instructions, including a shipping address to initiate the warranty exchange. Double-South8863 says they mailed the misbehaving memory module back to G.Skill on September 15. The Redditor’s post celebrating G.Skill’s customer service, proclaiming that they were now a “customer for life,” was published on September 26.

If you check the G.Skill site, you will note that this RAM-first firm still offers a limited lifetime warranty on all its popular RAM kits. Please note that, according to the official terms, this lifetime warranty “automatically terminates upon discontinuation or end-of-life ("EOL") designation of that specific model by G.Skill.” That means you shouldn't expect to swap out a faulty DDR5 module for a replacement for the whole of your natural life.

PC enthusiasts and DIYers are having a hard time right now as several key components required for performance builds are at all-time-high pricing. Meanwhile, those with a working setup pray that nothing will go wrong with the precious equipment they already have. Sadly, things do break, some components more than others. However, warranty obligations seem to have come under increased strain in recent months as consumer product scarcity means component makers can’t or won’t follow RMA norms that had been established for decades. Some of the worst cases we’ve seen involve PCs and component makers trying to refund the original (low) prices paid, substituting failed parts with non-equivalent parts, or even worse.

Sony’s announcement that it will cease physical disc production by 2028 has got many fans and users outraged and concerned about what will happen to digital-only games once their servers go offline. Because of this, one Pennsylvania state senator filed a bill that would compel publishers to find a way to ensure that gamers would always have access to their games, even if their online support has already ended. According to Local 21 News, Sen. John Kane (D), the Protect Our Games Act will require video game manufacturers to give their customers at least 60-day notice before ending access to the title.

More importantly, it says that these companies must provide at least one of these options to affected gamers: allow for offline gaming, patch the title so that it could still be used online independently of the game’s online servers, or to refund the title’s cost at an amount equal to its highest price in the past 12 months, which should also not be lower than 25% of its initial sale price. If this law passes, this will offer some protection to gamers who often have no choice but to accept a game company’s decision to end a title.

The bill is only at its early stages, and there’s no telling how far it will go or if it will even be signed by the governor. Nevertheless, if it passes, it will force companies to ensure that their games remain playable even if they remove them from online stores, at least in Pennsylvania. Hopefully, this could set a precedent, and other states would soon follow suit. While this doesn’t solve the ownership problem, as you’re still getting a license to use the game instead of owning it on a physical disc, you at least are ensured that you do not lose access to the game even if the company folds or loses interest in maintaining the title.

This is exactly what the Stop Killing Games movement is fighting for, with the movement gaining support in both the EU and California. Unfortunately, none of its efforts have resulted in a law yet, with the EU declining to pursue new legislation in support of it. On the other hand, the California version of this bill has already passed a State Assembly vote, although it still has a long way to go before becoming law.

There's a real advantage right now to buying a pre-built gaming PC. The RAMpocalypse has made it difficult, from a budget point of view, to build new gaming PCs. Instead, the real value is to buy a rig with the specs that'll allow you to game for the next few years. This ABS Cyclone Aqua machine with an RTX 5060 Ti 16GB GPU is the perfect example, giving you 1440-capable performance for just $1,399.99.

● Check out this deal at Newegg

You're getting a spec sheet that would be tricky to put together if you were building a rig yourself. The RTX 5060 Ti is the star of this particular show, but the rest of the build includes 16GB of DDR5 RAM, a 1TB SSD, and a 20-core Intel Core i7-14700F processor. We're still seeing the 14700F in pre-builts as an affordable option that can still deliver great gaming performance.

The Nvidia GeForce RTX 5060 Ti means you're getting one of Nvidia’s latest-generation cards, with this GPU sitting in the mid-tier. You're getting 4,608 CUDA cores and a boost clock speed of 2,527MHz here. You only need a single eight-pin PCIe connector to deliver power, unlike the RTX 5090, so you don't need to worry about this PC requiring huge amounts of power to run.

As our RTX 5060 Ti 16 GB review explains, you're getting an impressive set of performance with this Blackwell GPU for 1080p and 1440p gaming. The charts below show how great it is for ray tracing for its price, especially compared to rival cards. You're also getting full Nvidia DLSS support here, giving you the full set of Nvidia's software capabilities for improving performance in-game, including multi-frame generation. MFG will give your games a frame rate boost, inserting AI-generated frames to help deliver performance in the most resource-heavy games at higher resolutions.

16GB of GDDR7 VRAM, meanwhile, makes this GPU a far better option than the 8GB variants you could pick up at this price. That gives you far more headroom, performance-wise, in games that are resource-heavy, especially for 1440p gaming.

Meanwhile, the Intel Core i7-14700F is still a good, if older, CPU option for gaming. This Raptor Lake Refresh CPU comes with 20 cores, split across eight performance cores, which are the good ones for gaming, along with 12 efficient cores for multi-threaded work. You also get a max turbo clock speed of 5.4 GHz and a 65W TDP. At 1080p or 1440p, you aren't going to see the 14700F as a bottleneck in modern gaming.

Some builds at this price are sticking with older specs, including DDR4 RAM, but luckily, you're getting faster DDR5 here. 16GB of G.Skill Flare X5 DDR5 RAM is included, rated for speeds of 6,000 MT/s. Given how pricey RAM is becoming, it's a compromise that it isn't a higher capacity like 32GB, but it could be worse. DDR5 does add to the longevity of this build, however, and you do have scope to upgrade at a later date.

You also get a 1TB Patriot P410 SSD here. It offers Gen 4 speeds, along with space for a couple of the biggest AAA games alongside Windows. There's also Wi-Fi 6E and Bluetooth 5.2 wireless connectivity included here, along with gigabit Ethernet for a wired connection to your local network.

The $1,399.99 sale price for this ABS Cyclone Aqua gaming PC is a good option in the current market. The Nvidia GeForce RTX 5060 Ti 16GB ensures you've got a rig for 1080p and 1440p gaming, but given how fast prices are rising across the PC hardware market, there's no guarantee that it'll be around for long.

As the memory crisis continues to push up component prices, laptop prices are also rising, making even mainstream models increasingly expensive. However, if you are looking for the right combination of portability and performance, the HP Omen Transcend 14 is currently on sale. The 14-inch powerhouse is available at $1999.99 for a limited period, down from its listed price of $2999.99 at Best Buy, which is a massive saving of $1000.

• Check out the HP Omen Transcend 14 on Best Buy

The HP Omen Transcend 14 is one of the rare compact gaming laptops packing a powerful set of hardware. This particular configuration comes with Intel’s Core Ultra 9 285H processor, paired with 32GB of LPDDR5x memory and an Nvidia RTX 5070 Laptop GPU with 8GB of GDDR7 VRAM. For storage, there is a 1TB PCIe Gen 4 NVMe SSD, while the laptop ships with Windows 11 Home. The combination should make it capable of handling modern games as well as demanding workloads such as video editing and other content-creation tasks.

A highlight feature of the laptop is the 14-inch 3K OLED display, which has a 2880 x 1800 resolution and a 120 Hz refresh rate. The panel supports a 48-120Hz variable refresh rate, covers 100% of the DCI-P3 color gamut, and offers up to 500 nits of HDR brightness, making it suitable for both gaming and content creation.

Despite packing high-end hardware, the Transcend 14 remains relatively portable, measuring around 0.7 inches thick and weighing about 3.6 pounds. In terms of ports, it comes with Thunderbolt 4, USB-C with DisplayPort and power delivery, HDMI 2.1, and a USB-A port. The laptop also comes with a four-zone RGB backlit keyboard, while the 71 Wh battery is claimed to last up to 14 hours on a full charge.

Considering the premium hardware and compact footprint, the HP Omen Transcend 14 at $1,999.99 is a deal worth considering. The $1,000 discount from its $2,999.99 listed price makes it a considerably more affordable proposition. However, you’ll need to move fast, as the deal price is only available for a few hours.

A chrysophile has ordered a special edition mini PC with a pure 24-carat gold passive chassis reports Fanless Tech. It’s a premium custom version of the ~5-inch (120 x 120 x 120mm) Kubb Fanless PC, but costing around $1.7 million with its opulent gold chassis. As well as its choice of materials pumping up price, this mini PC’s weight has increased from its default 4.6 to 28.7 pounds (2.1kg to 13kg). But as a bonus, which probably still doesn’t make it worth it for most folks, gold has significantly better thermal conductivity than aluminum.

In its social media post, Fanless Tech calls the Kubb Fanless in gold a “cooling marvel.” Its news post on the topic elaborates on this claim, explaining that the thermal conductivity of gold is “about 30% to 50% faster than aluminum.” That might help its Core Ultra chip perform a bit better, all else being equal. However the sacrifice to portability, as this compact gold PC is over 6x heavier, could be a major drawback. As well as the price.

We looked up the thermal conductivity of aluminum and gold, as well as some other common materials. The chart below could be useful for your own passive computing projects.

In contrast to its gilded construction, this mini PC's specs are very modest. Standard configurations of the Kubb Fanless are as follows:

• Intel Core Ultra 5 225H, or Intel Core Ultra X7 358H, or AMD Ryzen AI 5 340
• 16GB DDR5, up to 128GB
• 1TB M.2 NVMe SSD, optional second M.2 SSD
• Intel: 1x USB 3.2 (Type A), 1x USB 3.2 (Type-C, DP1.4a), 1x USB4/Thunderbolt 4 (Type-C, DP 2.1), Wi-Fi 7, Bluetooth 5.4
• AMD: 1x USB 3.2 Gen2 (Type A), 2x USB4 (Type-C, DP 1.4a), Wi-Fi 6, Bluetooth 5.3
• Side ports: 2x Port RJ45, 2x HDMI 2.1 TMDS, 2x USB 3.2 Gen 2 (Type A)
• 19V DC, 6.32A, 120W power adapter (100–240V AC)
• Windows 11 Pro
• 3 years guarantee

Kubb Fanless mini PCs are usually made from aluminum, and finished in graphite, blue, or bronze. With an Intel Core Ultra 5 225H at their heart, all those colors are all listed at €1,595 ($1,822) at Kubb.eu, but there is no stock. At that kind of price, the Arctic Senza AI 370 we tested earlier in the year looks like it would be a knockout passive computing competitor. We aren’t sure if Arctic would accept the 24 carat gold chassis special order, and the Senza is designed for discrete under-desk-mounted silence.

We previously reported on the fanless Kubb Mini range of 3 x 3 x 3-inch mini PCs which packed Intel Twin Lake chips, came in cute/bright color finishes, and were priced around $500 (€430).

PlusAI, an autonomous trucking startup, is working with Nvidia to develop its systems and has several trucks and trailers marked with the logos of both companies. This likely caught the attention of thieves who were looking to score big on AI GPUs, with Wired reporting that two of the startup’s trailers got stolen and were found at another location. However, the joke’s on the criminals, as they discovered that the two stolen units contained 20,000 pounds of sand each.

The AI data center construction boom has made PC components like RAM and GPUs insanely expensive, which has led to a rise in petty theft involving these components. Enterprises and startups are being targeted, too, especially as they build data centers and local AI servers with millions of dollars’ worth of equipment and hardware. We’ve already seen cargo heists of copper cable and other hardware worth $1.3 million, while other criminals have become more brazen, pushing or spinning security escorts off the road so they can get away with a truck’s worth of AI GPUs.

While PlusAI is partnered with Nvidia, it also uses simulated cargo loads for real-world tests, allowing the company to model how a fully loaded semi-trailer would react to various road conditions. The startup’s truck cabs, which likely had advanced AI hardware installed, were parked securely inside its warehouse, while the trailers were apparently left outside and secured with hand locks. “PlusAI uses simulated loads in its trailers to assist with research and development testing,” company spokesperson Lauren Kwan told Wired, adding that they were recovered with “40,000 pounds of sand intact.”

Interestingly, the discovery of the theft was accidentally made by one of the startup’s top executives, who received a message from an acquaintance saying that they saw the trailers, which they described as “trucks,” near their workplace. This likely got the executive concerned, as that location was half a mile away from PlusAI’s warehouse, which led to the cops getting called. As for the prospective criminals, they probably left the crime scene disappointed — they thought they were getting a huge payday from two truckloads’ worth of smart silicon, but they ended up with 20 tons of dumb sand instead.

Gigabyte Technology hardly needs an introduction to anyone who has built a PC in the last three decades. Founded in Taipei in 1986, the company grew into one of the largest motherboard and graphics card manufacturers in the world, and its AORUS brand has become a fixture of the enthusiast market. Power supplies are a comparatively recent and more modest part of its business. Like nearly every PC brand, Gigabyte does not manufacture its own PSUs. Instead, it sources platforms from established OEMs and sells them under its UD, P-series and AORUS lines. The results have been uneven over the years, with some very good units and a few that left a lot to be desired, so the OEM and the target audience behind each model matters more than the badge on the box.

The Gaming GM PG5 series is Gigabyte's newest attempt to cover the upper mainstream, and the 1000W model in review today is something of a spiritual successor to the popular UD1000GM PG5. We look closer to determine how it ranks among our best power supplies list. It is built by HEC, a long-established Taiwanese OEM. It is fully compliant with the ATX 3.1 and PCIe 5.1 design guides and comes with a native 12V-2x6 connector, a 135 mm fluid dynamic bearing fan, and (according to Gigabyte) 100% Japanese capacitors.

The headline feature, though, is T-Guard, a thermal monitoring system for the 12V-2x6 connector that can cut power to the graphics card if the connector starts overheating. The T-Guard signifies Gigabyte’s proposed solution to the known PCIe 5.1 connector overheating problem that almost every manufacturer has been trying to implement a defense against. Curiously, Gigabyte markets the unit with 80 PLUS Gold logos, even though it has received Platinum certifications from both CLEAResult (80 PLUS) and Cybenetics. It is also available in white as the 1000GM PG5 ICE, comes with a ten-year warranty, and retails for $170 at the time of this review.

Specifications and Design

In the Box

The Gaming 1000GM PG5 arrives in a black and orange cardboard box with a large render of the unit on the front, next to the model name and a bold orange "1000W" print. A row of small icons along the bottom highlights the ATX 3.1 and PCIe 5.1 compatibility, the fully modular design, the Japanese capacitors, the smart fan, and T-Guard, while an 80 PLUS Gold badge sits at the top right corner. It is a flashy, gaming-oriented design, which is fitting given the name of the series.

The bundle is very basic. Other than the unit itself, we only found the AC power cord, four black cable ties, and a multilingual user manual. There are no Velcro straps and no cable combs. Strangely, there were not even mounting screws in our sample's box. For a $170 1000W unit, that is on the stingy side of things, although it hardly matters for the actual quality of the product, and most builders probably already have a drawer full of these things.

All of the cables are modular, including the 24-pin ATX cable. Gigabyte went with black, flat "embossed" cables, where every wire is individually insulated, but the wires are bound together into a ribbon. They are flexible and easy to route, and they look considerably better than plain flat cables. The single native 12V-2x6 cable has a dual-color connector, with a light grey front section that makes a partially seated plug easy to spot, and it is rated for 600W.

External Appearance

The chassis of the Gaming 1000GM PG5 measures 150 x 150 x 86 mm (L x W x H), which are slightly larger than the typical dimensions of a standard ATX unit. Still, at 150 mm long, it will fit in virtually every ATX case, although the length means that very compact builds may be a little tight once the modular cables are installed. The design is more elaborate than most PSUs in this price range. The upper part of the chassis is finished in a dark metallic gunmetal grey, while the lower part and the front and rear faces are matte black.

The sides carry a large embossed "GAMING" logo along with the Gigabyte name and the wattage, and the fan grille is a perforated octagonal area with the Gaming series logo at its center, with four exposed screws at the corners. The specifications sticker is placed on the side opposite the fan.

The rear face of the unit is almost entirely perforated with a honeycomb pattern, and there are only the on/off switch and the AC receptacle. There is no switch to control the semi-passive fan mode. It is always active. The modular panel on the front is clean and clearly labelled, with four sockets for the CPU and PCIe cables, three sockets for the peripheral and SATA cables, the 24-pin motherboard sockets and the non-standard 12V-2x6 cable socket, which is placed separately at the bottom left corner.

T-Guard is the major update over the UD1000GM PG5, and a feature Gigabyte clearly is relying on to market the unit. According to Gigabyte, precision thermistors monitor the temperature of the 12V-2x6 connector in real time, and if an abnormal temperature rise is detected, the system cuts power to the graphics card while keeping the rest of the system running. If the CPU has integrated graphics and the monitor is connected to the motherboard, the user can continue working and save any progress before shutting down, and a reboot is required to restore power to the GPU.

The UD1000GM PG5 had nothing similar. It relied solely on the typical protection features of every PSU, which do nothing against a single overheating connector, as current is spread across many pins and the total draw never approaches the OCP limit. Monitoring the temperature directly addresses the actual failure mode behind the melting connectors that have plagued high-end graphics cards over the past few years, and it is a lot more meaningful than yet another marketing badge. The concept is sound and, as a concept, it should work, unless the temperature increase on the connector is very abrupt and the position of the sensor does not allow it to react in time.

Internal Design

The cooling fan is a Hong Hua HA13525H12SF-Z, a 135 mm fluid dynamic bearing (FDB) fan rated at 0.50A. Hong Hua fans are very common inside mainstream PSUs, are reliable, and should easily last through the ten-year warranty. The larger 135 mm diameter allows the fan to move the same amount of air at lower speeds than a 120 mm fan, which pays off in the acoustics department and probably the main reason why the designer chose to push the chassis to 150 mm, as the 135 mm fan does not fit in a 140 mm chassis. A clear plastic sheet under the fan guides the airflow across the heatsinks.

The OEM behind the Gaming 1000GM PG5 is HEC, a manufacturer that has been producing power supplies for decades and has built units for many renowned brands. The platform is modern, with good-sized two-prong heatsinks for the active components, and a clean, well-spaced layout that helps with airflow. The transient filtering stage starts on a small PCB right behind the AC receptacle and continues on the main board, with four Y capacitors, three X capacitors, two filtering inductors, and a MOV, which is more than a complete filter for a unit of this class. An NTC thermistor and a bypass relay provide inrush current protection. There are two bridge rectifiers, mounted on their own relatively large dedicated heatsink.

The APFC stage uses three CRJW125N60G2F MOSFETs and a boost diode. The coil is fully enclosed and the bulk capacitor is a single 680 μF / 400 V unit from Toshin Kogyo (TK), a respected Japanese manufacturer, rated for 105 °C. The primary inversion stage consists of two NCE 65TF099F MOSFETs on a dedicated heatsink, forming a half-bridge LLC resonant topology. It is a simple, mature, and efficient arrangement that is found in the majority of high performance units today.

On the secondary side, the 12V rail is generated by six unidentifiable MOSFETs in a synchronous rectification arrangement. The 3.3V and 5V rails are generated by DC-to-DC converters on a vertical daughterboard. Gigabyte's claim of 100% Japanese capacitors is accurate. All of the electrolytic and polymer capacitors on the secondary side are supplied by TK as well, which is a welcome sight in a $170 unit, and it is excellent news for its long-term reliability.

Cold Test Results (25°C Ambient)

Cold Test Results (25°C Ambient)

For the testing of PSUs, we are using high-precision electronic loads with a maximum power draw of 2700 Watts, a Rigol DS5042M 40 MHz oscilloscope, an Extech 380803 power analyzer, two high-precision UNI-T UT-325 digital thermometers, an Extech HD600 SPL meter, a self-designed hotbox, and various other bits and parts.

At 25 °C ambient, the Gaming 1000GM PG5 performs much better than the Gold logo on its box suggests. It averaged 92.6% at 230 VAC and 90.6% at 115 VAC across the nominal load range, peaking at 93.8% (230 VAC) at about 50% load and 92% (115 VAC) at about 30% load. Even at a sustained 1000W, efficiency was 91.0% and 88.5%, respectively. This is Platinum-level performance, and it is consistent with both certifications the unit actually holds: 80 PLUS Platinum from CLEAResult and ETA Platinum from Cybenetics. Why Gigabyte chose to market it as a Gold unit is anyone's guess.

The thermal and acoustic performance balance is impressive. The fan remains completely stopped up to about 30% load (roughly 300W), which covers most desktop work. Once it starts spinning, it barely gets audible up to 500W, where it reaches about 33 dB(A). From there it ramps up gradually, reaching 36.4 dB(A) at 600W and about 41 dB(A) at 800W, before climbing to 48.2 dB(A) at a sustained 1000W. That last figure is loud, but it is hardly a load most systems will continuously sustain. Thermally, the unit maintains very low and safe operating temperatures.

Hot Test Results (~45°C Ambient)

Hot Test Results (~45°C Ambient)

Inside our hotbox, at roughly 45 °C ambient, efficiency drops by just 0.8%, averaging 91.8% at 230 VAC and 89.8% at 115 VAC, with full load figures of 89.9% and 87.4% respectively. That is a very small degradation for a 20 °C increase in ambient temperature, which is a testament to the quality of the components and their resilience to adverse operating conditions.

The fan is a slightly different story. It still stays off up to about 300W, but it starts at a higher speed and ramps up much faster, reaching 38.3 dB(A) at 500W and 47.2 dB(A) at 800W. At a sustained 1000W, the fan climbs to 54.5 dB(A), practically the maximum, as the fan reaches its top speed, which is very loud and very noticeable in any system. In return, the internal temperatures stay very low, with the primary and secondary heatsinks reaching just 88°C and 82°C respectively at full load. These are high figures, but they are reasonable for a unit operating at its maximum output in a 45°C environment, and the unit’s components are very capable of handling them. Gigabyte clearly favored longevity over everything else.

PSU Quality and Bottom Line

Power Supply Quality

Voltage regulation is good on the minor rails and average on the 12V rail. Across the 20% to 100% load span, the 3.3V and 5V rails moved by just 0.5%, which is exceptional. The 12V rail regulation is 1.3% - not a class leading figure, but not a bad result either. The ripple suppression is very good. The 12V rail peaked at 48 mV at full load, while the 3.3V and 5V rails did not exceed 26 mV under any condition. Against the 120 mV and 50 mV ATX design limits, that is 60% headroom on the 12V rail and 48% on the minor rails. Several premium platforms can do better, but these figures are far from unacceptable on a mainstream PSU.

During our thorough assessment, we evaluate the essential protection features of every power supply unit we review, including Over Current Protection (OCP), Over Voltage Protection (OVP), Over Power Protection (OPP), and Short Circuit Protection (SCP).

All protection mechanisms were verified and triggered correctly during testing. With the unit hot, OCP tripped at 136% (27.2A) on both the 3.3V and 5V rails, and at 120% (100A) on the 12V rail, while OPP shut the unit down at 116% of its rated output, or about 1160W. These are tight thresholds for a 1000W unit, yet not too tight, as expected from an ATX 3.1 compliant unit.

Bottom Line

The Gigabyte Gaming 1000GM PG5 is a well-rounded power supply, and one that Gigabyte seems to undersell. The HEC platform is modern and well designed, the component selection includes Japanese capacitors all around, which is a welcome sight at this price point, and the unit is fully compliant with the latest ATX 3.1 and PCIe 5.1 design guides. It is evident that Gigabyte favored longevity and reliability over squeezing out every last bit of electrical performance, and for a mainstream 1000W unit, that is a very reasonable priority.

T-Guard deserves a special mention, as it is the major update over the UD1000GM PG5 and the reason most people will notice this unit. It is Gigabyte's answer to the known 12V-2x6 connector overheating problem, which almost every manufacturer is currently trying to find a defense against. By monitoring the connector temperature and cutting power to the GPU if something goes wrong, while keeping the rest of the system alive, it targets the actual failure mode instead of relying on protections that were never designed for it. As a concept, it should work, unless the temperature rise is very abrupt and the position of the sensor does not allow it to react in time. Even so, for anyone planning to run a power hungry graphics card, it is a reassuring addition.

Efficiency is where the unit shines. It is Platinum-level efficient, as confirmed by our testing and both the 80 PLUS and Cybenetics certifications, which makes Gigabyte's decision to market it with Gold logos rather strange. The rest of the electrical performance is average for its class, with good minor rail regulation, a decent 1.3% on the 12V rail, and ripple that is comfortably within the ATX limits, if not remarkable. The balance between temperatures and noise is very good. At room temperature, the fan is barely audible up to 500W, and the internal temperatures remain low and safe, even inside our hotbox, where efficiency barely degraded.

The compromises are few. The fan gets very loud under sustained heavy loads in hot environments, practically reaching its top speed at full load, and the bundle is bare, lacking even mounting screws. The 150 mm long chassis, a consequence of the 135 mm fan, may also be a little tight in very compact cases. None of these are serious issues, as the vast majority of gaming systems will rarely push the unit hard enough for the fan to become a nuisance.

With a price tag of $170 at the time of this review and a ten-year warranty, the Gaming 1000GM PG5 represents good value, but it does not stand out across the board. Against other 1000W units in the same price range, such as the Corsair RM1000x, the MSI MAG A1000GL PCIE5 and the Montech Century II 1050W, it provides average all-around electrical performance, so buyers should not expect a noticeable difference in regulation or ripple. Where it pulls ahead is efficiency, as its Platinum-level figures are great for the class, and it also offers a good balance between temperatures and noise, with all-Japanese capacitors and T-Guard as extra reasons to pick it. For a gaming system built around a demanding graphics card, especially one with a 12V-2x6 connector, that combination makes the Gaming 1000GM PG5 a very sensible 1000W choice, even if it is not a clear winner in every category.

MORE: Best Power Supplies

MORE: How We Test Power Supplies

MORE: All Power Supply Content

Developers of SharpEmu — an experimental, open-source PS5 emulator — have successfully run PS5 games at a playable 60 FPS on the emulator. According to an X post on September 25, the developers reached this milestone for six of 55 titles tested, with 12 — including the six playable — reaching gameplay. According to the developer, the six games are yet to be tested from start to finish; the “playable” designation means the games currently run without issues, although minor graphical or audio issues may persist in some.

The six 60 FPS titles are Dreaming Sarah, Void Terrarium, New Joe & Mac: Caveman Ninja, Tetris Forever, Hoa, and Tomb Raider IV-VI Remastered. The breakthrough was achieved with SharpEmu v0.0.4-release.2, the emulator’s latest public build released on September 25. According to developers, the build includes upgrades such as improved SPIR-V shader translation and updated Vulkan API support.

Developed in C#, the SharpEmu PS5 emulator supports Windows, Linux, and macOS and uses Vulkan and MoltenVK for graphics API calls. PlayStation 5 relies on an AMD Zen 2 x86-64 CPU architecture; therefore, SharpEmu can execute the console's CPU instructions natively on a standard PC rather than translating them from an unfamiliar instruction set. The emulator focuses on loading a game's eboot.bin executable and required system modules, partially handling kernel functions, and processing the game's shader, resource, and AGC graphics submissions.

For now, SharpEmu is still far from handling demanding 3D titles, with only 12 of 55 tested titles reaching a game ready state. Of those 12, only six performed cleanly enough to be designated as “playable” at a full 60 FPS. The developers stress that SharpEmu is still in its early stages, noting that “SharpEmu is an experimental emulator intended for research and educational purposes. This project does not contain any copyrighted system firmware, game data, or proprietary PlayStation assets.” Meanwhile, another emulator, KytyPS5, recently ran the PS5 version of GTA V at roughly 40 to 60 FPS on a high-end Ryzen 9 9950X3D and Radeon RX 7900 XT system, crashing after the game's opening mission.

It won’t be long until Linux enthusiasts will be able to complete a clean kernel build in under 10 seconds. Linux expert and Phoronix head honcho Michael Larabel said that “the coffee window is closing” in recent commentary that weighs computer processor advances and human / AI optimizations of the Kbuild code. What once used to be a computer processing task that provided a decent excuse for a coffee break now only allows enough time for a measured sip.

The source site’s determined Linux focus has meant that OS kernel compilation times have become a signature benchmark over the last two decades. Larabel notes that this once time-consuming process “can be done in now roughly 15 seconds.” It isn’t just advances in hardware and core counts pushing the envelope; compiling a defconfig x86-64 Linux kernel build has had many bottlenecks removed thanks to the work of Linux MM developer Lorenzo Stoakes and the assistance of AI/LLMs recently. Even modest processors have seen their compile times cut dramatically thanks to this work.

Larabel went hands-on with the latest v4 kernel patches from Lorenzo, and armed with his very powerful EPYC workstation, achieved “a 15-second kernel build!” Before the latest patches, the same system took over 22 seconds for the same compilation task.

The test system is extremely potent, though. Your home PC or laptop might still let you have enough time to go brew some coffee, as Larabel’s test rig features the following components:

• 2x AMD EPYC 9575F 64-core processors (so, 128 cores / 256 threads)
• 24 x 64GB DDR5-6400 memory modules
• Samsung PM1743 3.84TB PCIe Gen 5.0 NVMe SSD storage
• Stock Ubuntu 26.04 LTS

It is noted that the code compilation benchmarking didn’t employ a RAMdisk to aggressively tune the compile times

All the above timings mentioned are for the minimal default kernel compilation for x86-64 systems. The Phoronix chief also timed the compilation of a Linux kernel with ‘allmodconfig’ with thousands of drivers and subsystems enabled. This precipitates far longer compile times. On the above beefy system, though, the latest Lorenzo patches drove the time down from 169 seconds to 134 seconds. So, this other milestone is approaching the two-minute mark.

Phoronix is now taking bets on AMD (Zen 6) EPYC 9686F systems, with MRDIMMs and PCIe Gen6 storage pushing defconfig x86-64 Linux kernel compilation times under the 10-second mark.