VideoCardz

Full SteamOS now runs on Apple Macs, Windows games tested at over 200 FPS

An independent project called FX Steam Launcher has made it possible to run Valve’s official SteamOS on Apple Silicon Macs. The software uses the ARM64 version of SteamOS developed for Valve’s Steam Frame, bringing the complete Steam gaming interface and Windows game compatibility to macOS. The project is open-source and available on GitHub.

Unlike a native installation, FX Steam Launcher runs SteamOS inside a virtual machine using Apple’s Hypervisor.framework and libkrun. So the idea is similar to DroidDeck, which is basically a full SteamOS running on Android.

The Apple version supports SteamOS Gaming Mode, KDE Plasma Desktop Mode, controllers, audio and GPU acceleration. Users can run SteamOS in a window or fullscreen, while retaining access to macOS. The launcher also downloads the official SteamOS image directly from Valve and supports operating system updates. 

SteamOS on Mac supports DirectX 12 through multiple translation layers

Windows game compatibility relies on Proton 11 for ARM64 and FEX for translating x86 instructions to ARM64. DirectX 9, 10 and 11 games use DXVK, while DirectX 12 games rely on vkd3d-proton. Graphics instructions then pass through Mesa Venus and either MoltenVK or KosmicKrisp before reaching Apple’s Metal API. The project also offers experimental DXMT support, which allows DirectX 10 and 11 games to use Metal without the Vulkan translation layer.

The launcher supports Apple Silicon Macs running macOS 15 or newer. However, KosmicKrisp requires macOS 26 or later and offers DirectX 12 feature level 12_0 support. Games requiring feature level 12_1 or Shader Model 6.6 remain unsupported. The developers also provide configurable CPU and memory allocations, MetalHUD monitoring, mouse and keyboard support, and compatibility with Xbox, DualSense and other controllers.

Apple M4 Max tested in Windows games, reaches over 200 FPS in Left 4 Dead 2

YouTuber ETA Prime tested the project on a MacBook Pro equipped with an Apple M4 Max featuring a 14-core CPU, 32-core GPU and 36GB of unified memory. The test included several Windows games running locally through SteamOS, with results ranging from more than 70 FPS in The Witcher 3 to over 200 FPS in Left 4 Dead 2. Stellar Blade also ran at over 70 FPS using High settings at 800p.

The results are not without issues. The Witcher 3 displayed incorrectly rendered trees, while several games experienced stuttering during shader compilation. DOOM Eternal launched but refused to run because it detected an unsupported GPU. Marvel’s Spider-Man: Miles Morales and Ghost of Tsushima also failed to start.

ETA Prime initially tested the software on an M6 Mac mini with 16GB RAM, which managed over 140 FPS in Left 4 Dead 2 but struggled with more demanding games due to memory swapping. On the M4 Max MacBook Pro, approximately 19GB was allocated to the VM. ETA Prime recommends at least 24GB RAM for more demanding games, although less demanding titles can work on 16GB systems. Increasing the VM allocation beyond the eight performance cores of the M4 Max also reduced performance in some tests.

FX Steam Launcher is still under active development. Version 1.9.8, released November 2, restores compatibility with SteamOS 0.6.0 preview after a Valve update prevented the virtual GPU from initializing. 

Apple Gaming

So, is this finally the moment when gamers can start taking Apple M-series devices seriously? Perhaps. The bigger question is why it took independent developers to make this happen, rather than Apple and Valve working together. I guess 2026 is the year where software crosses all proprietary hardware limitations. 

Source: FX Steam Launcher GitHub, ETA Prime

NVIDIA’s five-year-old RTX 3060 returns to focus

Graphics card manufacturers in China are reportedly advising distributors to focus on older GeForce RTX 30 series cards and AMD Radeon models as RTX 50 series supply becomes harder to secure. According to a report published by ChannelGate, some manufacturers are already running out of RTX 50 inventory, while others have only limited quantities left. Prices are also reportedly increasing.

RTX 3060 and RTX 3050 recommended as alternatives

ChannelGate points to increased cryptocurrency mining demand and China’s ongoing e-commerce promotional season as two reasons for the shortage. Yes, you heard that right, cryptocurrency is again on the rise, at least for board partners to notice.

With RTX 50 cards becoming harder to obtain, manufacturers are reportedly recommending older models instead. The list includes the GeForce RTX 3060 12GB, RTX 3060 8GB, RTX 3050 and unspecified AMD Radeon cards. These are being suggested as alternatives to help distributors keep up with demand.

NVIDIA already brought back the RTX 3060 this year

Interestingly, NVIDIA quietly brought back the GeForce RTX 3060 earlier this year, nearly five years after its original launch. As we reported in June, Colorful began shipping new RTX 3060 12GB cards in China, with more models later appearing in Europe and the US. PALIT also introduced its RTX 3060 Infinity 2 OC in July, showing that partners were still releasing new Ampere board designs rather than simply clearing out old stock.

Unfortunately, bringing back a five-year-old GPU has not made it cheaper. In August, we reported that the MSI GeForce RTX 3060 Ventus 2X 12GB had gone from $329.99 to $479.99 at Newegg in less than two months. That is a 45% increase, making the older Ampere card more expensive than newer options such as the RTX 5050.

ChannelGate expects older GPUs to maintain their current prices or become slightly more expensive as distributors look for alternatives. The report also focuses on China, so we do not know whether distributors in other regions are receiving similar advice. Still, the fact that manufacturers are reportedly recommending RTX 3060 and RTX 3050 cards in late 2026 was not on our Bingo card.

Source: Board Channels, Wccftech

New ZBOX CI431, CI451 and CI471 nano systems feature Intel Core Series 3 processors with up to six cores

ZOTAC has introduced three fanless ZBOX mini PCs based on Intel’s Wildcat Lake platform. The new ZBOX CI431 nano, CI451 nano and CI471 nano use Core Series 3 processors with passive cooling, DDR5 memory support and Thunderbolt 4 connectivity.

All three systems share the same chassis measuring 204 × 129 × 68 mm. ZOTAC has designed them without cooling fans, making them suitable for environments where low noise is required. The systems are offered as barebones configurations, allowing users to install their own memory and storage.

Three Intel Wildcat Lake configurations

The entry-level ZBOX CI431 nano uses the Intel Core 3 304, a five-core processor with one Performance core and four Low Power Efficient cores. The CI451 nano upgrades to the six-core Core 5 320, while the CI471 nano uses the Core 7 350 with a maximum boost clock of 4.8 GHz.

All three processors belong to Intel’s Wildcat Lake family. They have a 15W processor base power and a maximum turbo power of 35W. The integrated graphics use Intel’s Xe3 architecture, with one Xe-core on the Core 3 304 and two Xe-cores on the Core 5 and Core 7 models.

The new ZBOX systems support up to 64GB of DDR5-6400/5600 SO-DIMM memory and an M.2 NVMe SSD using a PCIe 4.0 x4 interface. Connectivity includes two Thunderbolt 4 ports, two USB 3.2 Gen 2 ports, three USB 2.0 ports and 2.5Gb Ethernet. Wireless connectivity includes Wi-Fi 7 and Bluetooth 5.4.

ZOTAC also supports three simultaneous displays through HDMI 2.0b, DisplayPort 2.0 and DisplayPort over Thunderbolt 4. The systems include mounting support for VESA 75 × 75 mm and 100 × 100 mm brackets. They are listed as Windows 11 ready, although the barebones models require additional components before use.

 ZOTAC has published product listings for all three configurations but has not provided pricing or a retail availability date.

Source: ZOTAC, FanlessTech

Nintendo Switch 2 games can now run natively on Apple Silicon

Nxnative is not a traditional console emulator. Instead, it runs the games’ ARM64 code directly on Apple Silicon while translating Nintendo’s graphics API to Apple’s Metal. The project originally focused on converting Switch games into applications for iPhone, iPad and Mac. Switch 2 compatibility introduces support for the newer NVIDIA Ampere GPU architecture and NVN2 graphics API.

Donkey Kong Bananza runs at 60 FPS on M4 Pro

According to the project’s documentation, Donkey Kong Bananza has been tested on a Mac powered by Apple’s M4 Pro chip. The game reportedly runs at 60 FPS through the opening sections, including the mining tunnel and subsequent areas. During demanding scenes, the game can temporarily drop to 30 FPS. Testing has covered the opening areas rather than the entire game.

The developers also confirmed that Splatoon Raiders runs on the M4 Pro. Testing covered the opening sequence, character creation and the first level’s beach area, including combat against Salmonids. The latest session lasted approximately 4.5 minutes without crashing. This game uses Switch 2 SDK 21.4.4, and its compatibility profile includes more than 27,000 shaders.

Unlike original Switch games, which can be packaged from NSP files using encryption keys, Switch 2 games currently require a folder containing extracted game files. The project can then build a macOS application from those files. The documented requirements include macOS 26.6 or later and Xcode 27.2. Official Switch 2 testing remains focused on macOS.

It looks like the gates have opened. Not only can you now play Switch 2 games such as Mario Kart World on PC, but emulators for Apple’s macOS now support them as well.

Source: nxnative

Lenovo’s 13-inch OLED Googlebook tablet could launch soon

Google has confirmed that detachable Googlebook devices are coming soon. During Android Dev Summit 2026, the company announced plans to expand its laptop platform beyond traditional clamshell and convertible designs, reports 9to5Google. The next category will include tablets with removable keyboards, allowing them to operate as standalone touchscreen devices.

Googlebook is Google’s new laptop platform built on Android technology with desktop components from ChromeOS. The first five models launched with ASUS, Dell, HP, Acer and Lenovo. Four use traditional laptop designs, while the Acer Googlebook 14 is currently the only convertible, starting at $899 in the US. However, none of these devices has a fully detachable keyboard.

Lenovo Googlebook Duet m could be the first detachable model

The first detachable Googlebook is expected to come from Lenovo. Earlier leaks revealed a device reportedly called the Lenovo Googlebook Duet m, featuring a 13-inch OLED touchscreen, a removable keyboard and a rear kickstand cover. Leaked images shared by Evan Blass also showed a fingerprint reader, a rear camera, a Glowbar indicator and support for the Lenovo Pen Pro stylus.

According to earlier hardware reports, the tablet is expected to feature the MediaTek Dimensity CX C10 Max, previously associated with the Kompanio Ultra 910. This 3nm processor features eight CPU cores, including one Cortex-X925, three Cortex-X4 and four Cortex-A720 cores. It also includes an Immortalis-G925 MC11 GPU and an NPU rated at up to 55 TOPS. The reported tablet configuration includes 16GB LPDDR5X memory, UFS 4.0 storage, Wi-Fi 7 and two USB-C ports. Lenovo has not officially confirmed these specifications.

Google has not announced a release date or pricing for its first detachable Googlebook.

Source: 9to5Google

RDNA 5+ could follow RDNA 5 in AMD’s mobile GPU lineup

AMD is reportedly planning another graphics architecture update called RDNA 5+, which could arrive in 2029 alongside its Zen 7-based Grimlock processors. The information comes from hardware leaker Kepler_L2, who suggests that AMD will introduce an updated version of RDNA 5 for future integrated graphics.

In a NeoGAF discussion about RDNA 5 availability in mobile processors, Kepler_L2 responded to speculation that the architecture would reach regular laptops in 2028 or later. His answer was brief: “RDNA5+ in 2029 with Grimlock.” Worth adding that this is not the first time Kepler has mentioned RDNA 5+, as the architecture was already referenced in September.

AMD has already used a similar approach with RDNA 3.5, which introduced architectural updates to RDNA 3 for integrated graphics. RDNA 5+ could follow the same strategy by improving the existing RDNA 5 design rather than introducing an entirely new GPU architecture. However, Kepler has not shared any technical differences between RDNA 5 and RDNA 5+.

The rumored schedule also fits AMD’s reported plans to separate its integrated GPU offerings. Lower-end Ryzen processors are expected to continue using RDNA 3.5 graphics, while premium designs such as Medusa Halo would move to RDNA 5. The 2029 Grimlock generation could introduce RDNA 5+ to a wider range of mobile processors.

Could RDNA 5+ appear in future gaming handhelds?

There is another interesting connection to recent rumors about Sony’s next PlayStation handheld. Digital Foundry recently mentioned discussions suggesting Sony may have an exclusive arrangement of sorts for RDNA 5 handheld chips. This would potentially leave the rumored PlayStation 6 handheld as the only portable gaming device with a custom RDNA 5 processor during its launch period.

If this information is accurate, RDNA 5+ could eventually become an option for competing handheld manufacturers. Sony could receive its custom RDNA 5 chip first, while other companies might adopt RDNA 5+ through later Ryzen mobile processors or custom designs. 

Source: Kepler_l2 (NeoGraf)

Lenovo launches Xiaoxin Pro 27 with Intel Core 5 320 for $1,126

Lenovo has introduced the Xiaoxin Pro 27 Vitality Edition in China, an updated version of its 27-inch all-in-one desktop powered by Intel’s Wildcat Lake processors. The new model uses the same chassis as the existing Xiaoxin 27 Vitality Edition but replaces its FHD screen with a higher-resolution panel.

Intel Core 5 320 with a 27-inch QHD display

The updated display features a 2560 × 1440 resolution, a 120Hz refresh rate, 400 nits of brightness and 100% sRGB coverage. Lenovo also includes a height-adjustable stand with tilt support. The display housing measures just 13mm at its thinnest point, while the base is 38% thinner than the previous generation.

The system is powered by an Intel Core 5 320 processor featuring two Performance cores and four Low Power Efficient cores, for six cores and six threads. The processor boosts up to 4.6 GHz and includes integrated Intel Graphics. The current Chinese configuration comes with 16GB of LPDDR5X-7467 memory and either 512GB or 1TB of SSD storage.

Limited upgrade options, but the display can be reused

As with most all-in-one PCs, the integrated screen is not designed to be upgraded separately. Buyers cannot simply replace the panel with a newer 4K or higher-refresh-rate monitor while keeping the rest of the system. The memory is also soldered and cannot be upgraded, although Lenovo provides a second M.2 slot for additional SSD storage.

The HDMI input allows the built-in display to work as an external monitor for another computer, supporting up to 1080p at 120Hz. This means the screen can still be used after the integrated PC hardware becomes outdated, although the HDMI input does not support the panel’s full QHD resolution.

The Xiaoxin Pro 27 Vitality Edition is listed in China for 7,499 yuan (approximately $1,126) with 512GB of storage and 7,999 yuan (approximately $1,201) for the 1TB version. Lenovo also lists comparable international configurations under the IdeaCentre AIO 27IWC11 name, including models with Core 5 320 and Core 7 350 processors. However, retail availability and pricing outside China have not been confirmed.

Source: Lenovo

ASRock preparing Z970 LiveMixer and Pro-A motherboards for LGA-1954

Two ASRock motherboards based on Intel’s upcoming Z970 chipset have appeared in shipping records. According to NBD customs data spotted by Club386, the company has shipped PCBs carrying the names Z970 LiveMixer WiFi and Z970 Pro-A WiFi. The records refer to shipments from Vietnam dated late September 2026.

Both names are already used by ASRock for its existing B860 motherboard lineup. The LiveMixer series targets content creators with additional connectivity options, while Pro-A covers more affordable desktop systems. These listings suggest that ASRock plans to carry both families over to Intel’s next-generation LGA-1954 platform.

Z970 expected to support CPU overclocking

We first reported Intel’s plans for Z990 and Z970 chipsets in February, based on information from the company’s roadmap. Subsequent leaks indicated that Z970 would share its basic I/O specifications with B960 but enable CPU multiplier overclocking. The more expensive Z990 is expected to support both multiplier and base-clock overclocking.

According to the previously leaked specifications, Z970 will offer 34 total PCIe lanes, including 14 PCIe 4.0 lanes from the chipset, four SATA ports and a PCIe 5.0 x2 DMI connection. Unlike Z990, the Z970 chipset itself will not provide PCIe 5.0 lanes, although PCIe 5.0 connectivity from the CPU will remain available.

As we reported in April, hardware leaker Jaykihn claimed that Z970 would replace most of the motherboard market currently served by B860. Meanwhile, B960 would cover lower-cost systems and OEM configurations. This would make CPU multiplier overclocking available on a wider range of Intel motherboards, provided users also have compatible unlocked processors.

These are not the first ASRock Nova Lake motherboards to appear in shipping records. In September, we covered six Z990 models listed in the same customs database, with shipments dating back to July. The new Z970 entries expand that list to eight known ASRock models across the two chipsets. Neither Intel nor ASRock has officially announced the Z970 motherboards, and the shipments do not confirm retail production or launch timing.

Source: Club386

Steam Hardware Survey reverses September’s RTX 5070 gains

Valve has published its October 2026 Steam Hardware Survey, and the GeForce RTX 3060 is back at the top of the GPU rankings. The RTX 5070 held first place for just one month before losing most of its September gains. Meanwhile, Linux has reached 3.95% of surveyed systems, approaching the 4% threshold again.

RTX 3060 returns to first place

The GeForce RTX 3060 now accounts for 3.62% of surveyed systems, up from 3.54% in September. The RTX 5070 dropped from 5.86% to 3.56%, losing 2.30 percentage points in a single month. This puts NVIDIA’s older Ampere card narrowly ahead of the Blackwell model. These figures use Steam’s default combined-platform survey, while the separate PC GPU table lists both cards at 3.78%.

The RTX 5070 is not the only Blackwell GPU to lose share. The RTX 5060 dropped from 4.20% to 2.97%, while the RTX 5060 Ti declined from 3.87% to 2.55%. Both cards have returned close to their August results. The changes suggest September’s unusually large increases were influenced by survey sampling rather than actual hardware adoption alone.

Linux approaches 4% of Steam users

The operating system statistics show Linux reaching 3.95% of all surveyed systems, compared with 93.70% for Windows and 2.35% for macOS. Steam’s current page reports a 0.90 percentage point increase over September’s originally published 3.05%. However, Valve subsequently revised September’s Linux share to 3.90%, making the increase just 0.05 points using the corrected figures.

Among Linux distributions, CachyOS accounts for 0.65% of all Steam systems, followed by SteamOS Holo at 0.55% and Arch Linux at 0.35%. A separate SteamOS entry accounts for another 0.32%, while Bazzite reaches 0.27%. Linux has previously exceeded 4% in the Steam survey, including earlier this year, so October’s result is not a new record.

Other hardware statistics also reversed September’s changes. Systems with 16GB of RAM returned to first place with 41.42%, while 32GB dropped to 37.06%.

Simplified Chinese language usage also fell by 9.41 percentage points to 21.98%. Increases in Chinese-language engagement often correlate with holidays in China, when more users are active on Steam. Such shifts can heavily skew the survey’s hardware statistics, as we have seen in previous months.

Source: Steam Hardware Survey

NVIDIA RTX 5080 24GB January launch in doubt

NVIDIA has reportedly missed its own schedule for providing GeForce RTX 5080 24GB design kits to board partners. According to BenchLife, the company was expected to distribute the necessary documentation in October, but manufacturers had still not received it by the end of the month. This raises questions about the rumored January 2027 launch.

We previously reported that NVIDIA was preparing to replace the RTX 5080 16GB with a 24GB model in the first quarter of 2027. More recent information from Chinese distribution channels suggested that NVIDIA had accelerated those plans, potentially targeting December 2026 or January 2027. However, BenchLife’s latest report suggests the schedule may be unrealistic.

Without the documentation, manufacturers cannot complete their designs, testing and validation ahead of mass production.

CES 2027 announcement still possible, but retail availability may come later

One board partner once told us that launching a new graphics card takes at least six weeks after receiving the necessary design information. That was during a period of strong competition in the gaming GPU market, when manufacturers were still focused on gaming products. Today, AI demand competes for production capacity, while memory and other component shortages complicate procurement. Under these conditions, even two months from receiving design kits to retail availability may be too optimistic.

BenchLife believes a January release is unlikely, although there are still nearly three months until the end of January 2027. We have also heard repeated suggestions that NVIDIA could announce its plans at CES 2027, scheduled for January 6 to 9. An announcement would not necessarily require retail-ready cards, so NVIDIA could still reveal the RTX 5080 24GB at CES and launch it several weeks later.

The bigger concern is what happens to the existing RTX 5080 and RTX 5090 during this transition. Both cards are already difficult to find at reasonable prices, and reports suggest that supply to board partners has been restricted. If NVIDIA is preparing to replace the RTX 5080 16GB while reducing RTX 5090 shipments, the next two to three months could bring even worse availability and higher prices. 

It definitely doesn’t look great for the NVIDIA high-end RTX 50 series for the remainder of 2026. 

Source: BenchLife

Valve’s canceled Portal prequel F-Stop is now playable after 19 years, thanks to leaked source code

Valve’s canceled Portal prequel, F-Stop, has resurfaced nearly 19 years after development began. Members of the Valve Cut Content community have assembled a playable version of the game using leaked assets, source code, and map files. The reconstructed build reportedly includes almost the entire original campaign, although it remains unfinished.

F-Stop originated in 2007 as part of Valve’s Directed Design Experiments. The company divided its developers into five teams for three months to explore different game concepts. One of those projects became F-Stop, which was planned as a prequel to Portal before Valve canceled development and shifted its attention to Portal 2.

F-Stop uses a camera instead of Portal’s iconic portal gun

Unlike Portal, F-Stop relies on an experimental Aperture Science camera. Players can photograph objects, recreate them elsewhere, and manipulate their size to solve puzzles. The concept shares similarities with Superliminal and Viewfinder, although Valve experimented with these mechanics years before either game was released. Some of F-Stop’s characters, dialogue, and ideas were later reused in Portal 2.

The playable reconstruction became possible following several separate leaks. In August 2026, approximately 12 TB of historical Steam depot data appeared online, including many of F-Stop’s original assets. However, the archive lacked the game’s source code and map source files, preventing the community from assembling a working version.

Those missing files reportedly appeared through the Valve Cut Content Discord server in late October. According to Kotaku, the person responsible for the map leak initially requested $1,000 before eventually sharing the files for free. The community subsequently verified the map data and combined the available material into a playable build.

Gameplay videos now show the reconstructed campaign running from beginning to end in less than an hour. Some textures, assets, and dialogue are still missing, but community members estimate that approximately 99% of the original prototype has been recovered. This remains an unofficial reconstruction rather than a release authorized by Valve. There is no indication that the company plans to distribute F-Stop itself.

Source: Kotaku

RPCS3 PS3 emulator adds KosmicKrisp Vulkan driver for macOS, fixing graphics issues

The RPCS3 team has released a graphics update for the macOS version of its PlayStation 3 emulator. The latest builds now include KosmicKrisp, a Vulkan driver built on Apple’s Metal 4 graphics API. According to the developers, the new driver fixes several graphical issues associated with the previously used MoltenVK implementation.

KosmicKrisp brings Vulkan 1.4 support to Apple Silicon

RPCS3 uses Vulkan for graphics rendering, but Apple does not provide native Vulkan support in macOS. The emulator previously relied on MoltenVK to translate Vulkan commands into Metal. KosmicKrisp provides an alternative implementation based on the open-source Mesa graphics stack.

Developed by LunarG, KosmicKrisp supports Vulkan 1.4 and has passed the Khronos conformance tests. The driver uses Metal 4 and targets Apple Silicon hardware. RPCS3 developers say the integration addresses graphical problems experienced with MoltenVK, although they have not provided game-specific performance comparisons.

The new driver requires macOS 26 Tahoe or newer. Users running macOS 15 Sequoia can continue using MoltenVK, which remains available in the emulator. KosmicKrisp does not support Intel-based Macs, so those systems must also continue using the older implementation.

The update addresses graphics compatibility rather than PlayStation 3 CPU emulation. There are still macOS-specific limitations that affect certain games, and the developers have not claimed a general performance increase from the new Vulkan driver. However, Apple Silicon users now have another graphics backend available without switching to another operating system.

Source: RPCS3 on X, LunarG KosmicKrisp, NotebookCheck

Windows 11 can now download and reinstall itself when boot fails

Microsoft has introduced Cloud rebuild, a new Windows 11 recovery feature that allows users to reinstall the operating system without USB installation media. The feature downloads a Windows image and device drivers directly from Windows Update. It can also work when the installed operating system fails to boot.

Cloud rebuild is included in the optional KB5127936 update released on October 28, 2026, for Windows 11 25H2 and 26H2. Microsoft is distributing the feature gradually, meaning it may not appear immediately on all supported systems.

Rebuild without installation media

Windows 11 already offers a Cloud download option through Reset this PC. However, Cloud rebuild goes further by downloading both the operating system and required drivers from Microsoft’s servers. It also operates independently of the existing Windows installation, provided the Windows Recovery Environment (WinRE) remains functional.

Users can start the process through Settings > System > Recovery > Rebuild PC. If Windows fails to boot repeatedly, the recovery environment can launch automatically. Users can then access Cloud rebuild through the troubleshooting menu and connect to the internet using Ethernet or a supported Wi-Fi network.

Cloud rebuild performs a full reinstall and removes locally stored files, installed applications, accounts and settings from the system disk. Microsoft also offers a drive sanitization option that uses the storage device’s hardware erase capability. This is intended for PCs being sold, recycled or reassigned. Users should back up their data before proceeding whenever possible.

The feature also supports remote management for enterprise devices. IT administrators can initiate Cloud rebuild through Microsoft’s Recovery CSP and select either the currently installed Windows build or a specified target build. Cloud rebuild requires a functional recovery environment, network access and compatible device drivers available through Windows Update. The feature is rolling out with the October preview update and will reach more devices through subsequent updates.

Source: Microsoft

Claude Code now refuses requests to recreate existing games

Anthropic has introduced new restrictions for Claude Code that prevent its AI coding assistant from recreating existing video games. The change applies to Claude Opus 5.5 and was confirmed by Anthropic’s Addy Osmani on October 30. Although the restriction targets unauthorized copies of existing games, some developers report that it is also preventing them from continuing their own projects.

The change was discovered by X user Chasm, who found a new instruction reportedly delivered through Anthropic’s servers around October 27. The configuration, named tengu_heron_brook, specifically tells Claude Code not to recreate an existing game, even if the user changes its name or describes it as an original project. According to Chasm, users cannot disable this restriction through Claude Code’s settings.

Anthropic confirms that the instruction was added to discourage intellectual property infringement. Osmani explained that original games, projects based on existing genres and games using the developer’s own designs and assets should continue to work. He also acknowledged that legitimate projects might be rejected and encouraged affected developers to report such cases.

Developers report Claude Code refusing existing projects

One such complaint appeared on Anthropic’s official GitHub repository on October 30. A developer representing Conflict Simulations LLC reported that Claude Code had started refusing work on personalized digital board-wargame modules commissioned by customers. According to the developer, the projects use original code and customer-provided materials, with finished modules intended for the customers’ personal use.

The developer clarified that these projects are not officially licensed by the original game publishers. Interestingly, the reported refusals started after the developer explicitly asked Claude Code to evaluate whether the new restrictions applied to the existing work. The developer claims that even after clearing the original session, Claude continued refusing requests involving the same repositories.

Legality of reverse engeering

I think the most common question is,”Is this even legal?”. Reverse engineering is legal in both the US and EU under certain conditions, including for software compatibility and interoperability. Providing users with reverse engineering or recompilation tools is also generally legal.

It is up to users to ensure they have legally obtained copies of the software and comply with applicable copyright laws. The original game files, code and other copyrighted assets cannot be redistributed without permission. These protections do not extend to every form of DRM circumvention, but reverse engineering itself is not an illegal activity. That’s why Nintendo can’t do much about decomp tools, but they will take down videos showing games running on emulators.

Source: Chasm/Addy Osmani, Claude (Github)

Lisuan LX 7G100 driver adds 58 changes, FreeSync control and fixes Phantom Blade Zero launch

Lisuan has released another Windows driver for its LX 7G100 graphics card. Pilot Release 30.0.2260.896 arrived on October 29 and includes 58 changes, consisting of 43 fixes, 13 game performance optimizations and two feature additions. The update is available through the company’s official driver download center.

The driver adds a FreeSync control option to the Lisuan eXtreme software, allowing users to manage variable refresh rate support. It also introduces hardware decoding support for MPEG-1 and MPEG-2 video. These additions expand the card’s existing display and multimedia capabilities.

Phantom Blade Zero and other games receive compatibility fixes

One of the listed changes addresses a startup problem in Phantom Blade Zero, which launched on October 29. Lisuan also lists startup fixes for F1 24, F1 25, Avatar: Frontiers of Pandora, Star Wars Outlaws, Evil West and Resonance: A Plague Tale Legacy. Other fixes target crashes in The Witcher 3: Wild Hunt, Halo: Campaign Evolved and Call of Duty: Modern Warfare 2 Campaign Remastered.

The company also addresses rendering problems in several games, including Raft, Gothic 1 Remake, Borderlands 4, Red Dead Redemption 2, Gray Zone Warfare and The Blood of Dawnwalker. Separate fixes target abnormal video memory usage in Monster Hunter: World, PUBG and Shadow of the Tomb Raider. Death Stranding receives a fix for performance drops.

Performance optimizations cover 13 games, including Marvel’s Spider-Man Remastered, Dying Light 2, Nioh 2, Wo Long: Fallen Dynasty, Borderlands 3 and Shadow of the Tomb Raider. Lisuan does not provide specific performance improvements or benchmark results for these changes.

As we reported in September, the previous Pilot Release 30.0.2260.773 already included 67 changes, split between 34 fixes and 33 performance optimizations. The two releases therefore account for 125 listed changes in just over a month.

Lisuan LX 7G100 Driver 30.0.2260.896

1. Added FreeSync (variable refresh rate) control to the Lisuan eXtreme control panel.
2. Added hardware decoding support for MPEG-1 and MPEG-2.

Game launch problems

• Phantom Blade Zero: Fixed an issue preventing the game from launching.
• Resonance: A Plague Tale Legacy: Fixed a launch issue.
• 朽木难雕: Fixed a launch issue.
• F1 24: Fixed a launch issue.
• F1 25: Fixed a launch issue.
• Avatar: Frontiers of Pandora: Fixed a launch issue.
• Corsair Cove: Fixed a launch issue.
• Star Wars Outlaws: Fixed a launch issue.
• JX3 Online: Fixed a launch issue.
• Evil West: Fixed a launch issue.
• Warcraft III: Reforged – Forsaken Kingdom: Fixed a launch issue.

Game crashes

• Halo: Campaign Evolved: Fixed a crash.
• Call of Duty: Modern Warfare 2 Campaign Remastered: Fixed a crash.
• The Witcher 3: Wild Hunt (Next-Gen): Fixed a crash.
• RuneScape: Dragonwilds: Fixed a DX11 crash.

Rendering and graphics issues

• Raft: Fixed rendering issues.
• Gothic 1 Remake: Fixed rendering issues.
• Crimson Moon: Fixed rendering issues.
• Wardogs: Fixed rendering issues.
• Silent Hill: Fixed abnormal color artifacts.
• Gray Zone Warfare: Fixed rendering issues.
• Control Resonant: Fixed rendering issues.
• No More Room in Hell 2: Fixed rendering issues.
• Transport Fever 3: Fixed rendering issues.
• BeastLink: Fixed rendering issues.
• EA Sports FC 27: Fixed rendering issues.
• Dinoblade: Fixed rendering issues.
• The Blood of Dawnwalker: Fixed rendering issues.
• Borderlands 4: Fixed rendering issues in certain scenes.
• Red Dead Redemption 2: Fixed DX12 rendering issues.
• Bus Simulator 27: Fixed DX12 rendering issues.
• Justice Online: Fixed DX11 rendering issues.

Android emulator issues

• Hot-Blooded Jianghu: Awakening: Fixed Vulkan rendering problems in MuMu Player.
• 王者万象棋: Fixed Vulkan rendering problems in MuMu Player.
• Goose Goose Duck: Fixed OpenGL rendering problems in MuMu Player.

Flickering issues

• Cairn: Fixed screen flickering.
• Delta Force: Fixed screen flickering in DX11.
• Mafia III: Definitive Edition: Fixed screen flickering in DX11.

Video memory and performance issues

• Monster Hunter: World: Fixed abnormal VRAM usage.
• PUBG: Battlegrounds: Fixed abnormal VRAM usage in DX11.
• Justice Online: Fixed abnormal VRAM usage in DX11.
• Shadow of the Tomb Raider: Fixed abnormal VRAM usage in DX11.
• Death Stranding: Fixed frame rate drops.

Game performance optimizations

• Nioh 2: Improved game performance.
• Total War: Pharaoh: Improved game performance.
• Sword Art Online: Fractured Daydream: Improved game performance.
• Wo Long: Fallen Dynasty: Improved game performance.
• Tiny Tina’s Wonderlands: Improved game performance.
• BloomKeeper: Improved game performance.
• Tropico 6: Improved game performance.
• Marvel’s Spider-Man Remastered: Improved game performance.
• Dying Light 2 Stay Human: Improved DX12 performance.
• World of Warcraft: Forever: Improved DX12 performance.
• Mafia III: Definitive Edition: Improved DX11 performance.
• Borderlands 3: Improved DX11 performance.
• Shadow of the Tomb Raider: Improved DX11 performance.

Source: Lisuan

Radeon RX 9070 GRE reaches 3.9 GHz without hardware modifications, but there’s a catch

An AMD Radeon RX 9070 GRE has reportedly reached 3.9 GHz using custom overclocking software and a lightweight DirectX 9 workload. The result comes from overclocker ShrimpBrime, who developed two utilities called R-Team OC (RTOC) and GPU Frequency Generator. The reported clock is around 40% above AMD’s official 2.79 GHz boost specification.

First of all, it’s worth pointing out that GPU frequency records have little to do with gaming performance, unfortunately. Such extreme frequencies are usually achieved using specialized workloads with minimal computational requirements, such as simple mathematical calculations (like Pi). This test follows a similar approach, using a lightweight workload. However, unlike typical frequency record attempts, the GPU is at least rendering an actual 3D scene.

The test used an XFX Radeon RX 9070 GRE Swift with its original cooling system and VBIOS. No hardware modifications were required. According to the developer, the GPU maintained frequencies approaching 3.9 GHz under the custom workload rather than simply reporting a brief clock spike. However, the driver crashed and recovered shortly after the screenshots were captured.

Custom DirectX 9 test maintains higher GPU clocks

The GPU Frequency Generator uses DirectX 9 to render a relatively simple scene containing 2,048 triangles per draw call and 2,048 draw calls per frame. This produces approximately 4.19 million triangles per frame, a lightweight workload for a modern graphics card. The developer also reproduced frequencies of approximately 3.85 GHz using AMD’s own Radeon overclocking controls.

ShrimpBrime compared readings from GPU-Z and HWiNFO64 against telemetry collected through AMD’s ADL and ADLX interfaces. The RTOC utility polls monitoring data every 2 ms, although new power-management telemetry reportedly becomes available approximately every 15.5 ms.

The developer later clarified that the GPU consumed approximately 130W while running the specialized workload at around 3.8 GHz. Still quite impressive what RDNA4 silicon can achieve, although it has no use for everyday gaming. It wouldn’t surprise me if RDNA4 hits 4.0 GHz with the stock cooler soon. 

Source: TechPowerUP

GeForce RTX 2080 Ti 22GB tested in modern games, doubled VRAM improves performance

Modified RTX 2080 Ti with 22GB GDDR6 tested in gaming and AI workloads.

The GeForce RTX 2080 Ti launched eight years ago with 11GB of GDDR6 memory. While NVIDIA never released a 22GB version, hardware modders have been offering upgraded cards through Chinese marketplaces for several years. PC Games Hardware has now tested one of these modified GPUs in modern games and local AI workloads, comparing it against the original RTX 2080 Ti and the newer RTX 5070 Ti.

The tested card was purchased through Alibaba and provided by a reader. The modification replaces eleven 1GB memory chips with eleven 2GB Samsung GDDR6 modules, increasing capacity to 22GB. It retains the original TU102 GPU, 352-bit memory interface and 616 GB/s bandwidth. According to the testing, changing a configuration resistor allows the card to recognize the additional memory without a modified BIOS. Standard GeForce drivers also work.

Twice the memory helps in demanding games

PCGH tested nine games with settings selected to increase VRAM usage. At 1440p with DLSS Quality, the differences were generally small, although the 22GB version delivered more consistent frame times. Microsoft Flight Simulator 2024 and Marvel’s Spider-Man 2 showed larger gains, with the original 11GB card delivering approximately 20% lower frame rates due to memory limitations.

The difference increased at 4K with DLSS Performance and AMD FSR Frame Generation. Several games experienced severe stuttering on the 11GB model, while Indiana Jones and the Great Circle crashed under the selected settings. The modified card avoided these memory-related problems, although GPU performance remained a limitation. Gears of War: E-Day, for example, ran at just 17 FPS with RTX Mega Geometry enabled.

PCGH also recorded up to 13.9GB of VRAM allocation in Spider-Man 2, followed by 13.2GB in Pragmata and 12.9GB in Microsoft Flight Simulator 2024. These figures show why the original 11GB configuration struggled with the selected settings. The results do not mean every modern game requires more than 11GB, as the benchmarks deliberately focused on memory-intensive scenarios.

AI benchmarks favor the newer RTX 5070 Ti

PCGH used ComfyUI with Krea-2 Turbo and Qwen 3-VL 4B to generate images. The modified RTX 2080 Ti performed better when the workload exceeded 11GB of VRAM. In the most demanding test, the 22GB model completed the first image in approximately 315 seconds, compared with nearly 540 seconds on the standard card. The RTX 5070 Ti completed the same task in around 85 seconds.

The reason is partly architectural. Turing GPUs lack native FP8 support and must use FP16 for the tested workflow. This increased memory usage to as much as 21.5GB on the modified RTX 2080 Ti, while the RTX 5070 Ti needed less than 16GB. The tested card also had cooling limitations, with its blower fan approaching 4,000 RPM and producing 9.8 sone under load.

We have previously covered these modified cards, including ready-to-use RTX 2080 Ti 22GB models selling for $499 and a Japanese repair shop offering the upgrade for around $280. The latest test confirms that the doubled memory is usable in games and AI applications. However, the benefits depend on the workload, and buyers must also consider the age of the underlying hardware, modified memory chips and cooling performance. These cards are still supported by NVIDIA drivers and have even received the latest DLSS upgrades (but likely forget about official multi frame generation and neural rendering).

Source: PC Games Hardware

Windows 11 gaming interface expands support for handhelds, upscaling tools and cloud gaming

Winhanced has announced version 1.0 Beta of its Windows 11 gaming interface. The software brings games from Steam, Xbox, Epic Games Store, GOG, Battle.net, Ubisoft and other platforms into a single controller-friendly library. The update also adds integrated OptiScaler support, AMD FSR 4.1 compatibility and a new in-game overlay.

The new Library settings page includes an option to install OptiScaler automatically in compatible games. Users can select supported upscaling technologies, including NVIDIA DLSS, AMD FSR and Intel XeSS, without manually copying files. Winhanced says the integration will skip games using anti-cheat software. Another option allows AMD handheld users to replace FSR 3.1 with FSR 4.1, including on Radeon hardware that AMD does not officially support for FSR 4.

New overlay, game compression and GeForce NOW integration

Winhanced 1.0 introduces an overlay accessible through the controller’s Guide button. It provides access to performance settings, Dynamic TDP, display refresh rates, battery information, music, achievements, Discord and game guides. Users can also manage controllers, take screenshots and record gameplay without leaving the game. However, the overlay will remain exclusive to Patreon and Discord Supporter+ members during the 30-day beta period.

The update also brings game compression and storage management directly into the library. Users can move games between drives, compress installed titles or enable automatic compression for new installations. NVIDIA GeForce NOW integration allows supported games to launch through cloud streaming from the same library. Winhanced also adds a Theatre section for streaming services, an on-screen keyboard and cloud save backups.

More handhelds supported, free download scheduled for October 31

The software supports Windows 11 PCs and handhelds, including the ASUS ROG Ally, Lenovo Legion Go, MSI Claw and OneXPlayer families. Version 1.0 expands hardware controls to additional OneXPlayer devices and offers power management on supported GPD, AYANEO and other AMD-based systems. Some features, including fan curves, battery charge limits and button remapping, depend on the device.

Winhanced 1.0 Beta is scheduled to become available as a free download on October 31. At the time of writing, the official download page still lists version 0.9.9.96 from October 18, although the new 1.0 patch notes are already online. The developers plan to make the overlay available to all users after the 30-day beta period.

Winhaned 1.0 beta 

• OptiScaler integration: Automatic installation in supported games.
• FSR 4.1 support: Enables FSR 4.1 on unsupported AMD GPUs.
• In-game overlay: Performance, TDP, FPS, HDR, Discord and music controls.
• Game compression: Compress installed games to save storage space.
• GeForce NOW: Cloud gaming integrated into the game library.
• Expanded launcher support: Battle.net integration and custom collections.
• Controller improvements: Button remapping, gyro aiming and mouse controls.
• Cloud saves: Automatic backups to Dropbox, OneDrive or local storage.
• Quick Resume: Lower memory usage when suspending games.
• Gameplay recording: Built-in recorder with 30-second replay capture.
• PC streaming: Connect to multiple PCs and access their game libraries.
• More handheld support: Expanded OneXPlayer and other Windows 11 device compatibility.

Sources: Winhanced

Intel’s unreleased Larrabee GPU gets another chance, drivers and development tools recovered

Intel’s unreleased Larrabee GPU has reportedly been brought back to life, complete with its original development software and working Direct3D acceleration. Even more surprising, the nearly 17-year-old prototype is claimed to be running Crysis. The discovery comes from a storage-unit auction, where a Redditor purchased what appeared to be an old Intel desktop without realizing what was inside.

The discovery was shared by Redditor pkellythompson initially asking the community to help identify an unusual Intel graphics card. The system contained an Intel DX58SO motherboard, a large engineering sample GPU and a separate diagnostic board. It was later recognized as an Intel Visual Adrenaline development system containing a Larrabee prototype. The GPU initially refused to display an image over HDMI, but connecting an older monitor through DVI worked. During startup, the card displayed the message “Intel(R) Larrabee customizable video bios products.”

Larrabee drivers and development software recovered

With help from a friend, the owner eventually managed to boot the system into Windows 7. The problem turned out to be a BIOS configuration reset caused by a dead motherboard battery. Changing the storage controller configuration from AHCI to IDE allowed Windows to boot. The graphics card was subsequently confirmed to be a C0 stepping engineering sample, reportedly one of the later Larrabee hardware revisions.

The discovery became even more interesting when two USB drives containing Intel’s original Larrabee software were found among the stored equipment. Both have since been preserved and uploaded to the Internet Archive, with archives measuring approximately 82 MB and 728 MB. According to the owner, the recovered files include drivers, documentation, development tools and instructions for updating the GPU’s video BIOS. Following the installation procedure, Windows 7 reportedly recognized Larrabee correctly and enabled Direct3D acceleration. The system was then used to run older Microsoft Direct3D samples.

But can it run Crysis Apparently, yes

The owner later shared screenshots claiming that Larrabee could run the original Crysis at 1280 × 720 resolution, Medium settings and without anti-aliasing. An earlier test reportedly achieved more than 17 FPS. After discovering another driver version buried in the recovered files, the person testing the system claimed considerably better performance, with subsequent screenshots suggesting around 36 FPS. T

View on Imgur

Larrabee was Intel’s attempt to develop a discrete GPU using multiple x86-based cores with wide vector processing units and a largely software-controlled rendering pipeline. Intel originally planned to compete with AMD and NVIDIA in the graphics market, but cancelled the consumer product in 2010 after missing its development targets.

Source: Reddit, HardOCP

Nintendo Switch 2 emulation advances with Eglantine and Eden-NX2

Nintendo Switch 2 emulation is progressing quickly following the recent jailbreak of the console. An experimental build of Eden, developed by Takoroka and released on October 29, has already enabled several Switch 2 games to run on PC. One of its forks, Eglantine, can reportedly run Mario Kart World at 60 FPS, although performance depends on the hardware and emulator settings. The developers are also testing Splatoon Raiders and other Switch 2 titles.

The Eden-NX2 project adds several components required for Switch 2 emulation. These include a native NVIDIA Ampere shader frontend with Maxwell fallback, support for six emulated CPU cores, additional graphics and system services, AV1 and H.264 decoding, and native shader caching. 

Mario Kart World performance and GPU requirements

Eglantine has already received additional modifications targeting Mario Kart World version 1.8.0. A separate fork, Eglantine-MKW, released an update on October 30 that addresses GPU memory allocation, shader compilation pauses and several crashes. The developers added automatic BC3 texture recompression for graphics cards with less than 12GB of VRAM. They also fixed a shader compilation memory leak and improved cache handling to prevent corrupted data after an interrupted session. According to the developer, testing with a simulated 8GB graphics card resulted in approximately 6.6GB of VRAM consumption.

There are already several community performance reports. One Redditor running a Ryzen 7 5800X3D and GeForce RTX 5090 reported approximately 60 FPS in Mario Kart World, with occasional shader stuttering and graphical problems. Another tester using a Ryzen 5 5600G and GeForce RTX 5070 reported 40 to 53 FPS during gameplay, dropping to 40 to 45 FPS with two players.

Other Switch 2 games are also being tested

Development is not limited to Mario Kart World. Eglantine’s developers confirm that Splatoon Raiders can run at variable performance levels, although some users report graphical problems. One Radeon RX 7600 XT user reported that the game displayed a black screen after entering gameplay while the interface remained visible. Donkey Kong Bananza is also being tested, but a GeForce RTX 5090 user reported that it remained stuck on the loading screen. Meanwhile, Pokémon Pokopia has reportedly reached its language selection screen using additional experimental fixes, but remains far from confirmed playable compatibility.

Reaching 60 FPS in Mario Kart World is already possible on some systems, but that does not mean the game can maintain this performance across all courses or complete longer gaming sessions without problems.

Source: Eglantine (Github)

TUXEDO Nano Pro Gen15 launches with Core Ultra X7 358H

The new Panther Lake mini-PC uses upgradeable DDR5 SO-DIMM memory despite Intel listing only LPDDR5X support for the processor.

German PC manufacturer TUXEDO Computers has launched the Nano Pro Gen15, a compact Linux desktop powered by Intel’s Core Ultra X7 358H processor. The system uses Panther Lake architecture with 16 CPU cores, including four Performance cores, eight Efficient cores and four Low Power Efficient cores. It also features Intel Arc B390 integrated graphics with 12 Xe3 cores and a maximum GPU clock of 2.5 GHz.

TUXEDO offers two upgradeable DDR5 SO-DIMM slots supporting up to 128GB of dual-channel memory, with speeds listed up to 7200 MT/s. However, Intel’s official specifications for the Core Ultra X7 358H list only LPDDR5X-9600 support and a maximum capacity of 96GB. TUXEDO is therefore advertising a configuration that goes beyond Intel’s published memory specifications, although the manufacturer has not explained this difference.

Compact Linux PC with upgradeable RAM and storage

The Nano Pro Gen15 measures 110 × 117.5 × 49 mm and weighs approximately 1 kg. It includes two M.2 SSD slots, one supporting PCIe 5.0 x4 with 2242/2280 drives and another supporting PCIe 4.0 x4 with 2242 drives. Connectivity includes a 40Gbps USB4/Thunderbolt 4 port, a 20Gbps USB-C port, three 10Gbps USB-A ports, dual 2.5GbE connections and Intel Wi-Fi 7.

The mini-PC supports four simultaneous 4K displays using its two HDMI outputs and two USB-C connections. There is one limitation for Linux users: TUXEDO specifies HDMI 2.0b operation at 4K 60Hz under Linux, while Windows supports HDMI 2.1 at up to 4K 120Hz. The system ships with TUXEDO OS or Ubuntu, while Windows 11 is available as an option. TUXEDO also provides its Control Center utility with fan controls.

TUXEDO is accepting preorders starting at €1,749 (approximately $1,960) in Germany, including 19% VAT. This configuration includes 16GB of DDR5 memory, a 500GB SSD and TUXEDO OS. Shipments are scheduled to begin in mid-November.

Sources: Tuxedo Computers

AOC U32G5D features Samsung’s Penta Tandem QD-OLED technology

AOC has introduced the U32G5D, a new 31.5-inch gaming monitor featuring a 4K QD-OLED panel with a 180Hz refresh rate. The monitor was announced in China on October 21 and is now listed on AOC’s official website with its full specifications.

The U32G5D uses a Penta Tandem QD-OLED panel based on EL Gen.3 technology, featuring a five-layer OLED structure. AOC also includes an anti-reflective coating, a 0.03ms gray-to-gray response time and Adaptive-Sync support, including AMD FreeSync. The monitor supports a native 10-bit color depth and a resolution of 3840 × 2160.

DisplayHDR True Black 400 and DisplayPort 2.1

According to AOC, the panel delivers 250 nits of brightness at 100% APL in SDR and 450 nits at 10% APL in HDR. It carries VESA DisplayHDR True Black 400 certification and covers 100% of the sRGB and 99% of the DCI-P3 color spaces. AOC also claims an average Delta E below 1 in sRGB mode based on factory calibration.

Connectivity includes two HDMI 2.1 ports operating at 48 Gbps, one DisplayPort 2.1 port supporting UHBR13.5 at 54 Gbps and a 3.5mm audio output. The monitor also features a USB 3.2 Gen 1 hub with two 5Gbps downstream ports. AOC specifies that HDMI requires Display Stream Compression (DSC) to achieve 4K at 180Hz with 10-bit RGB or 4:4:4 output, while DisplayPort does not have this requirement.

The U32G5D includes a height-adjustable stand with tilt, swivel and pivot adjustments, along with a hexagonal base and VESA 100 × 100mm mounting support. AOC also provides OLED Care features, including pixel refresh and graphene-based heat dissipation.

AOC U32G5D

• Panel: 31.5-inch Penta Tandem QD-OLED (EL Gen.3)
• Resolution: 3840 × 2160 (4K UHD)
• Refresh rate: 180Hz
• Response time: 0.03ms GtG
• Brightness: 250 nits SDR (100% APL), 450 nits HDR (10% APL)
• HDR: VESA DisplayHDR True Black 400
• Color: 10-bit, 100% sRGB, 99% DCI-P3, ΔE < 1
• VRR: Adaptive-Sync, AMD FreeSync
• Ports: 2× HDMI 2.1 (48 Gbps), 1× DisplayPort 2.1 (UHBR13.5), 2× USB 5Gbps, 3.5mm audio
• Stand: Height, tilt, swivel, pivot, VESA 100 × 100mm

The monitor is listed on AOC’s Chinese website, but pricing and availability outside China have not been confirmed.

Sources: AOC

New NVIDIA GPU AI unlocker automates performance tweaks for RTX 20, 30 and 40 series

UEFI-based tool combines earlier NVIDIA Tensor Core unlocks into one utility.

Last week, we reported on several projects that unlocked additional AI computing performance on NVIDIA GeForce GPUs. These modifications can nearly double the throughput of certain Tensor Core operations, although most implementations require manual configuration and primarily target Linux. A new project called nv-uefi-unlocker aims to simplify this process by bringing the existing techniques together into a single UEFI-based tool that also works with Windows.

Developed by physicsdolphin, the open-source utility runs before the NVIDIA graphics driver loads. It applies firmware-level register changes that remove restrictions on certain Tensor Core instruction rates. The process is handled automatically after booting from a prepared USB drive, and the tool then hands control back to the operating system. Unlike VBIOS modifications, it does not permanently change the graphics card’s firmware, and the unlock must be reapplied after a cold boot.

RTX 20, 30 and 40 series support and results

The project currently supports GeForce RTX 20, RTX 30 and RTX 40 series graphics cards, including desktop and laptop models. GTX 16 series support is also being developed, although compatibility remains experimental. The GeForce RTX 50 series is not supported because the existing unlocking methods do not work with Blackwell GPUs.

There are also some requirements: Secure Boot must be disabled for the standard setup, while Turing GPUs need an additional GSP firmware configuration step. Not every GPU and system configuration has been independently tested.

• RTX 20 series (Turing): Verified on RTX 2070 (+67-75% FP32 GEMM performance) and RTX 2060 SUPER. Requires GSP firmware mode and restart.
• RTX 30 series (Ampere): Fully compatible since v2.1.0. Verified on RTX 3060 Ti LHR, with additional reports covering RTX 3060, 3070 Laptop, 3080 and 3080 Ti Laptop. Earlier community tools showed gains of 58-85%.
• RTX 40 series (Ada): Verified on RTX 4060 Laptop, with FP16 GEMM performance increasing from 29.89 to 49.33 TFLOPS (+65%).

AI performance gains depend on the workload

However, users should not expect these improvements to translate directly into every AI application. The unlocks primarily benefit compute-heavy operations using FP16, BF16 and certain TF32 workloads. Image generation, model training and some AI inference tasks can benefit, depending on the software. For example, one RTX 3080 LHR user reported a 28.8% improvement in ComfyUI SDXL performance, considerably below the gains measured in synthetic Tensor Core benchmarks.

One final reminder: unless you have an older GPU that you can afford to experiment with, we would not recommend using these tools for performance gains that may be relatively small in real-world workloads. Should NVIDIA remove these restrictions for everyone? Probably. It’s not like everyone can afford to buy a newer graphics card anyway. But is the potential performance improvement worth risking still working card? I’m not sure.

Sources: NV UEFI Unlocker

PS1 games demonstrated on jailbroken Switch 2 using the console’s own filesystem

Nintendo Switch 2 homebrew development continues following the release of the ns2jb jailbreak last week. Developer Izmysh now claims to have created the first fully native PlayStation 1 homebrew emulator for Nintendo’s latest console. The demonstration reportedly includes Gran Turismo 2, Silent Hill and Tekken 3 running on Switch 2 hardware.

According to Izmysh, the emulator operates natively on the console rather than streaming gameplay from a PC. The developer says it uses the Switch 2 filesystem and can save disc images to a microSD card. He describes it as the first 100% native emulator developed for Switch 2,

Switch 2 homebrew continues progressing after jailbreak

The developer previously demonstrated DOOM running through his IZDOOM project, which uses an interpreter running on the console’s GPU. He has also released Super File Bros., a file manager designed for accessing Switch 2 storage through the existing jailbreak. The new PS1 demonstration follows his recent work on GPU-to-CPU execution, which could enable more demanding homebrew applications without full kernel access.

As we reported last week, the ns2jb exploit allows homebrew applications to run on Switch 2 firmware 23.0.1. However, the jailbreak still requires an external computer and does not provide permanent custom firmware or kernel-level access. 

The PS1 emulator remains a proof of concept, with no publicly available source code or downloadable build to verify its implementation or performance. The developer has teased additional progress but has not announced a release date. Still, this is another development for Switch 2 homebrew, coming just over a week after the first public jailbreak and subsequent demonstrations of custom applications, game ports and emulation projects.

Can Nintendo stop this now? Probably not, just as Sony has been unable to stop progress on jailbroken PS5 consoles. Will Nintendo fight back against emulators and modding? I mean it’s Nintendo, so they almost certainly will.

Sources: Izmysh, IZDOOM

Community-developed plugins replace the SteamOS home screen with Xbox and PlayStation interfaces

A new Decky Loader plugin allows Steam Deck owners to replace the default SteamOS home screen with an interface based on the Xbox Series X|S dashboard. The project, called Decky Xbox Dashboard, was developed by GitHub user kromossomom-debug and is available as an open-source download. It changes the appearance of SteamOS Game Mode without replacing the underlying operating system.

The plugin adds Xbox-style navigation with Home, My games & apps, Store, Community and System tabs. Players can switch between sections using LB and RB, while pressing Y opens a game’s information page. This includes Steam Deck compatibility, achievements, installation options and estimated completion times from HowLongToBeat. Users can also pin games, customize their home screen and apply Xbox colors to Steam’s menus.

The same developer also made a PS5 dashboard

Interestingly, the developer has released a separate PS5 Dashboard plugin. This replaces the SteamOS home screen with a PlayStation 5-style interface, including horizontal game tiles, full-screen artwork, achievement cards and PlayStation-themed menus. Both projects support English and Portuguese and allow users to change their accent colors. The PS5 version also supports other handhelds running SteamOS, according to its documentation.

Both plugins require Decky Loader and must be installed manually using a ZIP file. Users can enable Developer Mode in Decky Loader’s settings and select the installation option under the Developer menu. The Xbox version is currently at 0.0.9-alpha, released on October 28. The latest update fixes game installation commands and several menu actions. The developer also recommends disabling other dashboard themes before activating the PS5 version to avoid conflicts.

These are unofficial interface modifications, not Xbox or PlayStation emulators. They do not add native console game compatibility or Xbox Game Pass support. For that you will need a different mod.

Sources: Xbox version, PS5 Version

BIOSTAR launches fanless Wildcat Lake mini-PC with 40 TOPS AI performance and support for -20°C environments

EdgeComp MS-U320 features Intel Core 5 320, dual 2.5GbE LAN and up to 64GB DDR5 memory.

BIOSTAR has introduced the EdgeComp MS-U320, a compact industrial mini-PC powered by Intel’s Wildcat Lake processor. The system uses passive cooling without fans and supports operating temperatures from -20°C to 50°C. It is designed for industrial automation, edge computing and environments where conventional cooling systems may not be suitable.

The MS-U320 features an Intel Core 5 320 processor with six cores and six threads, consisting of two Performance cores and four Low Power Efficient cores. The chip is manufactured using Intel 18A technology and offers a maximum boost clock of 4.6 GHz, with a 15W processor base power. As we previously reported, the same processor has already appeared in consumer laptops from Honor, ASUS and HP.

Up to 40 TOPS of combined AI performance

BIOSTAR advertises up to 40 TOPS of combined AI performance across the CPU, GPU and NPU. However, Intel officially rates the processor’s NPU at 16 TOPS, while its integrated graphics provide up to 20 TOPS. The system supports one DDR5-6400 SO-DIMM with up to 64GB of memory. Storage is handled by an M.2 2242/2280 slot connected through PCIe 4.0 x2.

Four serial ports and dual 2.5GbE networking

The mini-PC includes two 2.5GbE LAN ports using Intel I226-LM and I226-V controllers. It also offers four COM ports, two USB 3.2 Gen 2 ports, four USB 2.0 ports, HDMI 2.1 and DisplayPort 1.4. Additional M.2 slots support Wi-Fi, Bluetooth and optional 4G/5G connectivity through a SIM-equipped cellular module.

The fanless enclosure measures 167 × 135 × 60 mm and weighs 1.2 kg. It supports 9-24V DC power input and ships with a 90W adapter. BIOSTAR also offers optional DIN rail, wall and VESA mounting accessories. Supported operating systems include Windows 11, Windows 11 IoT Enterprise 24H2 and Ubuntu 24.04.

The company has not announced pricing or retail availability.

Sources: Biostar

GeForce RTX 5070 Ti drops below 150W using mVolt+ undervolting tool

A Redditor has managed to run a GeForce RTX 5070 Ti at just 0.669V using mVolt+ Extreme v0.47.2. The card was configured to operate at approximately 1,897 MHz with a -145mV voltage offset and a +300MHz core clock offset. According to the shared monitoring screenshot, GPU utilization reached 98%, while maximum board power was recorded at just 142.8W. That is less than half of the card’s official 300W power rating.

The monitoring data also shows a maximum GPU temperature of 47.8°C, with the hotspot reaching 51.8°C and memory temperature topping out at 56°C. However, there are no gaming benchmarks or extended stability tests to demonstrate how well the card performs at these settings. 

GB203 is also used in 175W laptop GPUs

The desktop RTX 5070 Ti uses NVIDIA’s GB203 processor, which is also found in the desktop RTX 5080 and the GeForce RTX 5080 and RTX 5090 Laptop GPUs. Both laptop variants are configured for considerably lower power consumption, with a maximum GPU power of 150W plus another 25W through Dynamic Boost, reaching 175W. These mobile configurations use different clock speeds, voltage settings and core configurations, but it’s no surprise it’s technically possible to run this GPU at a much lower power envelope.

Before you try this at home

As much as we enjoy seeing experiments like this, we would not recommend running desktop graphics cards at such low voltages without extensive testing. Other users in the Reddit thread have already reported driver resets and freezes when attempting to go below 0.8V. Aggressive undervolting can cause intermittent problems, including random freezes, driver crashes and black screens, even when a card appears to work normally in some games.

The experiment also shows that recently developed GPU tuning tools are not only being used for overclocking. mVolt+ has received several updates, including versions 0.50.8 and 0.50.9 released on October 27 and 28, which restored experimental voltage controls for RTX 50 series GPUs running driver 621.01 and later.

Enthusiasts frequently combine undervolting and overclocking to adjust power consumption and GPU frequency. For users who prefer NVIDIA’s own tuning software, the NVIDIA App includes a built-in Automatic Tuning feature under System > Performance. This remains NVIDIA’s official option for automatic GPU overclocking without using third-party tools.

Sources: Reddit, MVolt (Github)

Modders add Neural Rendering to S-GAME’s new action RPG

S-GAME released Phantom Blade Zero on PC and PlayStation 5 this week, and modders have already started adding features to the PC version. Among the first modifications is an unofficial NVIDIA DLSS 5 Neural Rendering implementation. Other mods offer more control over combat difficulty, including parry timing and an option to unlock the game’s hardest mode immediately.

Phantom Blade Zero officially supports DLSS 4.5 with Super Resolution, Dynamic Multi Frame Generation and Ray Reconstruction. However, the game does not include DLSS 5 Neural Rendering. A mod uploaded to Nexus Mods on October 29 attempts to add this feature using the existing NVIDIA integration rather than modifying the game’s executable.

The package, called DLSS5 MOD Install with one click, combines ReShade 6.8, the RenoDX DLSS addon and NVIDIA’s DLSSNR runtime. According to its creator, the mod was validated on a GeForce RTX 5090 D and reportedly also works with RTX 40 series GPUs. It requires DLSS Quality or Balanced mode and uses a DXGI proxy to redirect existing DLSS calls to the Neural Rendering runtime.

This is another example of the unofficial DLSS 5 implementations we have covered since modders first gained access to the NVIDIA runtime in August. However, the mod has not been independently benchmarked, and some users have already reported installation problems. 

Sources: PCCentral, DLSS5 Mod, DSO Gaming

One of just 10 ASUS Trinity prototypes from 2008 is now available to collectors

A rare ASUS EAH3850 Trinity graphics card from 2008 has resurfaced. The card belongs to collector Dragos Ionut, who contacted VideoCardz with photographs of what he describes as a working engineering sample marked “RD Sample #01.” The prototype also comes with its original Thermaltake Bigwater 760i liquid-cooling kit and accessories.

The ASUS Trinity was an experimental graphics card featuring three Radeon HD 3850 GPUs on a single PCB. Each RV670 GPU was installed on a separate MXM module with 512MB of GDDR3 memory, giving the card 1.5GB of total VRAM. The board also featured four DVI outputs and an 8-pin PCIe power connector.

ASUS developed the card as a proof of concept rather than a retail product. According to coverage from 2008, only 10 units were produced. The unusual cooling system used heatpipes to transfer heat from the three GPU modules to a shared liquid-cooled block.

The owner says the surviving sample includes handwritten engineering labels on both the card and its packaging. He is also open to selling the complete kit, although no asking price has been set. You can contact Dragos here. 

Photos courtesy of Dragos Ionut.

OpenAI not so open to sharing memory with PC users

Altman argues that cloud data centers can make better use of DRAM than personal computers.

OpenAI CEO Sam Altman has an interesting take on the ongoing memory shortage. According to him, having large amounts of RAM installed in personal computers is not the best use of the world’s limited DRAM supply. He believes data centers can make better use of that memory by sharing it across many users.

During a conversation with Stripe co-founder John Collison on the Cheeky Pint podcast, Altman discussed local AI and whether running open-source models on personal computers could reduce demand for cloud infrastructure. He wasn’t convinced that local AI was necessarily the better approach.

“from a global allocation perspective, actually it’s quite terrible to have everybody, like, have a bunch of the world’s limited supply of DRAM sitting in their own computer and being used five percent of the time”

— Sam Altman

Altman’s argument is that cloud data centers can keep their memory busy serving multiple users rather than leaving it sitting idle inside individual PCs. The 5% utilization figure was his own estimate, and he didn’t provide data to support it. He also didn’t offer a direct comparison with memory utilization in AI data centers.

Meanwhile, the AI industry is buying up memory

PC memory prices continue to rise as manufacturers shift production toward server products. Suppliers are reducing PC DRAM output in favor of server memory, further limiting availability for consumer hardware. So far, none of the big CEOs have shown any signs of slowing down their collaboration with ChatGPT creators. In fact, they invite Altman to special events almost every few months. 

Altman is wrong

Of course, PC memory isn’t used only for running AI models. Games, browsers, operating systems and professional applications all need RAM, even when users aren’t actively using them. Local AI also gives users the option to run models without sending their data to remote servers.

Altman wasn’t suggesting that people should stop buying RAM, but his argument clearly favors cloud infrastructure over running AI locally. Needless to say, we disagree with his view. We hope to see a shift in the conversation toward recognizing the importance of personal computing, rather than treating it as an inefficient use of hardware resources.

2025 and 2026 have been brutal years for the consumer PC market, and the sooner the AI spending boom slows down, the better. Some of the damage to the consumer market may take years to reverse, assuming it can be reversed at all. Manufacturers have already adjusted their production priorities and product roadmaps, and shifting those back toward consumers won’t happen overnight.

Source: Cheeky Pint