<![CDATA[Geeks for your information

<![CDATA[Geeks for your information - Tech News]]> https://www.geeks.fyi/ Tue, 28 Apr 2026 18:01:37 +0000 MyBB <![CDATA[AMD EXPO 1.2 now available, adds partial CUDIMM support and three new Chinese memory]]> https://www.geeks.fyi/showthread.php?tid=21919 Sun, 26 Apr 2026 11:40:52 +0000 harlan4096]]> https://www.geeks.fyi/showthread.php?tid=21919 Quote:1usmus reveals AMD EXPO 1.2 details

AMD EXPO 1.2 reportedly adds MRDIMM support, full CUDIMM support expected with Zen 6

AMD EXPO 1.2 is now said to bring a wider memory profile update than previously expected. According to 1usmus, the new revision adds support for module geometry and MRDIMM memory. CUDIMM and CSODIMM are also part of the EXPO 1.2 path, although CUDIMM support is still limited on current firmware.

Only partial support, full support with Zen6The developer claims AGESA 1.3.0.1 does not yet include full CUDIMM support. AMD is reportedly preparing that part for Zen 6, which suggests current AM5 platforms may expose EXPO 1.2 metadata before full clocked module support is available. This follows earlier coverage when HWiNFO added AMD EXPO 1.20 support and 1usmus linked the update to CUDIMM memory support.

Continue Reading... ]]> Quote:1usmus reveals AMD EXPO 1.2 details

AMD EXPO 1.2 reportedly adds MRDIMM support, full CUDIMM support expected with Zen 6

AMD EXPO 1.2 is now said to bring a wider memory profile update than previously expected. According to 1usmus, the new revision adds support for module geometry and MRDIMM memory. CUDIMM and CSODIMM are also part of the EXPO 1.2 path, although CUDIMM support is still limited on current firmware.

Only partial support, full support with Zen6The developer claims AGESA 1.3.0.1 does not yet include full CUDIMM support. AMD is reportedly preparing that part for Zen 6, which suggests current AM5 platforms may expose EXPO 1.2 metadata before full clocked module support is available. This follows earlier coverage when HWiNFO added AMD EXPO 1.20 support and 1usmus linked the update to CUDIMM memory support.

Continue Reading... ]]> <![CDATA[DLSS 4.5 SDK arrives with Dynamic Multi Frame Generation support]]> https://www.geeks.fyi/showthread.php?tid=21890 Wed, 22 Apr 2026 06:43:40 +0000 harlan4096]]> https://www.geeks.fyi/showthread.php?tid=21890 Quote:NVIDIA releases DLSS 4.5 SDK with Dynamic Multi Frame Generation and 6X mode

[Image: NVIDIA-DLSS-4.5-DYNAMIC-FRAME-GEN-HERO-1200x630.jpg]

[Image: NVIDIA-DLSS-4.5-DYNAMIC-FRAME-GEN-HERO-1200x630.jpg]

NVIDIA has released the DLSS 4.5 SDK for developers, adding support for DLSS 4.5 Dynamic Multi Frame Generation and Multi Frame Generation 6X. The SDK is now available and gives developers access to its newest Super Resolution and Frame Generation updates.

According to NVIDIA, the SDK is built on Streamline and is meant to give studios one integration path across DLSS features. NVIDIA also says the package includes updated APIs, documentation, and sample code, which should make it easier to add features such as Ray Reconstruction or Dynamic Multi Frame Generation to both new and existing games.

[Image: NVIDIA-DLSS-310_6_0-SDK-HERO-850x367.png]

[Image: NVIDIA-DLSS-310_6_0-SDK-HERO-850x367.png]

Source: Github

Continue Reading... ]]> Quote:NVIDIA releases DLSS 4.5 SDK with Dynamic Multi Frame Generation and 6X mode

[Image: NVIDIA-DLSS-4.5-DYNAMIC-FRAME-GEN-HERO-1200x630.jpg]

Source: Github

Continue Reading... ]]> <![CDATA[AMD ROCm 7.2.2 Brings Optimization Guide For Ryzen AI / RDNA 3.5 Hardware]]> https://www.geeks.fyi/showthread.php?tid=21879 Tue, 21 Apr 2026 06:57:25 +0000 harlan4096]]> https://www.geeks.fyi/showthread.php?tid=21879 Quote:ROCm 7.2.2 is out today as a small point release to this open-source AMD GPU compute stack. There are a few code changes but most notable is arguably on the documentation side.

It's been just a few weeks since ROCm 7.2.1 and thus ROCm 7.2.2 is on the very lightweight side. ROCm 7.2.2 brings a fix for a ROCTracer reporting failure, updated user-space/driver/firmware dependency details, and ROCm documentation updates.

On the documentation side there is now an AMD RDNA 3.5 system optimization page that outlines optimization details for AMD Ryzen AI NPUs with RDNA 3.5 graphics. This is focused on both Ryzen AI systems with RDNA 3.5 like Strix Point and the flagship Ryzen AI Max "Strix Halo" platforms that have seen much ROCm/compute interest from the community thanks to its very powerful integrated graphics.

The AMD RDNA 3.5 system optimization page outlines recommended memory settings/tunables, configuring shared memory limits under Linux, operating system support, Linux kernel version requirements, and other details.

Continue Reading... ]]> Quote:ROCm 7.2.2 is out today as a small point release to this open-source AMD GPU compute stack. There are a few code changes but most notable is arguably on the documentation side.

It's been just a few weeks since ROCm 7.2.1 and thus ROCm 7.2.2 is on the very lightweight side. ROCm 7.2.2 brings a fix for a ROCTracer reporting failure, updated user-space/driver/firmware dependency details, and ROCm documentation updates.

On the documentation side there is now an AMD RDNA 3.5 system optimization page that outlines optimization details for AMD Ryzen AI NPUs with RDNA 3.5 graphics. This is focused on both Ryzen AI systems with RDNA 3.5 like Strix Point and the flagship Ryzen AI Max "Strix Halo" platforms that have seen much ROCm/compute interest from the community thanks to its very powerful integrated graphics.

The AMD RDNA 3.5 system optimization page outlines recommended memory settings/tunables, configuring shared memory limits under Linux, operating system support, Linux kernel version requirements, and other details.

Continue Reading... ]]> <![CDATA[Mozilla's MZLA Technologies Launches Thunderbolt, an Open-Source Self-Hosted Enterpri]]> https://www.geeks.fyi/showthread.php?tid=21877 Mon, 20 Apr 2026 08:43:41 +0000 harlan4096]]> https://www.geeks.fyi/showthread.php?tid=21877 Quote:MZLA Technologies, a subsidiary of Mozilla, has launched Thunderbolt, an open-source AI client aimed at enterprise use on private infrastructure. The tool offers a unified interface for chat, search, and research functions, connecting to internal AI models, data sources, and automation pipelines.

The source code is available on GitHub, and organizations interested can join a waitlist at thunderbolt.io.

What Thunderbolt Does

[Image: Screenshot-2026-04-20-at-09.25.47-scaled.png]

[Image: Screenshot-2026-04-20-at-09.25.47-scaled.png]

Thunderbolt functions as an AI workspace that connects directly to an organization's existing systems, instead of routing data through external hosted AI services. Administrators can choose which AI models to run behind the interface, supporting commercial services, open-source models, and locally hosted systems.

Integration options include DeepSet's Haystack platform for backend orchestration, Model Context Protocol servers, and Agent Client Protocol agents. The Haystack integration connects the Thunderbolt interface with backend tools responsible for model selection, retrieval, and automated responses.

This setup allows organizations to connect internal pipelines without needing to rewrite existing infrastructure.

Automation Features and Cross-Platform Support

Thunderbolt offers automation features for scheduling and repeating tasks. This includes generating daily briefings, monitoring specific topics, compiling reports, and triggering actions based on incoming data. These functions can be set up within the client environment without relying on external cloud services.

Native applications are available for Windows, macOS, Linux, iOS, and Android, enabling staff to access the same AI environment across desktop and mobile devices.

Continue Reading... ]]> Quote:MZLA Technologies, a subsidiary of Mozilla, has launched Thunderbolt, an open-source AI client aimed at enterprise use on private infrastructure. The tool offers a unified interface for chat, search, and research functions, connecting to internal AI models, data sources, and automation pipelines.

The source code is available on GitHub, and organizations interested can join a waitlist at thunderbolt.io.

What Thunderbolt Does

[Image: Screenshot-2026-04-20-at-09.25.47-scaled.png]

Thunderbolt functions as an AI workspace that connects directly to an organization's existing systems, instead of routing data through external hosted AI services. Administrators can choose which AI models to run behind the interface, supporting commercial services, open-source models, and locally hosted systems.

Integration options include DeepSet's Haystack platform for backend orchestration, Model Context Protocol servers, and Agent Client Protocol agents. The Haystack integration connects the Thunderbolt interface with backend tools responsible for model selection, retrieval, and automated responses.

This setup allows organizations to connect internal pipelines without needing to rewrite existing infrastructure.

Automation Features and Cross-Platform Support

Thunderbolt offers automation features for scheduling and repeating tasks. This includes generating daily briefings, monitoring specific topics, compiling reports, and triggering actions based on incoming data. These functions can be set up within the client environment without relying on external cloud services.

Native applications are available for Windows, macOS, Linux, iOS, and Android, enabling staff to access the same AI environment across desktop and mobile devices.

Continue Reading... ]]> <![CDATA[Intel shows Texture Set Neural Compression, claims up to 18x smaller texture sets]]> https://www.geeks.fyi/showthread.php?tid=21779 Mon, 06 Apr 2026 07:39:15 +0000 harlan4096]]> https://www.geeks.fyi/showthread.php?tid=21779 Quote:TSNC Variant A delivers 9x texture compression, lower-quality B variant up to 18x

[Image: INTEL-TEXTURE-SET-NEURAL-COMPRESSION-TSN...00x624.jpg]

[Image: INTEL-TEXTURE-SET-NEURAL-COMPRESSION-TSN...00x624.jpg]

Intel has shared new details on Texture Set Neural Compression, or TSNC, a neural texture compression technology first shown as an Intel Labs research demo at GDC 2025. The company now says it has rebuilt the project into a standalone SDK, with a new decompression API that can compile to C, C++, or HLSL. Intel says the decoder supports both a fallback fused multiply-add path for CPU and GPU, and a linear algebra path that uses XMX acceleration on supported Intel GPUs.

This is yet another compression algorithm based on neural networks, followed by NVIDIA, who also held a GDC talk and mentioned up to 85% compression. Intel is working on two variants, one offering up to 9x compression and another up to 18x.

TSNC vs BC1

The basic idea is to go beyond standard block compression by training a neural network to encode and decode a set of textures together. Intel stores the latent data in a four-layer BC1-based feature pyramid, then reconstructs the original texture channels through a three-layer MLP decoder.

Continue Reading... ]]> Quote:TSNC Variant A delivers 9x texture compression, lower-quality B variant up to 18x

[Image: INTEL-TEXTURE-SET-NEURAL-COMPRESSION-TSN...00x624.jpg]

Intel has shared new details on Texture Set Neural Compression, or TSNC, a neural texture compression technology first shown as an Intel Labs research demo at GDC 2025. The company now says it has rebuilt the project into a standalone SDK, with a new decompression API that can compile to C, C++, or HLSL. Intel says the decoder supports both a fallback fused multiply-add path for CPU and GPU, and a linear algebra path that uses XMX acceleration on supported Intel GPUs.

This is yet another compression algorithm based on neural networks, followed by NVIDIA, who also held a GDC talk and mentioned up to 85% compression. Intel is working on two variants, one offering up to 9x compression and another up to 18x.

TSNC vs BC1

The basic idea is to go beyond standard block compression by training a neural network to encode and decode a set of textures together. Intel stores the latent data in a four-layer BC1-based feature pyramid, then reconstructs the original texture channels through a three-layer MLP decoder.

Continue Reading... ]]> <![CDATA[NVIDIA launches DLSS 4.5 Dynamic Multi Frame Generation and 6X mode through app beta]]> https://www.geeks.fyi/showthread.php?tid=21702 Wed, 01 Apr 2026 07:16:14 +0000 harlan4096]]> https://www.geeks.fyi/showthread.php?tid=21702 Quote:NVIDIA adds even more AI frames

[Image: NVIDIA-DLSS-MFG-6X-HERO-1200x656.jpg]

[Image: NVIDIA-DLSS-MFG-6X-HERO-1200x656.jpg]

NVIDIA has started rolling out DLSS 4.5 Dynamic Multi Frame Generation and Multi Frame Generation 6X through the opt-in NVIDIA app beta. The new overrides are limited to GeForce RTX 50 Series GPUs, and NVIDIA says GeForce Game Ready Driver 595.97 WHQL or newer is required. Needless to say, this update is only for RTX 50 owners, NVIDIA even provided a chart that shows RTX 40 clearly falling behind in supported features.

[Image: nvidia-dlss-4-5-feature-support-chart-850x465.png]

Source: NVIDIA

The new Dynamic mode replaces a fixed frame multiplier with an adaptive system that changes in real time based on GPU load and the display target. NVIDIA says the feature is designed to generate only the number of frames needed to reach a chosen cap or the monitor refresh rate, instead of staying locked to one preset multiplier throughout gameplay.

Continue Reading... ]]> <![CDATA[Google has officially moved its advanced ransomware detection and file restoration fe]]> https://www.geeks.fyi/showthread.php?tid=21701 Tue, 31 Mar 2026 09:00:20 +0000 harlan4096]]> https://www.geeks.fyi/showthread.php?tid=21701 Quote:Anti-Lag 2 is no more?

[Image: FSR-LATENCY-REDUCTION-2-HERO-1200x624.jpg]

[Image: FSR-LATENCY-REDUCTION-2-HERO-1200x624.jpg]

As you may recall, around November, just weeks before the Redstone release, AMD made a quiet change to its technology stack. With AMD, it is often hard to keep track of what the company is doing, because it tends to leave many changes unannounced. The quiet renaming of various FidelityFX technologies to simply “FSR” was not officially highlighted and was only noticed after we spotted the change on AMD’s website.

As a result, FidelityFX Super Resolution simply became FSR Upscaling. Frame Generation also received the FSR prefix, and AMD added FSR Ray Regeneration and Radiance Caching alongside the Redstone release. Redstone is easily AMD’s slowest technology rollout yet. It is currently available in just two games, and not even with the full stack, because Radiance Caching is not available anywhere yet.

Continue Reading... ]]> <![CDATA[AMD marks AM4’s 10th anniversary, says AM5 will follow the same philosophy]]> https://www.geeks.fyi/showthread.php?tid=21695 Mon, 30 Mar 2026 09:38:21 +0000 harlan4096]]> https://www.geeks.fyi/showthread.php?tid=21695 Quote:Zen4, Zen5 and Zen6: AM5 here to stay for long time

AMD is again pointing to platform longevity as one of the main selling points of socket AM5. In comments tied to the Ryzen 9 9950X3D2 Dual Edition announcement, Jack Huynh said AMD sees long-term socket support as a customer commitment rather than a simple marketing line. That fits AMD’s existing public guidance, which states AM5 support will continue through 2027 and beyond.

The timing of that message matters because AM4 is still active almost a decade after its debut. We just covered the Ryzen 5 5500X3D appearing at UK retailer, even though AMD has previously treated that processor as a regional part. That listing was another reminder that AM4 has outlived the normal desktop socket cycle and is still getting fresh retail movement in 2026.

Quote:“With AM4, we built a platform that made it easy for customers to upgrade their PC across multiple generations of processors. And this year, we celebrate its ten-year anniversary, supported by an expansive line of processors powering millions of systems around the world.

That same philosophy continues with socket AM5. Long-term platform support isn’t just a marketing message for us, it’s a commitment to protecting our customer’s investments and giving enthusiasts a platform that can grow with them.”

— Jack Huynh, Senior Vice President and General Manager of Computing and Graphics at AMD

Continue Reading... ]]> Quote:Zen4, Zen5 and Zen6: AM5 here to stay for long time

AMD is again pointing to platform longevity as one of the main selling points of socket AM5. In comments tied to the Ryzen 9 9950X3D2 Dual Edition announcement, Jack Huynh said AMD sees long-term socket support as a customer commitment rather than a simple marketing line. That fits AMD’s existing public guidance, which states AM5 support will continue through 2027 and beyond.

The timing of that message matters because AM4 is still active almost a decade after its debut. We just covered the Ryzen 5 5500X3D appearing at UK retailer, even though AMD has previously treated that processor as a regional part. That listing was another reminder that AM4 has outlived the normal desktop socket cycle and is still getting fresh retail movement in 2026.

Quote:“With AM4, we built a platform that made it easy for customers to upgrade their PC across multiple generations of processors. And this year, we celebrate its ten-year anniversary, supported by an expansive line of processors powering millions of systems around the world.

That same philosophy continues with socket AM5. Long-term platform support isn’t just a marketing message for us, it’s a commitment to protecting our customer’s investments and giving enthusiasts a platform that can grow with them.”

— Jack Huynh, Senior Vice President and General Manager of Computing and Graphics at AMD

Continue Reading... ]]> <![CDATA[Geekbench flags Intel Binary Optimization Tool results as potentially invalid]]> https://www.geeks.fyi/showthread.php?tid=21687 Thu, 26 Mar 2026 07:41:09 +0000 harlan4096]]> https://www.geeks.fyi/showthread.php?tid=21687 Quote:Geekbench Browser now flags Intel BOT-capable CPU results

Primate Labs says the Geekbench Browser will now attach a warning to Geekbench 6 CPU results from processors that support Intel’s Binary Optimization Tool, or BOT. The company says some Geekbench 6 workload scores can rise by up to 40% under BOT, while overall scores can increase by up to 8%. Primate Labs argues those runs are not comparable with standard Geekbench 6 submissions, and says it currently has no way to tell whether BOT was enabled for a specific result.

Intel launches BOT

Intel introduced BOT with its Core Ultra 200S Plus and 200HX Plus launches as an optional feature inside the advanced mode of Intel Application Optimization. Intel describes it as a binary-level optimization layer meant to improve IPC and native performance in select games and applications, even when the workload was tuned for another x86 processor, a console, or an older architecture. Intel’s support page says BOT requires Intel Dynamic Tuning Technology and currently works with a short list of titles, including Geekbench 6.3+, which Intel labels as a “Proof of Concept.”

Continue Reading... ]]> Quote:Geekbench Browser now flags Intel BOT-capable CPU results

Primate Labs says the Geekbench Browser will now attach a warning to Geekbench 6 CPU results from processors that support Intel’s Binary Optimization Tool, or BOT. The company says some Geekbench 6 workload scores can rise by up to 40% under BOT, while overall scores can increase by up to 8%. Primate Labs argues those runs are not comparable with standard Geekbench 6 submissions, and says it currently has no way to tell whether BOT was enabled for a specific result.

Intel launches BOT

Intel introduced BOT with its Core Ultra 200S Plus and 200HX Plus launches as an optional feature inside the advanced mode of Intel Application Optimization. Intel describes it as a binary-level optimization layer meant to improve IPC and native performance in select games and applications, even when the workload was tuned for another x86 processor, a console, or an older architecture. Intel’s support page says BOT requires Intel Dynamic Tuning Technology and currently works with a short list of titles, including Geekbench 6.3+, which Intel labels as a “Proof of Concept.”

Continue Reading... ]]> <![CDATA[AMD releases FSR 4.1 for Radeon RX 9000 series, uses algorithm shared with Sony’s upd]]> https://www.geeks.fyi/showthread.php?tid=21673 Fri, 20 Mar 2026 07:16:34 +0000 harlan4096]]> https://www.geeks.fyi/showthread.php?tid=21673 Quote:AMD releases Adrenalin 26.3.1 for Crimson Desert, highlights FSR 4.1 update

[Image: AMD-FSR41-CRIMSON-HERO-1-1200x633.jpg]

[Image: AMD-FSR41-CRIMSON-HERO-1-1200x633.jpg]

AMD has released Radeon Software Adrenalin Edition 26.3.1, the same day Crimson Desert went live (I would say it was long overdue, but that’s just AMD’s thing). Pearl Abyss says the game launched globally today and that the PC version supports FSR 3/4 and AMD FSR Ray Regeneration.

In AMD Gaming’s launch message for Crimson Desert, the company describes the update as FSR Upscaling 4.1 and says it improves ML-based upscaling with finer detail, smoother camera motion, and better Ultra Performance Mode behavior.

Source: AMD

That also clarifies the current hardware target for FSR 4.1. There were some rumors that AMD might unlock FSR4 for older architectures; however, latest drivers clearly mention support only for the Radeon RX 9000 series.

Continue Reading... ]]> <![CDATA[AMD talks RyzenClaw and RadeonClaw, local agents on AMD hardware, if only you had 128]]> https://www.geeks.fyi/showthread.php?tid=21657 Sat, 14 Mar 2026 10:33:17 +0000 harlan4096]]> https://www.geeks.fyi/showthread.php?tid=21657 Quote:AMD explains how to run OpenClaw on their hardware

[Image: AMD-RYZEN-CLAW-RADEON-CLAW-HERO-1200x624.jpg]

[Image: AMD-RYZEN-CLAW-RADEON-CLAW-HERO-1200x624.jpg]

AMD has published a new OpenClaw guide built around what it calls “RyzenClaw” and “RadeonClaw,” two AMD hardware paths for running local AI agents on Windows. The first is a Ryzen AI Max+ system with 128GB of unified memory, while the second uses a Radeon AI PRO R9700 graphics card. AMD is pitching both as ways to run OpenClaw locally through WSL2, LM Studio, and a local model setup rather than relying on the cloud.

[Image: AMD-RYZENCLAW-RADEON-CLAW-3-768x432.jpg]

Source: AMD

RyzenClaw & RadeonClawAMD’s “RyzenClaw” setup is not just any Ryzen AI Max+ laptop or mini PC, it specifically calls for a 128GB configuration, and AMD’s own instructions say users should reserve 96GB of variable graphics memory for that platform. That setup runs the Qwen 3.5 35B A3B model at about 45 tokens per second, supports a 260K token context window, and can handle up to six concurrent agents.

[Image: AMD-RYZENCLAW-RADEON-CLAW-1-768x433.jpg]

[Image: AMD-RYZENCLAW-RADEON-CLAW-4-768x429.jpg]

Source: AMD

Continue Reading... ]]> Quote:AMD explains how to run OpenClaw on their hardware

[Image: AMD-RYZEN-CLAW-RADEON-CLAW-HERO-1200x624.jpg]

Source: AMD

Continue Reading... ]]> <![CDATA[Scientists invent entirely new kind of storage – and it could change the course of hu]]> https://www.geeks.fyi/showthread.php?tid=21655 Fri, 13 Mar 2026 07:51:35 +0000 schreckdeividas]]> https://www.geeks.fyi/showthread.php?tid=21655 Quote:Scientists have invented a new kind of storage – and suggest that it could change the course of human history.

The storage uses laser-modified glass to encode information. And that information could last for more than 10,000 years, they say.

The world is generating far more information than it ever has. But storing that information has proven difficult: the hard disks inside of our computer, for instance, degrade relatively quickly, leading to fears that the vast amount of information we are generating could be lost one day soon.

Researchers have suggested in the past that storing that information in glass could be a helpful way of keeping it around for civilisation in the future. But it has so far proven impossible to actually write or retrieve that data.
Source----@Independent]]> Quote:Scientists have invented a new kind of storage – and suggest that it could change the course of human history.

The storage uses laser-modified glass to encode information. And that information could last for more than 10,000 years, they say.

The world is generating far more information than it ever has. But storing that information has proven difficult: the hard disks inside of our computer, for instance, degrade relatively quickly, leading to fears that the vast amount of information we are generating could be lost one day soon.

Researchers have suggested in the past that storing that information in glass could be a helpful way of keeping it around for civilisation in the future. But it has so far proven impossible to actually write or retrieve that data.
Source----@Independent]]> <![CDATA[AMD reveals “FSR Diamond” for Next-Gen Xbox, but is it RDNA5 exclusive?]]> https://www.geeks.fyi/showthread.php?tid=21653 Thu, 12 Mar 2026 10:15:08 +0000 harlan4096]]> https://www.geeks.fyi/showthread.php?tid=21653 Quote:AMD confirms FSR Diamond as a new codename during Project Helix reveal

[Image: FSR-DIAMON-HERO-VIDEOCARDZ-1200x620.jpg]

[Image: FSR-DIAMON-HERO-VIDEOCARDZ-1200x620.jpg]

Non-official AI renderAMD has now attached a new name to its next FSR stack. In a social media post published after Microsoft’s Project Helix presentation at GDC 2026, AMD SVP and GM Jack Huynh referred to the technology as AMD FSR Diamond, tying it directly to Xbox’s next-generation platform plans. Microsoft had already said Helix will use a custom AMD chip and a next-gen FSR path with machine learning features. Neither AMD nor Microsoft has confirmed RDNA5 as the GPU backbone of the new console.

Huynh says FSR Diamond as being built for next-generation neural rendering, with next-gen ML-based upscaling, a new ML-based multi-frame generation path, and next-gen ray regeneration for ray tracing and path tracing.

Continue Reading... ]]> Quote:AMD confirms FSR Diamond as a new codename during Project Helix reveal

[Image: FSR-DIAMON-HERO-VIDEOCARDZ-1200x620.jpg]

Non-official AI renderAMD has now attached a new name to its next FSR stack. In a social media post published after Microsoft’s Project Helix presentation at GDC 2026, AMD SVP and GM Jack Huynh referred to the technology as AMD FSR Diamond, tying it directly to Xbox’s next-generation platform plans. Microsoft had already said Helix will use a custom AMD chip and a next-gen FSR path with machine learning features. Neither AMD nor Microsoft has confirmed RDNA5 as the GPU backbone of the new console.

Huynh says FSR Diamond as being built for next-generation neural rendering, with next-gen ML-based upscaling, a new ML-based multi-frame generation path, and next-gen ray regeneration for ray tracing and path tracing.

Continue Reading... ]]> <![CDATA[Intel XeSS 3.0 SDK released, still not open source]]> https://www.geeks.fyi/showthread.php?tid=21646 Tue, 10 Mar 2026 10:22:48 +0000 harlan4096]]> https://www.geeks.fyi/showthread.php?tid=21646 Quote:Intel promised open-source XeSS four years ago, but XeSS 3.0 arrives as Windows binaries

[Image: INTEL-ARC-XESS3-HERO-1200x624.jpg]

[Image: INTEL-ARC-XESS3-HERO-1200x624.jpg]

Phoronix reports that Intel has updated its XeSS Software Development Kit on GitHub. The new release adds XeSS 3.0 support, including multi-frame generation and updated frame generation models. The first question was whether Intel had finally turned XeSS into an open-source project.

That still has not happened. In 2021, Intel’s Anton Kaplanyan said XeSS would be open source. Four years later, XeSS 1, XeSS 2, and now XeSS 3.0 have all arrived without a public source release for the core runtime components.

Intel does host the XeSS SDK on GitHub, and the package can be downloaded freely under its license. However, the actual release still relies on precompiled Windows DLL files rather than source code. Phoronix says the new SDK ships with binary-only Windows libraries, including libxess.dll, libxell.dll, and libxess_fg.dll.

Continue Reading... ]]> Quote:Intel promised open-source XeSS four years ago, but XeSS 3.0 arrives as Windows binaries

[Image: INTEL-ARC-XESS3-HERO-1200x624.jpg]

Phoronix reports that Intel has updated its XeSS Software Development Kit on GitHub. The new release adds XeSS 3.0 support, including multi-frame generation and updated frame generation models. The first question was whether Intel had finally turned XeSS into an open-source project.

That still has not happened. In 2021, Intel’s Anton Kaplanyan said XeSS would be open source. Four years later, XeSS 1, XeSS 2, and now XeSS 3.0 have all arrived without a public source release for the core runtime components.

Intel does host the XeSS SDK on GitHub, and the package can be downloaded freely under its license. However, the actual release still relies on precompiled Windows DLL files rather than source code. Phoronix says the new SDK ships with binary-only Windows libraries, including libxess.dll, libxell.dll, and libxess_fg.dll.

Continue Reading... ]]> <![CDATA[JEDEC publishes UFS 5.0 spec with up to 10.8 GB/s sequential throughput]]> https://www.geeks.fyi/showthread.php?tid=21618 Sun, 01 Mar 2026 08:17:26 +0000 harlan4096]]> https://www.geeks.fyi/showthread.php?tid=21618 Quote:KIOXIA starts sampling UFS 5.0 devices after JEDEC publishes the 5.0 spec

[Image: KIOXIA-UFS5-1200x624.jpg]

UFS 5.0 has been published, and the spec targets up to 10.8 GB/s for sequential read and write throughput. That is close to a 2x jump over the 5.8 GB/s figure commonly quoted for UFS 4.0 class devices, and it also exceeds the 8.0 GB/s one-direction bandwidth limit of a PCIe 4.0 x4 link on paper.

The speed increase is tied to the physical layer update. KIOXIA’s announcement points to MIPI M-PHY v6.0 and its HS-GEAR6 mode (up to 46.6 Gbps per lane), which is how a 2-lane configuration reaches about 10.8 GB/s effective performance.

The UFS 5.0 features include improved signal integrity, improved power integrity, and new data protection features, rather than only raising the peak transfer number.

Continue Reading... ]]> Quote:KIOXIA starts sampling UFS 5.0 devices after JEDEC publishes the 5.0 spec

[Image: KIOXIA-UFS5-1200x624.jpg]