PyTorch 2.8 Released With Better Intel CPU Performance For LLM Inference

PyTorch 2.8 released today as the newest feature update to this widely-used machine learning library that has become a crucial piece for deep learning and other AI usage. There are a few interesting changes worth highlighting with the new PyTorch 2.8 release.

Piquing my interest with PyTorch 2.8 is improved Intel CPU performance. In particular, a focus on high performance quantized large language model (LLM) inference for Intel CPUs using the native PyTorch version. The change outlines the LLM quantization work done by Intel engineers
Continue reading...

Linux 6.17 KVM Additions Include Intel LKGS From FRED, Smarter AMD SEV Cache Flushing

The KVM feature changes were merged a few days ago with all of their enhancements for the in-development Linux 6.17 kernel. Some nice improvements made it this cycle for enhancing the open-source Linux virtualization stack.

KVM on the x86 (x86_64) side of the Kernel-based Virtual Machine includes sharing device posted IRQ code between AMD SVM and Intel VMX code to harden it against bugs and run-time issues. There are also improvements around the MMIO stale data mitigation
Continue reading...

Open Source LLM

Windows Subsystem For Linux "WSL" Updated For A Yet-To-Be-Public Security Vulnerability

Microsoft today released an updated version of Windows Subsystem for Linux "WSL" that allows running Linux binaries atop Windows 11. There is only one change noted and it's for a yet-to-be-public security vulnerability.

It looks like next week Microsoft will be making public a heavy-hitting WSL security vulnerability that they have gone ahead to release the new version of WSL in advance while confirming the CVE security disclosure details will be made public on 12 August. That date also happens to be this
Continue reading...

Open Source LLM

Microsoft Announces Open-Source "Wassette" Using Rust + WebAssembly To Help AI Agents

Microsoft formally announced today its newest open-source project: Wassette. The Wassette is MIT-licensed and includes Linux and macOS support alongside Windows while being a Rust+WebAssembly-based project focused on securing AI agents.

Microsoft's Wassette is built atop the Wasmtime run-time and allows for AI agents to autonomously fetch WebAssembly components from Open Container Initiative (OCI) registries and to be able to execute them. Via the WebAssembly runtime, Microsoft aims to have Wassette as secure as browser sandboxing. Plus Wassette provides additional
Continue reading...

Mesa 25.2 Released With Many Improvements For RADV, Intel & NVK Drivers

Mesa 25.2 is now available as the newest quarterly feature release for this set of open-source Linux graphics drivers predominantly for OpenGL, Vulkan, and video acceleration support on the Linux desktop.

Like usual most of the changes are dominated by improvements to the AMD Radeon RADV/RadeonSI drivers and Intel Iris/ANV drivers but the open-source NVIDIA "NVK" Vulkan driver continues picking up more hardware support and features too. The various smaller graphics drivers for IP found on different SoCs continues to be enhanced too

Intel Xe Enabling SR-IOV PF By Default, Marking Driver As "Broken" For Non-4K Kernels

On top of all the Linux 6.17 kernel graphics driver feature changes that landed last week like promoting Panther Lake's Xe3 graphics to on-by-default, SR-IOV for Battlemage GPUs, multi-GPU preparations, Wildcat Lake enablement work, and more, some additional Intel Xe kernel graphics driver changes were submitted today for merging to Linux 6.17 in the coming days.

Sent out today were the drm-xe-next-fixes for the week as primarily some "fixes"
Continue reading...

AlmaLinux Introduces Native NVIDIA Support Using Open-Source Kernel Driver

The AlmaLinux project announced today that there is now "native" NVIDIA graphics driver support for AlmaLinux 10 and AlmaLinux 9 using NVIDIA's open-source kernel modules that are now conveniently packaged in an AlmaLinux repository for easy usage complete with NVIDIA's closed-source user-space packages like CUDA.

AlmaLinux worked with NVIDIA to provide this native NVIDIA driver support on AlmaLinux 9 / 10 built around their modern open-source but out-of-tree kernel drivers for Turing and newer. With using the open
Continue reading...

Linux 6.17 Fixes A Performance Bottleneck In The Futex Code

Merged last week for Linux 6.17 were the FUTEX locking changes that include addressing an observed performance bottleneck.

Prominent Linux engineer Peter Zijlstra at Intel adapted the FUTEX code to use a RCU-based, per-CPU reference counting to address a performance bottleneck found within the existing code that used a single instance variant.

Peter explained with the patch addressing the FUTEX performance bottleneck:
"The use of rcuref_t for reference counting introduces a performance bottleneck when accessed concurrently by multiple threads during futex operations.

Replace rcuref_

Btrfs Sees Urgent Fix Following Recent Reports Of Log Tree Corruption

On Linux 6.15.3+ there have been increased reports of log tree corruption being hit by users of the Btrfs file-system. Fortunately, a fix has now been submitted for Linux 6.17 Git and then for back-porting to the recent stable kernel versions.

Peter Jung of CachyOS raised one month ago increased reports of corruption for the log tree with Btrfs. There have also been other reports too of Btrfs log tree corruption on Linux 6.15.3+ kernels. Fortunately

Valve's ACO Compiler Used By AMD Drivers Optimize Scheduling Heuristic For Newer GPUs

Merged today for Mesa 25.3-devel to benefit the RADV Vulkan and RadeonSI Gallium3D AMD drivers are improved scheduling heuristics for the ACO compiler back-end developed by Valve.

The ACO compiler can now enjoy improved scheduling heuristics to help with performance on newer AMD Radeon graphics processors. The existing ACO scheduling heuristics were catering to aging Polaris GPUs while now the code is better adapted for more recent GPUs.

Daniel Schürmann opened the merge request nearly one year ago to improve the scheduling heuristic for ACO.
Continue reading...