AI Briefing#vulkan
The latest AI and developer news about #vulkan, with the original source and a short summary.
The latest AI and developer news about #vulkan, with the original source and a short summary.
The release removes deprecated APIs and adds FP8 support, YaRN truncation controls, and performance improvements across CUDA, Metal, Vulkan, and other backends.
The new kernel leverages native matrix core shapes and SLM architecture, delivering significant speedups on ARL-H, PTL-H, and B70 Pro platforms.
The open-source project replicates the 71-block Swin/ViT architecture of DLSS-NR build 310.8.0 using FP8 tensor cores, achieving byte-for-byte intermediate matches with the original.
ik_llama.cpp has added support for DSpark models and Vulkan quantization features.
llama.cpp has rolled out an update including Vulkan backend performance improvements and support for new models such as Granite and LFM2.5.
An ExecuTorch integration has been released that lets React Native apps run Gemma 4 offline.
The CppStudio framework has been released to help AI coding agents perform specialized GPU programming.
A build method for using CUDA and ROCm simultaneously with llama.cpp on Windows was shared.
A TurboQuant inference path was added to Runtime Valkyr, achieving 120 tok/s on Gemma with an RTX 3090.
A llama.cpp benchmark comparing ROCm 6.4.2 and Vulkan on the 6900 XT was published.
A Mesa PR boosted llama.cpp pp performance on Intel Xe2's Linux Vulkan by 37~130%.
Qwen3.6 35B-A3B showed a prompt processing speed of 282 tokens per second on a Radeon 780M iGPU.
In llama.cpp benchmarks, the Arc Pro B70 is significantly slower than the 3090, but SYCL improves things somewhat.
Combining an RX 9070 and RTX 5070 Ti, the author ran Qwen 35B Q4 at over 100tps.
Comparing Vulkan and ROCm on a Strix Halo 128GB machine showed the winner varying by model.