AI Briefing#rocm
The latest AI and developer news about #rocm, with the original source and a short summary.
The latest AI and developer news about #rocm, with the original source and a short summary.
New Docker image bundles vLLM fork and ROCm 10.1 for Strix Halo, achieving 40-43 tok/s decode for Qwen3.8-Flash-Next.
The open-source plugin targets gfx1151 hardware to optimize MoE GEMMs and attention layers, reducing 256k prompt prefill time from over two hours to four minutes.
A CUDA-compatible build compiled for Windows based on ZLUDA has been shared, showing approximately 3% slower performance compared to native according to initial evaluations.
vLLM v0.28.0 adds large-scale performance optimizations and inference capabilities for the Kimi-K3 and DeepSeek V4 models.
AMD has released a fully open 16B MoE model trained on ROCm.
MSLK has organized Attention, GEMM, and MoE kernel usage in one place.
audio.cpp 0.5 adds expressive TTS, multilingual voice transfer, and AMD GPU support.
PyTorch Monarch has been extended to AMD Instinct GPUs and the ROCm environment, strengthening fault tolerance for large-scale distributed training.
llama.cpp is preparing an update that boosts prompt processing speed by 15% on ROCm environments and resolves certain quantization bugs.
Unsloth now officially supports local LLM inference and fine-tuning using AMD GPUs.
PyTorch Monarch now supports AMD GPU's ROCm environment, enabling uninterrupted, resilient distributed training even in the event of node failures.
Zluda 6, which helps run CUDA applications even without Nvidia GPUs, has been released.
This analyzes the FP8 standard mismatch and kernel optimization issues that arise when running DeepSeek-V4-Flash on AMD MI300X.
A native HIP W4A16 kernel that optimizes performance on AMD GPUs has been added to vLLM.
A major update for AMD/ROCm environments has been released in llama.cpp version B9387.
hipEngine, an open-source engine that maximizes inference performance for Qwen 3.6 models on AMD RDNA3 GPU environments, has been released.
ROCm 7.13 nightly adds Strix Halo optimizations and open-sources ROCprof.
PyTorch 2.12 significantly strengthens CUDA performance, export, and ROCm support.
CyberSecQwen-4B, a specialized 4B-scale model offering privacy protection and cost efficiency for cybersecurity defense, has been released.
A build method for using CUDA and ROCm simultaneously with llama.cpp on Windows was shared.
Starting with Ubuntu 26.04, Canonical will let users install CUDA and ROCm with just apt.
A llama.cpp benchmark comparing ROCm 6.4.2 and Vulkan on the 6900 XT was published.
GHOST v2.1 adds native Windows support, simplifying running AI models on AMD GPUs.