Tag
xdna-top is a terminal monitor that shows both NPU and iGPU activity on Ryzen AI Max/Strix Halo systems, providing an honest view of NPU counter deltas instead of fake utilization percentages.
A researcher discovered a remote code execution vulnerability in AMD's AutoUpdate software due to insecure HTTP download links and lack of certificate validation. AMD initially dismissed it as out of scope but later agreed to issue a CVE and fix after public attention.
Reports running DiffusionGemma 26B on four AMD 7900 XTX GPUs using vllm, achieving 100 tps generation with overall 45-60 t/s, sharing performance metrics and setup commands.
AMD's Lemonade SDK for local AI adds NVIDIA CUDA support in version 10.7, enabling the same local AI server experience on competitor GPUs. The release also introduces lemonade bench for cross-backend LLM benchmarking and broader Vulkan support.
AMD touts unified memory architecture as a key enabler for next-gen products like the Ryzen AI MAX 400 series (Gorgon Halo), shaping their product roadmaps for AI and compute workloads.
A user benchmarks Google's Gemma 4 QAT models on an AMD 7900 XTX, reporting up to 45% faster generation, 83% higher throughput, and significant VRAM savings (e.g., 5.7GB for the 12B QAT model) with no quality loss compared to standard weights.
Review of AMD's Radeon RX 9070 GRE GPU, which offers fewer cores, less memory, and lower bandwidth than the RX 9070 at the same price, making it a disappointing release due to shrinkflation.
NVIDIA has released a 550B parameter model (Nemotron-3-Ultra-550B), prompting commentary that AMD and Intel should similarly release their own AI models as model development becomes a commodity for hardware companies.
The US Commerce Department closed a loophole that let Chinese AI companies purchase advanced Nvidia and AMD chips through overseas subsidiaries, extending export-licence rules to cover entities headquartered in China regardless of physical location. The guidance targets future shipments and does not affect existing hardware.
At Computex 2026, AMD announced extended AM5 socket support through at least 2029 and re-released the Ryzen 7 5800X3D as a 10th Anniversary Edition for AM4, offering budget-friendly CPU upgrade options.
A roundup of major announcements from Computex 2026, including Nvidia's RTX Spark laptop chips, Microsoft Surface Laptop Ultra, Dell XPS 13, and other hardware launches.
AMD at Computex 2026 promises AM5 motherboard support through 2029, relaunches the Ryzen 7 5800X3D for AM4's 10th anniversary, introduces the Ryzen 7 7700X3D, and expands the Radeon RX 9070 GRE globally, emphasizing the longevity of its existing platforms.
A developer announces winning an AMD Strix Halo laptop via the Lemonade SDK Developers Contest and invites others to participate by building projects on AMD hardware.
A new packed16 K technique for llama.cpp on RDNA3 GPUs reduces KV cache VRAM by 47% compared to Vulkan fp16, using int8 packing and native dot4 instructions to maintain fp16-quality K values with minimal KLD loss.
Intel's upcoming Panther Lake processor is positioned to compete in local AI with up to 180 TOPS, surpassing AMD's Ryzen AI 400, amid the Microsoft-Nvidia-Windows-Arm N1X discussions.
A community post requests that AMD restore Linux support for the free (BASIC) version of the Vivado FPGA design suite.
The author compares various GPUs for LLM inference, critiquing common benchmarks and emphasizing the importance of prefill performance over generation speed, offering recommendations for different budgets and use cases.
llama.cpp version b9387 introduces MFMA support for AMD CDNA architecture (MI100, MI200, MI300 series), improving processing pipeline performance on datacenter AMD GPUs.
A rejected PR for llama.cpp provides up to 30% faster prompt processing for MOE models on AMD Strix Halo hardware, with gains diminishing at higher context lengths.
Converted the Qwen 3.6 35b a3b model to ROCmfp4 format, leveraging MTP benefits for improved performance on AMD hardware.