AMD GPUs

Evidence (8)

• 8Hacker News·9/28/2026funding
  AMD to Acquire World Labs

  World Labs has signed a definitive agreement to join AMD, bringing its team and research into the chipmaker. The companies say the deal will combine World Labs' AI software and foundation-model work with AMD's hardware platform, with Dr. Fei-Fei Li joining AMD as Executive Vice President and Chief Scientist. The transaction is expected to close by the end of 2026 pending regulatory approval.

• 6Hacker News·9/13/2026open_source
  CUDA Compatibility Layer for AMD GPUs on Windows

  An open-source project aims to let CUDA-based applications run on AMD GPUs under Windows by translating CUDA calls to AMD-compatible GPU APIs. The repository has drawn strong attention on Hacker News, indicating interest in broader GPU software compatibility outside NVIDIA hardware. The effort is early-stage and primarily targets developers who want to run CUDA workloads on AMD systems.

• 5Hacker News·9/7/2026research
  Speculative Decoding Performance in vLLM on AMD GPUs

  This article explains how speculative decoding works in vLLM and compares several draft-and-verify methods, including native MTP, Gemma 4 MTP, EAGLE-3, DFlash, and DSpark. The authors report that throughput gains varied significantly depending on the drafting approach, proposal length, model family, draft checkpoint, workload, and token acceptance behavior on AMD Instinct MI300X and MI355X GPUs.

• 8Hacker News·8/6/2026funding
  AMD acquires Taalas for silicon-based AI inference chips

  AMD has acquired Toronto-based startup Taalas, which designs model-specific chips that etch model weights directly into silicon to accelerate AI inference. Early demos of Taalas' HC1 chip reportedly reached 16,960 tokens per second, and AMD may integrate the technology into its rack-scale AI systems alongside Instinct GPUs.

• 7Hacker News·8/4/2026open_source
  DeepSeek V4 Flash Running on a Single AMD MI300X

  A GitHub project claims to demonstrate DeepSeek V4 Flash running on a single AMD MI300X GPU. The repository appears to focus on making a large model fit and run efficiently on one accelerator, which is notable given the model’s scale and hardware demands. The post drew substantial Hacker News attention, suggesting strong interest in practical high-performance inference techniques.

• 7Hacker News·8/2/2026research
  Running Kimi K3 on AMD MI355X with Better Performance per Dollar Than B300

  The post reports benchmark results for serving the Kimi K3 model on AMD MI355X GPUs, claiming better performance per dollar than Nvidia B300 in this setup. It also describes a ROCm-side bug in speculative decoding infrastructure and notes that a small PyTorch fix was needed to stabilize the scheduler.

• 6Hacker News·7/3/2026research
  Wafer reports faster GLM5.2 inference on AMD MI355X

  Wafer says it ran GLM5.2 on AMD MI355X with 2626 tokens per second per node at 2.4 RPS, while claiming the setup cost is more than 2x lower than Blackwell-based alternatives. The post also describes using MXFP4 quantization with AMD Quark and serving the model with sglang on ROCm after fixing speculative decoding support.

• 8Hacker News·6/28/2026research
  LineShine Becomes No. 1 on the TOP500 at ISC 2026

  At ISC 2026, the TOP500 list crowned LineShine in Shenzhen, China as the new No. 1 supercomputer, powered by a large CPU-only Armv9 system. The report also notes LineShine's strong HPCG performance, Italy's new HPC7 system entering the top ranks, and Fugaku remaining highly competitive despite its age.