{
  "id": 8546463,
  "title": "AMD Local LLM Setup on Windows/Linux: ROCm Overrides & Vulkan vs. HIP Benchmarking",
  "url": "https://urgent.news/2026/09/19/amd-local-llm-setup-on-windows-linux-rocm-overrides-vulkan-vs-hip",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-19T21:47:27.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/xanpavle/amd-local-llm-setup-on-windowslinux-rocm-overrides-vulkan-vs-hip-benchmarking-15p2"
  },
  "original_language": "en",
  "account": "Local artificial intelligence (AI) setups on AMD hardware can run into two main challenges: inadequate drivers or unrecognized GPU architectures. To address these issues, two solutions emerged - ROCmFix and InferBench.\n\nROCmFix is a Python utility that automatically resolves the HSA_OVERRIDE_GFX_VERSION problem. It works by querying PCI IDs through Windows Registry or Linux lspci. The tool allows setting permanent or temporary environment variables for different command lines, such as CMD, PowerShell, Bash, Zsh, and Fish. Users can also run rocmfix doctor to check the installed HIP SDK and Vulkan components.\n\nInferBench is another solution that automates the testing of local LLM inference speeds on AMD GPUs. The benchmarking workflow includes executing warm-up queries and forcing VRAM unloads between runs to avoid memory fragmentation or caching bias. InferBench then calculates the median tokens per second (tok/s) and Time-to-First-Token (TTFT), providing valuable insights for optimizing performance. Both ROCmFix and InferBench are available on GitHub for further exploration and use.",
  "summary": "The Problem with Local AI on AMD Hardware Running models locally with tools like Ollama or LM Studio on AMD GPUs often leads to two roadblocks: Missing drivers or unrecognised GPU architectures requiring HSA_OVERRIDE_GFX_VERSION manual tweaks. Uncertainty over whether to run Vulkan backends or ROCm/HIP backends for maximum tokens per second. Solution 1: ROCmFix ROCmFix is a single-file Python…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}