{
  "id": 2619307,
  "title": "Why Ollama runs on CPU after an update: the missing sm_86 CUDA kernels",
  "url": "https://urgent.news/2026/08/22/why-ollama-runs-on-cpu-after-an-update-the-missing-sm-86-cuda-kernels",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-22T17:05:59.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/milkyway008/why-ollama-runs-on-cpu-after-an-update-the-missing-sm86-cuda-kernels-39bn"
  },
  "original_language": "en",
  "account": "After updating Ollama, the model you use daily might start running slower on your GPU. Instead of utilizing the GPU, which is often the case with RTX 30-series owners, it begins using the CPU almost exclusively. This change goes unnoticed for many users. The issue stems from the missing sm_86 CUDA kernels in the updated version (0.32.14) of Ollama. The affected hardware includes GPUs with compute capability 8.6, which are found in cards like the RTX 3090, 3080, 3070, 3060, A40, A6000, A5000, A10, and similar models. The problem wasn't entirely new, but the addition of a safety net in previous builds was broken in 0.32.14. Instead of falling back to the CUDA 12 library, Ollama skips to using the CPU, without any errors appearing. To resolve the issue, users are advised to revert to the last version where the CUDA 12 fallback worked, which is version 0.32.13. The fix involves uninstalling Ollama, installing the v0.32.13 version from the GitHub release page, and restarting the Ollama service. To verify if the GPU is now being used, users can run nvidia-smi during a generation. If the VRAM allocated to the ollama process is nonzero, and the log shows library=CUDA compute=8.6, then the GPU is being utilized correctly.",
  "summary": "Why Ollama runs on CPU after an update: the missing sm_86 CUDA kernels You update Ollama, pull the model you use every day, and everything gets slower. Not broken, not erroring. Just slow. ollama ps says something like 12% CPU / 88% GPU , so you shrug and carry on. Then you open nvidia-smi mid-generation and see Ollama using 0 MiB of VRAM. It's been running on CPU this whole time. Nobody told…",
  "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."
}