[GH-ISSUE #10178] Option to disable CPU fallback for SOC with unified memory #68735

Open
opened 2026-05-04 15:02:07 -05:00 by GiteaMirror · 1 comment
Owner

Originally created by @Hanselltc on GitHub (Apr 8, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/10178

For SOCs with unified memory like Apple Silicon or AMD APUs, the behaviour of falling back to system RAM when VRAM is insufficient causes Ollama to not use the GPU without the benefit of utilizing extra system RAM.

Please add an option to disable CPU fallback, either as an environmental variable or an updated fallback behaviour for these SOCs.

Originally created by @Hanselltc on GitHub (Apr 8, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/10178 For SOCs with unified memory like Apple Silicon or AMD APUs, the behaviour of falling back to system RAM when VRAM is insufficient causes Ollama to not use the GPU without the benefit of utilizing extra system RAM. Please add an option to disable CPU fallback, either as an environmental variable or an updated fallback behaviour for these SOCs.
GiteaMirror added the feature request label 2026-05-04 15:02:07 -05:00
Author
Owner

@JasonHonKL commented on GitHub (Apr 8, 2025):

#10155

<!-- gh-comment-id:2786970239 --> @JasonHonKL commented on GitHub (Apr 8, 2025): #10155
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#68735