[PR #15875] docs: clarify ROCm visible device behavior #62036

Open
opened 2026-04-29 16:59:32 -05:00 by GiteaMirror · 0 comments
Owner

📋 Pull Request Information

Original PR: https://github.com/ollama/ollama/pull/15875
Author: @samrusani
Created: 4/29/2026
Status: 🔄 Open

Base: mainHead: codex/clarify-rocm-cuda-visible-devices


📝 Commits (1)

  • 2414652 docs: clarify ROCm visible device behavior

📊 Changes

2 files changed (+11 additions, -1 deletions)

View changed files

📝 docs/gpu.mdx (+10 -0)
📝 envconfig/config.go (+1 -1)

📄 Description

Summary

  • Clarify that ROCm may honor CUDA_VISIBLE_DEVICES for HIP compatibility on hybrid NVIDIA/AMD systems
  • Point users toward OLLAMA_LLM_LIBRARY=rocm when they want to prefer ROCm over CUDA
  • Update the server config description for CUDA_VISIBLE_DEVICES so it no longer implies NVIDIA-only behavior

Duplicate check

  • Searched all PR states for #15752, CUDA_VISIBLE_DEVICES ROCm, CUDA_VISIBLE_DEVICES AMD, HIP_VISIBLE_DEVICES, and ROCR_VISIBLE_DEVICES
  • Found related GPU discovery/configuration work, but no PR documenting this specific ROCm/HIP compatibility behavior

Validation

  • git diff --check
  • go test ./envconfig could not be run locally because go is not installed in this environment

Refs #15752


🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.

## 📋 Pull Request Information **Original PR:** https://github.com/ollama/ollama/pull/15875 **Author:** [@samrusani](https://github.com/samrusani) **Created:** 4/29/2026 **Status:** 🔄 Open **Base:** `main` ← **Head:** `codex/clarify-rocm-cuda-visible-devices` --- ### 📝 Commits (1) - [`2414652`](https://github.com/ollama/ollama/commit/24146528c5d41f9e6032ed962001c692dfe9272b) docs: clarify ROCm visible device behavior ### 📊 Changes **2 files changed** (+11 additions, -1 deletions) <details> <summary>View changed files</summary> 📝 `docs/gpu.mdx` (+10 -0) 📝 `envconfig/config.go` (+1 -1) </details> ### 📄 Description ## Summary - Clarify that ROCm may honor `CUDA_VISIBLE_DEVICES` for HIP compatibility on hybrid NVIDIA/AMD systems - Point users toward `OLLAMA_LLM_LIBRARY=rocm` when they want to prefer ROCm over CUDA - Update the server config description for `CUDA_VISIBLE_DEVICES` so it no longer implies NVIDIA-only behavior ## Duplicate check - Searched all PR states for `#15752`, `CUDA_VISIBLE_DEVICES ROCm`, `CUDA_VISIBLE_DEVICES AMD`, `HIP_VISIBLE_DEVICES`, and `ROCR_VISIBLE_DEVICES` - Found related GPU discovery/configuration work, but no PR documenting this specific ROCm/HIP compatibility behavior ## Validation - `git diff --check` - `go test ./envconfig` could not be run locally because `go` is not installed in this environment Refs #15752 --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
GiteaMirror added the pull-request label 2026-04-29 16:59:32 -05:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#62036