Skip to content

fix: ROCm Support for RDNA2 Device Models#2410

Open
gchriswill wants to merge 1 commit into
lemonade-sdk:mainfrom
gchriswill:rocm-rdna2-devices
Open

fix: ROCm Support for RDNA2 Device Models#2410
gchriswill wants to merge 1 commit into
lemonade-sdk:mainfrom
gchriswill:rocm-rdna2-devices

Conversation

@gchriswill

@gchriswill gchriswill commented Jun 25, 2026

Copy link
Copy Markdown

PR fixes #2403

This pull request fixes support for AMD RDNA2 GPUs (RX 6000 series) in the ROCm architecture detection logic. The fix includes updated device-to-architecture mapping logic in both the code and documentation.

Device detection and mapping improvements:

  • Updated the identify_rocm_arch_from_name function in system_info.cpp to explicitly include all relevant RDNA2 model numbers (6950, 6900, 6800, 6750, 6700, 6650, 6600, 6500) in the mapping to gfx103X, ensuring accurate detection of these GPUs.
  • Updated the backend support table in README.md to document the addition of gfx103X (RDNA2) support and list all corresponding Windows-compatible Radeon RX 6000 series GPUs.

@github-actions github-actions Bot added engine::llamacpp llama.cpp backend (LlamaCppServer); GPU/CPU LLM inference (Vulkan, ROCm, Metal) runtime::rocm AMD ROCm runtime enhancement New feature or request labels Jun 25, 2026
@jeremyfowers jeremyfowers requested a review from superm1 June 25, 2026 14:41

@superm1 superm1 left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think drop the test and then it's fine.

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Honestly I don't see the point of this test. Missing hardware IDs will always happen and the test won't avoid them in future.

@gchriswill gchriswill Jun 25, 2026

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Since there was a modification in the conditional statement logic that checks RDNA2 IDs, and further noticing the absence of test coverage, I thought it might have been good idea to have it. 😅

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

But you're right about the IDs and the test. And the only way for the test to catch anything requires the same maintenance (update to constants) as the code itself.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I will remove it now.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@superm1 Done ✅

Fixes RX 6000 series detection in system_info.cpp to include 6950, 6900, 6750, and 6650 for gfx103X mapping. Adds test_identify_rocm_arch_from_name.py with comprehensive unit tests replicating the C++ logic, covering all RDNA2 keywords, case handling, vendor guard, and bypass scenarios.

Signed-off-by: Christopher Gonzalez <4576678+gchriswill@users.noreply.github.com>
@gchriswill gchriswill force-pushed the rocm-rdna2-devices branch from 5dd45d6 to 33fb9f2 Compare June 25, 2026 18:23
@superm1 superm1 enabled auto-merge June 25, 2026 18:45
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

engine::llamacpp llama.cpp backend (LlamaCppServer); GPU/CPU LLM inference (Vulkan, ROCm, Metal) enhancement New feature or request runtime::rocm AMD ROCm runtime

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Missing ROCm Support for RDNA2 Device Models (6950 XT, 6900 XT, 6900, 6750 XT, 6650 XT)

2 participants