feat(server): support ROCM for /api/usage endpoint

[Set27]

I have read the CLA Document and I hereby sign the CLA

📝 Summary

Add AMD gpu stats supported
#9237

📋 Pre-Review Checklist

- [ ] For large changes, or changes that affect the public API: this change was discussed or approved through an issue, on Discord, or the community discussions (Please provide a link if applicable).

• Any AI generated code has been reviewed line-by-line by the human PR author, who stands by it.
• Video or media evidence is provided for any visual changes (optional).

✅ Merge Checklist

• I have read the contributor guidelines.
• Tests have been added for the changes made.
• [not sure if any] Documentation has been updated where applicable, including docstrings for API changes.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
marimo-docs	Ready	Preview, Comment	Jun 11, 2026 3:21pm

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[Set27]

I have read the CLA Document and I hereby sign the CLA

[Copilot]

Pull request overview

This PR extends the server /api/usage endpoint GPU reporting to support AMD GPUs via ROCm (rocm-smi), alongside the existing NVIDIA (nvidia-smi) implementation, so the frontend can show GPU memory usage on ROCm systems.

Changes:

• Detects whether NVIDIA or ROCm tooling is available and selects the appropriate stats collector.
• Refactors GPU collection into dedicated parsing helpers for nvidia-smi and rocm-smi.
• Adds a ROCm command definition for fetching VRAM stats via CSV output.

[Set27]

• Rewrite using json output instead of csv
• Fix current test
• Add new test

[Set27]

I would highly appreciate example output for nvidia-smi to add test for it

[mscolnick]

@cubic-dev-ai

[cubic-dev-ai]

@cubic-dev-ai

@mscolnick I have started the AI code review. It will take a few minutes to complete.

[cubic-dev-ai]

1 issue found across 2 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="marimo/_server/api/endpoints/health.py">

<violation number="1" location="marimo/_server/api/endpoints/health.py:452">
P2: ROCm GPU stats silently report all-zero memory when expected JSON keys are missing</violation>
</file>

Architecture diagram

sequenceDiagram
    participant Client as Client
    participant HealthAPI as /api/usage endpoint
    participant GpuDetect as _is_gpu_available()  
    participant NvidiaSMI as nvidia-smi process
    participant RocmSMI as rocm-smi process
    participant GpuParser as GPU stats parser
    
    Note over Client,GpuParser: GET /api/usage with ROCM GPU support
    
    Client->>HealthAPI: GET /api/usage
    HealthAPI->>HealthAPI: Collect CPU, memory, network stats
    
    HealthAPI->>GpuDetect: Check GPU availability
    alt No GPU tools found
        GpuDetect-->>HealthAPI: return False
        HealthAPI-->>Client: GPU stats = []
    else NVIDIA GPU detected
        GpuDetect->>NvidiaSMI: subprocess.run(_NVIDIA_GPU_STATS_CMD)
        NvidiaSMI-->>GpuDetect: CSV stdout
        GpuDetect-->>HealthAPI: return "nvidia"
        HealthAPI->>GpuParser: _parse_nvidia_smi_stats()
        GpuParser->>GpuParser: Parse CSV lines, handle [N/A]
        GpuParser-->>HealthAPI: list of GPU dicts
    else AMD ROCM GPU detected
        GpuDetect->>RocmSMI: subprocess.run(_AMD_GPU_STATS_CMD)
        RocmSMI-->>GpuDetect: JSON stdout (with possible WARNING prefix)
        GpuDetect-->>HealthAPI: return "rocm"
        HealthAPI->>GpuParser: _parse_rocm_smi_stats()
        GpuParser->>GpuParser: Strip warning lines, parse JSON
        Note over GpuParser: Extract card#, Card Series, VRAM bytes
        alt JSON parse error
            GpuParser-->>HealthAPI: return []
        else Success
            GpuParser-->>HealthAPI: list of GPU dicts
        end
    end
    
    alt GPU process failure
        HealthAPI->>HealthAPI: Log warning, continue
    end
    
    HealthAPI-->>Client: JSON response with GPU stats

Loading

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[mscolnick]

@Set27 looks good. some failing CI tests

[Set27]

@Set27 looks good. some failing CI tests

I can't reproduce CLI failing running the same command locally; I guess I run intro transient, so I rebase on the latest main.
Colud you start workflows one more time?

[mscolnick]

thank you for this feature @Set27 !