Dashboard API: /api/gpu/detailed returns 503 on Apple Silicon (router never calls get_gpu_info_apple)

[yasinBursali]

Summary

GET /api/gpu/detailed returns 503 "No GPU data available" on Apple Silicon installs, while GET /api/features.gpu correctly reports the Apple M-Series GPU on the same service. The detection function exists and works — the router just doesn't wire it for GPU_BACKEND=apple.

Reproduction

On a healthy Apple Silicon install (GPU_BACKEND=apple in .env):

API_KEY=$(grep '^DASHBOARD_API_KEY=' ~/dream-server/.env | cut -d= -f2-)
# Fails:
curl -s -H "Authorization: Bearer $API_KEY" http://127.0.0.1:3002/api/gpu/detailed
# → {"detail":"No GPU data available"}  (HTTP 503)

# But this succeeds on the same backend:
curl -s -H "Authorization: Bearer $API_KEY" http://127.0.0.1:3002/api/features | jq .gpu
# → {"name":"Apple M-Series (24 GB Unified)","vramGb":24.0,"tier":"Strix Halo Compact"}

Root cause

dream-server/extensions/services/dashboard-api/routers/gpu.py:42-49:

def _get_raw_gpus(gpu_backend: str) -> Optional[list[IndividualGPU]]:
    """Return per-GPU list from the appropriate backend, with fallback."""
    if gpu_backend == "amd":
        return get_gpu_info_amd_detailed()
    result = get_gpu_info_nvidia_detailed()
    if result:
        return result
    return get_gpu_info_amd_detailed()

No apple branch. When GPU_BACKEND=apple the code falls through to NVIDIA → AMD detection (both fail on Apple hardware) → empty list → HTTPException(503) at line 108.

Meanwhile gpu.py::get_gpu_info_apple() exists and works (proven by /api/features.gpu).

Suggested fix

Add an Apple branch in _get_raw_gpus(). Since Apple Silicon is a single integrated GPU with unified memory, wrap get_gpu_info_apple() into a single-element list[IndividualGPU]:

def _get_raw_gpus(gpu_backend: str) -> Optional[list[IndividualGPU]]:
    if gpu_backend == "apple":
        info = get_gpu_info_apple()
        return [_to_individual(info)] if info else None
    if gpu_backend == "amd":
        return get_gpu_info_amd_detailed()
    result = get_gpu_info_nvidia_detailed()
    if result:
        return result
    return get_gpu_info_amd_detailed()

_to_individual() would convert the existing GPUInfo into an IndividualGPU (set gpu_id=0, copy memory/name fields). Alternatively, add get_gpu_info_apple_detailed() in gpu.py that returns list[IndividualGPU] directly.

Not a regression

git blame shows _get_raw_gpus() was introduced by commit c3e54d23 which is already on upstream/main. This bug pre-dates the current batch of open PRs; it's a feature gap from when multi-GPU monitoring was added (NVIDIA+AMD only).

Workaround

Dashboard Monitor / per-GPU cards will simply not populate on Apple. Use /api/features (already works) or the CLI (dream-cli gpu status — Apple-aware since PR Light-Heart-Labs#999/Light-Heart-Labs#1016).

[yasinBursali]

Shipped as upstream PR: Light-Heart-Labs#1025