Providers¶
Providers detect vendor-specific hardware and telemetry while keeping the public API concept-first.
| provider | platform | status |
|---|---|---|
AppleGPU |
Apple Silicon macOS | GPU model, cores, Metal support, unified memory |
AppleNPU |
Apple Silicon macOS | Neural Engine identity and Core ML backend |
NvidiaGPU |
Linux + CUDA | CUDA architecture, SM count, memory, clocks where supported |
AMD, Intel, and Qualcomm providers are import-safe stubs today. They return unavailable so imports and CI do not require hardware or vendor SDKs.
Provider details should add telemetry, not rename the public concepts. A GPU is still a GPU. CUDA, Metal, ROCm, Level Zero, and Core ML are backend details.