MCPcopy
hub / github.com/Andyyyy64/whichllm / _is_vulkan_only_gpu

Function _is_vulkan_only_gpu

src/whichllm/engine/compatibility.py:37–48  ·  view source on GitHub ↗

Return True for legacy NVIDIA GPUs with no modern CUDA support. Kepler cards (compute capability 3.x) were dropped by CUDA 12 and current llama.cpp CUDA builds, so they only run through the Vulkan backend. Matches ``VULKAN_ONLY_GPUS`` entries as case-insensitive substrings of the GPU na

(gpu: GPUInfo)

Source from the content-addressed store, hash-verified

35
36
37def _is_vulkan_only_gpu(gpu: GPUInfo) -> bool:
38 """Return True for legacy NVIDIA GPUs with no modern CUDA support.
39
40 Kepler cards (compute capability 3.x) were dropped by CUDA 12 and current
41 llama.cpp CUDA builds, so they only run through the Vulkan backend. Matches
42 ``VULKAN_ONLY_GPUS`` entries as case-insensitive substrings of the GPU name,
43 the same convention used by the bandwidth/compute-capability lookups.
44 """
45 if gpu.vendor != "nvidia":
46 return False
47 name_upper = gpu.name.upper()
48 return any(marker.upper() in name_upper for marker in VULKAN_ONLY_GPUS)
49
50
51def _fit_candidate_gpus(gpus: list[GPUInfo]) -> list[GPUInfo]:

Callers 1

check_compatibilityFunction · 0.85

Calls

no outgoing calls

Tested by

no test coverage detected