GPU Feature Cap

What a stock macOS VM reports

The paravirtualized device runs work on the host GPU but can report a conservative feature level. In our tested workload, selected newer paths stayed behind that reporting boundary.

Apple 9 SIMD-matrix · bfloat16 · 64 KB Tested profile
Apple 8Metal 3 / 4
Apple 7Metal 3 / 4 · SIMD-group matrix
Apple 6Metal 3
Apple 5 Metal 2 · no matrix path Stock VM reports here
Apple 4Metal 2
Apple 3Metal 2
Apple 2Metal 1
Apple 1Metal 1
Paths behind the cap
  • Metal 3 / Metal 4
  • SIMD-group matrix multiply
  • bfloat16
  • 64 KB threadgroup memory
    (up from 32 KB)
Tested capability profile
Stock guest report

Feature levels are illustrative. Apple groups GPU capability into numbered families with Metal feature sets layered on top. This graphic ends at Apple 9, the ceiling used in our test. The stock guest reported roughly Apple 5 with SIMD-group matrix support unavailable. The configured profile changed selected answers so llama.cpp could choose newer paths. Each additional feature requires its own virtualization test.