Some Nomenclature

NVIDIA: 32 threads per warp - threads execute in lockstep with same instruction

AMD: 64 threads per wavefront (32 on RDNA) - similar to NVIDIA but larger groups

Apple Silicon: 32 threads per SIMD group

8
1
1
1
Thread Grid Layout
Threadgroup Boundary
SIMD Group Boundary (32 threads execute in lockstep)
Quad Group Boundary (4 threads for fragment derivatives)
Thread Details
Click on a thread to see details
Total Threads: -
Threads per Threadgroup: -
SIMD Groups per Threadgroup: -
Quad Groups per Threadgroup: -