GPU Compute Model Terminology / Quick Reference / CheatSheet

GPU Compute Model Terminology.md

from Landom Thomas https://landonthomas.net/docs/gpu_compute_model_terms_quick_ref.pdf

Compute Abstraction Hierarchy	Microsoft HLSL	Khronos GLSL	Khronos OpenCL C	Nvidia,CUDA C	AMD,HIP	Apple,MSL
Entire Kernel/Shader Compute Space	Dispatch	Dispatch, Compute Space	NDRange, Index Space	Grid	Grid	Grid
Major Compute Group	Group, Thread Group	Workgroup, Local Workgroup	Work-group	Block, Thread Block	Block	Threadgroup
Minor Compute Group (& Device Minor SIMD Unit)	Wave	Subgroup	Sub-group	Warp	Wavefront, Wave, Warp	SIMD-group
Quad Compute Object	Quad Wave	Subgroup Quad	?	Quad	Quad	Quad-group
Single Compute Object	Thread	Invocation	Work-item	Thread	Thread	Thread

Hardware Abstraction Hierarchy	Microsoft HLSL	Khronos GLSL	Khronos OpenCL C	Nvidia,CUDA C	AMD,HIP	Apple,MSL
GPU / Compute Device	Device	Physical Device	Compute Device	Device	Device	Device
Major SIMD / Multi Processor Unit	SIMD Processor	Compute Unit (CU)	Compute Unit (CU)	Streaming Multiprocessor (SM)	Compute Unit (CU)	Compute Unit (CU)
Minor SIMD Unit(& Compute Minor Group)	Wave	Subgroup	Sub-group	Warp	Wavefront, Wave, Warp	SIMD-group
Single SIMD Processor	Lane	?	Processing Element (PE)	Streaming Processor (SP), Lane	Processing Element (PE), Lane	Lane

Memory Abstraction Hierarchy	Microsoft HLSL	Khronos GLSL	Khronos OpenCL C	Nvidia,CUDA C	AMD,HIP	Apple,MSL
Contiguous Device Memory ~L2+ cache	Device Memory	Buffer, Image Memory	Global Memory	Global Memory	Global Data Share (GDS), Global Segment	Device Memory / Address Space
Faster, Partitioned Memory ~L1 cache	Thread Group Shared Memory (TGSM)	Shared Memory	Local Memory	Shared Memory	Local Data Share (LDS), Local Segment	Threadgroup Memory / Address Space
Fastest, Smallest Memory ~L0 cache	Temporary Registers	Subgroup Memory	Private Memory	Local Memory	L0 vector cache, Private Segment	Thread Memory / Address Space