GPU architectures

[lucaparisi91]

This is meant to exaplain the main concepts of a GPU architecture that are relevant to GPU programming.
Base material on L11-Accelerated_Architectures.pptx.

For instance explaining:

• presence of multiple SMs

• High bandwidth, high latency . Hide latency with multiple threads and fast switching contexts

• SIMT execution ( blocks are assigned to SMs, threads execute in warps/wavefronts )

• overview of memory hierarchy: global memory, shared memory, caches

• GPU and CPU have separate address spaces. Maybe mention unified shared memory ( Grace/Hopper, MI300 architecture )

• Example architectures ( Volta, AMD MI200 )

[NMannall]

Initial introduction could be based on sections 1.01 from the main ARCHER2 GPU course: https://github.com/EPCCed/archer2-gpu-course/tree/main

Likely worth condensing a little