Further reading on accelerators

• AMD GPUs

  • Whitepaper: Introducing AMD CDNA^TM 2 Architecture which includes information on how the MI250X connects to the CPU, and on the vector and matrix cores of the GPU.

  • HPCwire article on the AMD MI300 APU that will be used in the El Capitan supercomputer.

  • AnandTech article on the MI300, based on information from AMD's 2022 Investors Day. The article shows some not very detailed slides about the memory architecture.

  • GPU comuting at the AMD Financial Analyst Day 2022 on YouTube (not an official AMD recording). The CDNA3 part starts at 1:07:50.

  • MI300 @ AMD CES'2023 keynote (YouTube). The MI300 part starts at 1:29:42.

• NVIDIA GPUs

  • Whitepaper: Summit and Sierra Supercomputers: An Inside Look at the U.S. Department of Energy’s New Pre-Exascale Systems for information on the physically unified memory space made possible by the NVLink connections between the CPUs and GPUs.

  • Whitepaper: NVIDIA Grace Hopper Superchip Architecture for all information on the combination of the Grace ARM-based CPU and the Hopper GPUs.

• NVIDIA CUDA

  • NVIDIA CUDA toolkit

• AMD ROCm

  • AMD ROCm information portal

• Intel oneAPI

  • Intel oneAPI overview

• OpenCL

  • Khronos Group OpenCL documentation

• OpenACC

  • The OpenACC Organization

  • Clacc, OpenACC in LLVM project page

• OpenMP

  • OpenMP Architecture Review Board web site

• SYCL

  • Khronos Group SYCL documentation

  • AdpativeCpp (formerly hipSYCL)

  • Intel oneAPI DPC++/C++ compiler

  • Intel oneAPI DPC++ compiler GitHub

• Kokkos

  • Kokkos ecosystem home page

• RAJA framework

  • RAJA home page

  • RAJA documentation

• Alpaka abstraction library

  • Alpaka information on the CASUS web site

  • Alpaka information on GitHub. The README is the best source of information on compatibility with various compilers and backend technologies.

  • Alpaka documentation

  • Alpaka training from 2020 on YouTube

• Libraries

  • MAGMA

  • heFFTe