AMD GPU servers

Overview

The Hopefog and Livestong PODs both have two AMD GPU servers, which enable powerful Machine Learning (ML) workflows.

2U, Gigabyte G291 enclosures https://www.gigabyte.com/GPU-Server/G291-Z20-rev-100#ov
8 AMD Radeon Instinct MI50 GPUs (https://www.amd.com/en/products/professional-graphics/instinct-mi50)
- each with 32 GB HBM2 memory
48-core/96-hyperthread EPYC 7642 CPU
512 GB RAM
1.9 TB Samsung NVMe SSD

Best starting places:

ROCm Video series
- https://community.amd.com/t5/instinct-accelerators-blog/rocm-open-software-ecosystem-for-accelerated-compute/ba-p/418720
- Especially the Introduction to AMD GPU Hardware: Link
  - Provides hardware background and terminology used throughout other guides
- Also
  - GPU Programming Concepts
    - Part 1 - HIP framework (like NVIDIA CUDA): Link
    - Part 2 - Device management, synchronization, MPI programming: Link
    - Part 3 - Device code, shared memory & thread synchronization: Link
  - GPU Programming Software (compilers, libraries & tools): Link
AMD ROCm resources Learning Center: https://developer.amd.com/resources/rocm-resources/rocm-learning-center/
- Especially:
  - Introduction to ROCm (Video, PDF)
  - Introduction to HIP (Video, PDF)
  - Introduction to Deep Learning on ROCm (Video, PDF)