Skip to content

AMD

Work in Progress

This page is not yet complete. Content is being added gradually. Functionality may be incomplete or subject to change. Thank you for your patience!

The FTP provides access to a range of AMD compute nodes combining high-core-count EPYC Milan CPUs with AMD Instinct GPU accelerators. These systems are well-suited for HPC workloads, GPU-accelerated computing, and evaluation of the ROCm software stack.

All AMD nodes are installed in the x86 cluster and are accessible via the x86 login node.

Hardware Overview

Four node types are available, differing in the attached GPU accelerator:

Node type CPU GPU Queue
AMD Milan + MI100 AMD EPYC Milan (2×64 cores) AMD Instinct MI100 amd-milan-mi100
AMD Milan + MI210 AMD EPYC Milan (2×64 cores) AMD Instinct MI210 amd-milan-mi210
AMD Milan + MI250 AMD EPYC Milan (2×64 cores) AMD Instinct MI250 amd-milan-mi250
AMD Instinct MI300A AMD EPYC (integrated) AMD Instinct MI300A (APU) amd-milan-mi300

The MI300A is an Accelerated Processing Unit (APU) that integrates CPU cores, GPU compute units, and HBM3 memory on a single package, eliminating the traditional CPU–GPU data transfer bottleneck.

Access

AMD nodes are available to all FTP users. See the FTP overview for how to request access.

Using AMD Nodes

Connect to the x86 login node and submit jobs to one of the AMD partitions:

ssh username@ftp-x86-login.scc.kit.edu

Use sbatch or salloc to access the desired partition, e.g.:

sbatch -p amd-milan-mi250 script.sh

or, for interactive use:

salloc -p amd-milan-mi300 -t 30

Software for AMD

AMD GPU nodes support the ROCm (Radeon Open Compute) software stack, which provides HIP, OpenCL, and various AI/ML frameworks. Existing CUDA code can often be ported using the hipify tool.

Hello World Example