Bouchet AMD nodes
YCRC has added additional compute nodes to the Bouchet HPC cluster. YCRC has historically deployed Intel-based compute nodes. All the new nodes, however, will utilize AMD processors for the first time in YCRC clusters. We have added:
- 26 CPU nodes: 128-core AMD EPYC 9575F processors with 2.2TB RAM per node
- 8 GPU nodes: 128-core AMD EPYC 9575F processors with 8 NVIDIA RTX Pro 6000 Blackwell Server Edition GPU cards (96GB vRAM each) per node
- 6 GPU nodes: 128-core AMD EPYC 9575F processors with 8 NVIDIA B200 GPU cards (193GB vRAM each) per node
Both Intel and AMD processors use the x86-64 instruction set, but there are architectural differences that can impact the software compilation process and application performance.
Partitions
For testing, the AMD nodes are in the following partitions:
- CPU-only AMD nodes:
day_amd - RTX Pro 6000 Blackwell nodes:
gpu_rtx6000 - B200 nodes:
gpu_b200
For detailed information about job limits and available compute nodes in each partition, please refer to our Bouchet partition documentation.
Software modules
We have recompiled most software modules with compilation flags that enabled them to run on both Intel and AMD nodes. If you are running any software compiled with CUDA on RTX Pro 6000 Blackwell or B200 GPUs, please use the modules installed with CUDA 12.8.0. Software built with older CUDA versions may fail or result in poorer performance. If you see any missing modules or experience issues with software modules, please reach out to us.
Compiling your code
If you compile your code optimized for Intel node architecture, it will not run on AMD nodes, and vice versa. To run your code on both Intel and AMD nodes, you need to compile it with a compiler flag that specifies a compatible instruction set. You can compile your code on either types of compute node.
Recommended compiler flag:
- GCC compilers (
gcc,g++,gfortran):-march=x86-64-v4 - Legacy Intel compilers (
icc,icpc,ifort):-axCORE-AVX512 - New Intel compilers (
icx,icpx,ifx):-march=x86-64-v4 - NVHPC compiler (
nvc,nvc++,nvfortran):-tp x86-64-v4
If you are using Intel compilers, we recommend compiling and running your workflow on the Intel nodes.
If you would like to compile your code optimized for a specific node architecture to potentially achieve better performance, please request an intereactive compute sesssion on that architecture (devel for Intel nodes and day_amd for AMD nodes) and compile your code with --martch=native for GCC and -xHost for Intel compilers. Please run your code only on the same architecture on which it was compiled.
If you need any assistance compiling your code, please reach out to us.
Report Issues
If you discover issues when running your workflow or experience performance issues, feel free to contact us for assistance. Please include the working directory, the commands that were run, the software modules used, and any more information needed to reproduce the issue.