A software stack is built for each architecture (x86-64, Arm) in the high-security zone. The complete list of modules is shown in each link:
- x86-64: virtual machines (VMs), non-GH200 compute nodes in Rio
- Arm: GH200 compute nodes in Rio

Running software on GH200

The Arm software stack is automatically mounted to /apps when you are on a GH200 node. You cannot use modules from this stack elsewhere.

Conda users must load the miniforge module on a GH200 node to create conda environments. You cannot use conda environments created on x86.

Prebuilt binaries must be Arm-based. Look for arm, arm64, aarch64, etc. in the download options.

Containers must be Arm-based. In your pull command add --arch arm64 or else it will default to x86. (You can pull on an x86 machine.)

To build from source for Arm, load the compiler of choice (gcc or nvhpc) on a GH200 node with compiler flags -mcpu=neoverse-v2 (gcc) or -tp=neoverse-v2 (nvhpc) for optimization. If you get errors related to ld, also load binutils.