Skip to the content.

Vista Jupyter Kernels

R, C, C++, Bash, and Julia notebooks on TACC Vista, installed by one script.

Vista’s notebook environment gives you Python and nothing else. Everything else installs per-user through conda, and several kernels do not work once installed until you patch them. These pages document what actually works, including the failure modes and their fixes.

Quick start

On Vista, from a compute node (idev -p gh -N 1 -n 1 -t 01:00:00):

curl -O https://raw.githubusercontent.com/ashleyscruse/vista-jupyter-kernels/main/setup-kernels.sh
bash setup-kernels.sh --all

Then shut your Jupyter session down completely and start a new one from the TACC Analysis Portal. Restarting the kernel is not enough; the Jupyter server caches kernelspecs at startup.

Guides

Language Kernel name Guide
R Python (R) R
C, C++ 8 variants, including OpenMP C and C++
Bash Bash Bash
Julia Julia Julia

Fortran is unavailable: lfortran has no linux-aarch64 build. gfortran still works for compiled code.

The three things that trip everyone up

Restart the Jupyter session, not the kernel. Kernelspecs are cached when the server starts. This costs people more time than anything else here.

Kernelspecs must carry their own environment. Jupyter calls an environment’s Python directly and never runs conda activate, so PATH is not what you expect. Symptoms: rpy2’s Unable to determine R_HOME, or C++ cells failing with use of undeclared identifier 'std'.

Vista is ARM64. NVIDIA Grace CPUs. Use the aarch64 Miniforge installer, and expect some packages to have no build at all.

Source

github.com/ashleyscruse/vista-jupyter-kernels