Skip to Content
Changelog Execution target plugins and HPC support

Execution target plugins and HPC support

New in q8s version v0.14.0

You can now install execution targets as separate plugins and run quantum workloads on HPC CPUs and AMD GPUs.

Execute workloads across Kubernetes and HPCThe selected execution target routes a Python workload through Kubernetes to CPU or NVIDIA GPU nodes, or through interLink and Slurm to HPC CPUs or AMD GPUs, with HPC execution tested on LUMI.Qubernetes clusterHPC SYSTEM · SLURMQuantum workloadq8sctl execute app.py --target …Kubernetes APISubmit the workloadCPU nodes--target cpuNVIDIA GPU nodes--target gpuinterLinkHPC virtual nodeHPC CPUs--target hpc-cpuAMD GPUs · ROCm--target hpc-rocmHPC execution pathHPC execution tested on LUMI

Select an execution target to run your workload on Kubernetes nodes or on HPC resources through interLink.

Install execution target plugins

The core q8s package includes the cpu target. Install additional plugins for the execution environments you use:

PackageTargetExecution environment
q8scpuKubernetes CPU nodes
q8s-cudagpuKubernetes nodes with NVIDIA GPUs
q8s-hpc-cpuhpc-cpuHPC CPU nodes
q8s-hpc-rocmhpc-rocmHPC AMD GPUs using ROCm

For example, to run workloads on NVIDIA GPUs:

pip install "q8s==0.14.0" q8s-cuda

CUDA execution has moved to the q8s-cuda plugin. Install it when upgrading to keep using --target gpu in your existing projects.

Run workloads on HPC resources

The new hpc-cpu and hpc-rocm targets submit workloads to an HPC system through a Kubernetes cluster connected using interLink . To use them, you need access to a cluster with this integration and the HPC resources.

HPC execution has been tested on the LUMI supercomputer .

Install the plugin for your HPC environment:

pip install q8s-hpc-cpu

For AMD GPUs using ROCm, install q8s-hpc-rocm instead.

Add the selected target and its Python dependencies to your Q8Sproject, then build its container image. Create an HpcConfig file alongside Q8Sproject with the virtual node and Slurm allocation settings for your HPC environment:

HpcConfig
--q8s-node your-hpc-virtual-node --account your-project --partition your-partition --time 00:30:00

Execute your program with the selected target:

q8sctl execute app.py --target hpc-cpu

Use --target hpc-rocm to run on AMD GPUs. To use a configuration file in another location, pass --hpc-config /path/to/HpcConfig.

Extend Qubernetes with new targets

Qubernetes discovers target plugins automatically through the q8s.targets Python entry point group. You can add new execution targets in separate Python packages without changing the q8s core.

Created on Last updated on