For more than two decades, Warewulf has been a quiet workhorse of high-performance computing. Originally built at Lawrence Berkeley National Lab in 2001, it became a staple for provisioning and managing clusters at scale. Now, as AI reshapes compute demands, CIQ is reintroducing Warewulf as Warewulf Pro — a modernized version that bridges HPC’s legacy with AI’s future.

Jonathon Anderson, Principal HPC Engineer at CIQ and chair of the Warewulf project’s technical steering committee, explained why the platform continues to matter. “The main thing that’s helped Warewulf stay relevant is just that it has been in the trenches, being used by the community that whole time,” he said. Each generation has rethought how provisioning works, and Warewulf 4 — launched in 2020 — leaned into container ecosystems like Docker and Podman, making node image building repeatable and automatable.

From Research Labs to AI Clusters
Cluster computing used to be the domain of national labs and academic institutions. Today, those same HPC-style environments underpin everything from autonomous vehicles to large-scale AI training. Anderson noted that AI workloads “look an awful lot like HPC at an infrastructure level,” but increasingly they’re being repatriated from cloud platforms back to local data centers. That shift has renewed interest in Warewulf as a lightweight, flexible tool for provisioning GPU-heavy clusters.

Equally transformative has been the rise of containers. By building on container runtimes — including Apptainer, the Singularity successor that CIQ helps maintain — Warewulf allows organizations to create, test, and share node images with CI/CD pipelines. “It used to be a very manual and tightly coupled process with your local environment,” Anderson said. “Now it’s something you can automate…and share throughout your infrastructure.”

What’s New in Warewulf Pro
Warewulf Pro builds on this foundation with enterprise-ready capabilities. At the top of the list is a long-requested web interface. Backed by an open REST API, it makes cluster configuration more approachable and discoverable, revealing features many users didn’t realize Warewulf supported. Pre-built node images are another cornerstone, offering ready-to-deploy stacks for OpenHPC with schedulers like Slurm and PBS, as well as CIQ’s own Fuzzball hybrid cloud platform. Configuration overlays let teams customize behavior without editing images directly, and Apptainer support comes bundled to encourage containerized HPC from day one.

Beyond software, CIQ is pairing Warewulf Pro with support and training. Enterprises can get commercial backing for existing deployments, or use CIQ’s images and overlays to simplify adoption. “Nothing would create a hard cut that would be a barrier to entry,” Anderson emphasized. Organizations can adopt gradually, while still benefiting from professional services and curriculum-based training.

Looking Ahead
Warewulf Pro is not limited to HPC. Teams are already experimenting with Kubernetes, Proxmox, and even storage clusters. CIQ plans to expand first-class support to these use cases, adding auto-discovery features, more resilient high availability, and integration with file systems like IBM’s GPFS. As Anderson put it, the philosophy is to keep Warewulf “as unopinionated as possible, so that you can take it as a tool in your toolbox and use it for things we’d never imagined.”

As HPC and AI continue to converge, that flexibility may prove decisive. Whether in research labs or enterprise AI deployments, the demands on infrastructure are only intensifying. Warewulf Pro positions CIQ to meet that demand — with an open-source foundation, enterprise polish, and a future that looks firmly toward the expanding edge of AI.