CIQ launches Fuzzball 4.0 for HPC & AI orchestration

CIQ has released Fuzzball 4.0 for HPC and AI orchestration, aimed at national laboratories, HPC centres and research teams.

The new version adds direct integration with existing parallel file systems, a built-in object cache, an integrated container registry and broader cloud support. The update is intended to reduce the specialist infrastructure work research organisations have typically needed before running workloads.

Advanced computing environments have often required teams to assemble and manage separate storage systems, registries, cloud connections and operational tools. That work has usually fallen to administrators and infrastructure specialists before researchers can begin using compute resources for modelling, analysis or AI workloads.

Fuzzball 4.0 connects to established storage environments including Lustre, GPFS and BeeGFS, which are widely used in large research computing estates. Organisations can run workloads against data where it already resides, without migrating or duplicating information or changing their storage architecture.

A generic hostpath storage driver extends that approach to other file system backends. Volumes are externally managed and imported dynamically, allowing the software to sit alongside existing infrastructure rather than requiring a replacement project.

Platform changes

The release also introduces a native object cache designed to retain ingressed data for reuse across workloads. That should reduce repeated data movement for iterative and recurring jobs, while giving users a direct staging path for workflow inputs and outputs through the command line and web interface.

An integrated container registry has also been added. This allows organisations to push container images directly into Fuzzball rather than relying on a separate external registry, which can pose an operational risk in production environments.

CIQ has also expanded deployment options across cloud providers. With the addition of Azure, Fuzzball 4.0 now supports AWS, Google Cloud, Oracle Cloud, CoreWeave and Azure, alongside on-premises clusters built with Warewulf, VMware or bare metal.

The same workflow can run on an on-premises cluster or any supported cloud without modification. CIQ has also redesigned the command-line interface and rebuilt the web interface so administrators and users can manage workflows, dependencies and service visibility in one place.

Research use

The release reflects growing pressure on research institutions to make better use of existing compute estates while also tapping cloud capacity when demand spikes. For many laboratories and research groups, one of the main barriers has been the operational burden of stitching together software layers across storage, scheduling, containers and cloud access.

CIQ, known as a support and services partner for Rocky Linux, has been building a broader portfolio around Linux infrastructure, cluster provisioning and container tools for scientific computing. Fuzzball sits within that portfolio as software intended to manage job-based orchestration for HPC and AI workloads.

Gregory Kurtzer, Chief Executive Officer and Founder of CIQ, described the new version as a shift in how research organisations can approach advanced computing infrastructure.

"Every transformative platform in computing history introduced an abstraction layer that made something previously out of reach, consumable. Beowulf democratized access to compute by replacing specialized hardware with commodity systems. Fuzzball democratizes access to outcomes by replacing specialized operational expertise with intelligent infrastructure abstraction. For the first time, organizations can treat HPC, AI training, AI inference, cloud resources and on-premises infrastructure as a single consumable capability. What we have built is the supersuit for scientific computing: the most powerful HPC and AI platform in the world, finally as easy to deploy, operate and scale as the science demands," said Gregory Kurtzer, Chief Executive Officer and Founder of CIQ.

The latest release is built to meet production requirements in national laboratories and HPC centres, where environments can span multi-petabyte storage systems and a mix of local and cloud compute resources. In those settings, software that can connect to existing systems without forcing migration or major redesign can reduce the disruption associated with platform changes.

That may be particularly relevant for institutions trying to support a wider mix of workloads, including traditional simulation and modelling jobs alongside AI training and inference. For operators, the challenge is often less about raw compute availability than about making varied resources usable through a common operational layer.

By combining storage access, caching, container image management and multi-cloud deployment in one product, CIQ is seeking a larger role in the software stack used by research computing teams. The company says organisations no longer need to stand up and integrate separate services before research teams can begin work.

Fuzzball 4.0 is available for deployment on major cloud platforms and on-premises environments, including clusters running Warewulf, VMware and bare metal.