The CoSTAR Cluster

This section covers the core technical infrastructure of the CoSTAR cluster and how to use it. If you are new to High-Performance Computing (HPC), it helps to think of the system in terms of the “3 Ss”.

• Scheduler sends your jobs to the right hardware.

• Software provides the tools and environments you need.

• Storage keeps your data available and secure.

Understanding how these three components work together is the key to running successful jobs.

Select a component below to read the detailed guides and configurations:

Scheduler

Learn how Slurm allocates compute nodes and GPUs. Find guides on submitting batch scripts and managing your queue.

Job Scheduler

Software

Find out about recommended software workflows. Learn to build containers, manage virtual environments, etc.

Software

Storage

Understand the high performance file systems. Check your quotas and learn where to store your code versus your massive datasets.

Storage Options

Hardware Specifications

The CoSTAR cluster is a heterogeneous environment featuring 10 standard compute nodes, three pilot nodes, and 100 state-of-the-art NVIDIA H200 GPUs.

Compute Nodes

Nodes	Processor	RAM	GPUs (per node)
10 × Standard Nodes costar04 – costar13	2× AMD EPYC 9535 (128 cores total)	3 TB	8× H200 (141 GB HBM3e)
2 × Intel Pilot Nodes costar01 – costar02	2× Intel Xeon 8480+ (112 cores total)	2 TB	8× H200 (141 GB HBM3e)
1 × AMD Pilot Node costar03 [*]	2× AMD EPYC (64 cores total)	1.1 TB	4× H200 (141 GB HBM3e)

[*]

costar03 is currently unavailable due to a faulty GPU. When it returns to the production pool, it will offer additional MIG (Multi-Instance GPU) slices (e.g., 35GB and 71GB instances) to support smaller workloads.

Available Partitions

The CoSTAR cluster provides several partitions (queues) for different types of workloads:

Partition	Resources	Max Walltime	Description
main	4 nodes, 32 GPUs	4 days	Default partition for general long-running workloads.
risk	8 nodes, 64 GPUs	4 days	Larger resource pool but preemptible (jobs may be stopped and re-queued if higher-priority work needs the resources). Only use this if your workflow implements checkpointing and can tolerate interruption.
debug	1 node, 8 GPUs	4 hours	For short test and debugging jobs you can use this partition. It is good practice to validate job scripts before running longer running jobs.
mig	1 node, 24 MIG instances	4 days	For smaller GPU workloads consider using MIG (Multi-Instance GPU) slices rather than full GPUs. For example a 35GB or 71GB MIG instance can be used for inference workloads which would otherwise waste resources on a full 141GB GPU.

If you are unsure which partition is best for your workload, please contact the CoSTAR Support Team.

[†]

Industry Users should use the partition name provided during the onboarding process. If you are unsure, please contact the CoSTAR Support Team.

Cluster Connectivity & Storage

Login Node	costar-login01.surrey.ac.uk
Operating System	Rocky Linux 9.6
Interconnect Fabric	Mellanox InfiniBand NDR (400 Gb/s)
Parallel Storage	750 TB WekaFS (NVMe-backed)
Total Production Cores	1,568
Total Production GPUs	100× NVIDIA H200 (141 GB HBM3e)