Infrastructure#

VUB-HPC is a member of the Vlaams Supercomputer Centrum (VSC), a partnership between the five Flemish universities and their university associations to provide HPC resources to the academic and industrial community in Flanders.

VSC Tier-2#

As a member of VSC, all researchers in VUB and UZ Brussel can access all Tier-2 clusters in the VSC network.

Full list of Tier-2 clusters is available in VSCdocTier-2 Hardware

Hydra#

Hydra is the local VSC Tier-2 cluster managed by VUB-HPC. The computational resources of Hydra are distributed following an exclusive policy, meaning that each computational job running in the cluster is allocated dedicated resources.

The cluster is a general purpose heterogeneous cluster that is maintained under a policy of permanent upgrades. Therefore, over the years it has gained several features:

  • 3000+ CPU cores for generic compute

  • 20+ NVIDIA data center GPUs

  • High-memory node with 1.5 TB of RAM

  • Fast InfiniBand Interconnect (EDR)

Detailed hardware information is available in VSCdocHydra Hardware

Anansi#

Anansi is a small cluster designed for interactive use and test/debug jobs. It sits next to Hydra and both clusters share the same network and storage system, which simplifies testing those jobs that will be run on Hydra.

The particularity of Anansi is that its computational resources are distributed following a non-exclusive policy. This means that resources such as CPU cores and GPUs allocated to jobs in Anansi might be shared with other jobs. The only resource that is kept exclusive is system memory.

Shared resources combined with a maximum walltime of 12 hours maximizes the availability of this cluster for quick interactive use and test/debug tasks, avoiding wait time in queue.

Technical characteristics of Anansi:

  • 32 Intel Broadwell CPU cores (128 slots for jobs)

  • 4 Nvidia GeForce 1080Ti GPUs (16 slots for jobs)

  • 512 GB of system memory

VSC Tier-1#

Tier-1 clusters offer computational resources in a bigger scale to tackle bigger problems

  • large parallel computing jobs that require a high-bandwidth low-latency interconnect

  • large arrays of jobs with a multitude of small independent tasks

  • large arrays of jobs that require the use of 1 or more GPUs

As a member of VSC, all researchers in VUB and UZ Brussel can request access to any of the Tier-1 clusters in the VSC network. Resources in Tier-1 are granted to research projects that need the large scale HPC and can efficiently use its computational resources. More information in Getting Access.

The current Tier-1 cluster, VSCdocHortense, consists of both CPU-only and GPU nodes.

Full list of Tier-1 clusters is available in VSCdocTier-1 Hardware