2.2.1. Hydra hardware#

The VUB Hydra cluster is an heterogeneous cluster with a mixture of nodes with varied hardware. The majority of nodes are non-GPU nodes for generic multi-purpose compute, they are distributed in partitions depending on their CPU microarchitectures and network interconnects. The cluster also contains a number of nodes with NVIDIA GPUs, which are also distributed in partitions depending on their GPU generation.

CPU-only nodes#

Slurm partition

nodes

processors per node

memory

local disk

network

broadwell_himem

1

4x 10-core INTEL E7-8891v4 (Broadwell)

1.5 TB

4 TB HDD

10 Gbps

skylake

22

2x 20-core INTEL Xeon Gold 6148 (Skylake)

192 GB

1 TB HDD

10 Gbps

skylake_mpi

32

2x 20-core INTEL Xeon Gold 6148 (Skylake)

192 GB

1 TB HDD

EDR-IB

skylake_mpi

16

2x 14-core INTEL Xeon Gold 6132 (Skylake)

192 GB

450 GB HDD

EDR-IB

zen4

20

2x 32-core AMD EPYC 9384X (Genoa-X)

384 GB

450 GB SSD

25 Gbps

GPU nodes#

Slurm partition

features

nodes

GPUs per node

GPU memory

processors per node

CPU memory

local disk

network

pascal_gpu

4

2x Nvidia Tesla P100 (Pascal)

16 GB

2x 12-core INTEL E5-2650v4 (Broadwell)

256 GB

2 TB HDD

10 Gbps

ampere_gpu

6

2x Nvidia A100 (Ampere)

40 GB

2x 16-core AMD EPYC 7282 (Zen2 - Rome)

256 GB

2 TB SSD

EDR-IB

ampere_gpu
big_local_ssd

4

2x Nvidia A100 (Ampere)

40 GB

2x 16-core AMD EPYC 7282 (Zen2 - Rome)

256 GB

5.9 TB SSD

EDR-IB

Login nodes#

  • nodes: 2 (fair share between all users)

  • processors per node: 2x 12-core INTEL Xeon Gold 6126 (Skylake)

  • memory: 96GB (maximum per user: 12GB)

  • 10GbE network connection

  • Infiniband EDR connection to the storage