New HW in MetaCenter

New HW in MetaCenter

The MetaCenter has been recently expanded with two new powerful clusters:

1) Masaryk University (CERIT-SC) added 20 additional nodes with a total of 960 CPU cores and 32x NVIDIA H100 with 94 GB of GPU RAM suitable for AI-intensive computing.

2) The Institute of Physics of the Academy of Sciences added a new cluster magma.fzu.cz consisting of 23 nodes with 2208 CPU cores and 1.5 TB RAM each

 

 Configuration and access 

1) Cluster bee.cerit-sc.cz

There are 10 nodes involved in the MetaCenter batch system, with a total of 960 CPU cores and 20x NVIDIA H100, with the following configuration of each node:

CPU 2x AMD EPYC 9454 48-Core Processor
RAM 1536 GiB
GPU 2x H100 s 94 GB GPU RAM
disk 8x 7TB SSD with BeeGFS support
net Ethernet 100Gbit/s, InfiniBand 200Gbit/s
note

Performance of each node is according to SPECrate 2017_fp_base = 1060

owner CERIT-SC

The cluster supports NVidia GPU Cloud (NGC) tools for deep learning, including pre-configured environments, and is accessible in regular gpu queues.

We are also preparing a change in access the DGX H100 machine, which will remain in a dedicated queue gpu_dgx@meta-pbs.metacentrum.cz. It will be usable on demand and only by users who can prove that their jobs support NVLink and are able to use at least 4 or all 8 GPU cards at once. We will keep you posted on the upcoming change.

 


2) Cluster magma.fzu.cz

There are new 23 nodes involved in the MetaCenter batch system, with a total of 2208 CPU cores with the following configuration for each node:

CPU 2x AMD EPYC 9454 48-Core Processor CPU @ 2.7GHz
RAM 1536 GiBidia
disk 1x 3.84 NVMe
net Ethernet 10Gbit/s
note

The performance of each node is according to SPECrate 2017_fp_base = 1160

owner FZÚ AV ČR

The cluster is accessible in the priority queue of the owner luna@pbs-m1.metacentrum.cz and for other users in short regular queues.
 

Complete list of the available HW: http://metavo.metacentrum.cz/pbsmon2/hardware.

 


Ivana Křenková, 18. 11. 2024