Wilson Cluster-Institutional Cluster

CPU-only worker features

partition(s)	node count(s)	processor	cores / threads	memory [GB]	mem / core [GB]
wc_cpu / wc_test	90 / 10	dual Xeon E5-2650 v2 2.6 GHz	16 / 16	128	8

GPU-equipped worker features

The workers below share the wc_gpu slurm partition. The gres specification is used to request a particular worker node type.

node count	slurm gres	GPU type	GPU count	processors	cores / threads	host memory [GB]
2	gpu:a100	NVIDIA A100	2	dual AMD EPYC 7543 2.8 GHz	64 / 64	512
1	gpu:a100	NVIDIA A100	4	dual AMD EPYC 7543 2.8 GHz	64 / 64	512
4	gpu:v100	NVIDIA V100	2	dual Xeon Gold 6248 2.5 GHz	40 / 40	192
1	gpu:p100nvlink	NVIDIA P100	2	dual Xeon E5-2680 2.4GHz	28 / 28	128
1	gpu:p100	NVIDIA P100	8	dual Xeon E5-2609 v4 1.7GHz	16 / 16	768

IBM Power9 (ppc64le) GPU worker

The IBM Power9 worker ppc64le architecture is not binary compatible with the widely used x86_64 architecture of AMD and Intel processors, hence, this node is in a separate slurm partition named wc_gpu_ppc. GPUs are requested using --gres=gpu:v100nvlinkppc64:n where n is the number of GPUs needed. This node’s design is similar to the worker nodes in the OLCF Summit supercomputer.

node count	GPU type	GPU count	processors	cores / threads	host memory [TB]
1	NVIDIA V100	4	dual IBM Power9 3.8 GHz	32 / 128	1.1