Compute

Hardware Overview

The HPC cluster is connected to the university network over 10 Gbps Ethernet (GbE) for high-speed data transfer and provides access to over 1,024 CPU nodes, 50,000 CPU cores, and over 200 GPUs. Compute nodes are wired with 10 GbE or a high-performance HDR200 InfiniBand (IB) interconnect running at 200 Gbps (with some nodes running HDR100 IB if the HDR200 IB is not supported).

If you would like to purchase hardware, please schedule a Consultation with the RC team first.

CPU Nodes

The table below shows the feature names, number of nodes by partition type (public and private), and the RAM memory range per node. The feature name follows archspec microarchitecture specification. If you are interested in more information about the different partitions on the cluster, including the number of nodes per partition, running time limits, job submission limits, and RAM limits, see Partitions.

Feature Name	Number of Nodes (Public, Private)
skylake	0, 170
zen2	40, 292
zen	40, 300
ivybridge	64, 130
sandybridge	8, 0
haswell	230, 62
broadwell	756, 226
cascadelake	260, 88

GPU Nodes

The table below shows the GPU types, architecture, memory and other features of the GPUs on the HPC cluster. For more information about GPUs, see Working with GPUs.

GPU Type	Public Nodes (x # GPUs)	Private Nodes (x # GPUs)
V100 PCle	4(x2)	1(x2), 16GB
V100 SXM2	24(x4)	10(x4), 16GB 8(x4), 32GB
T4	2(x3-4)	1(x4)
A100	3(x4)	15(x2-8)
Quadro RTX 8000	0	2(x3)
A30	0	1(x3)
RTX A5000	0	6(x8)
RTX A6000	0	3(x8)