Performance

Headline figures

Note

The Crop Diversity HPC compute cluster has a theoretical peak CPU performance of approximately 40 TFLOPs (40 trillion floating point operations per second).

This is a cumulative figure estimated via Intel’s LINPACK benchmarking tool, run per-node rather than simultaneously cross-cluster.

There is also an additional 14 TFLOPs of (double precision) GPU performance available, rising to 224 TFLOPs if only counting Tensor (Deep Learning) performance.

CPU benchmarks

The following graphs show the relative performance of the different CPU types available within the cluster, scored using CPU PassMark (as of March 2023). Higher numbers are better.

Details of which nodes use which CPUs are below:

CPU	Node name begins with
Intel Xeon Silver 4116 @2.1GHz 12C	n19-32-*
Intel Xeon E5-2680 v4 @2.4GHz 14C	n17-28-*
AMD EPYC 7252 @3.1GHz 8C	n21-16-*
Intel Xeon Silver 4216 @2.1GHz 16C	n19-24-*
Intel Xeon Gold 6132 @2.6GHz 14C	n19-28-*
Intel Xeon Gold 6242 @2.8GHz 16C	n19-64-*
AMD EPYC 7502 @2.5Ghz 32C	n21-64-*
AMD EPYC 7543 @2.8Ghz 32C	n23-64-*

Tip

The System Overview page gives a full breakdown of the nodes and their various specifications.

Network throughput

The cluster’s compute and storage nodes are linked together using network switches that support 25-gigabit speeds (over 400 times faster than the UK’s average broadband speed of 60-megabits), with dual-bonding employed across the storage nodes. Interlinks between the switches are currently limited to 100 Gbps (gigabits per second), which allows for maximum data transfer of around 12 GB/s (gigaBYTEs per second), although this is generally only seen when running very parallel tasks.