Here we present the results of our benchmarking of Wilkes3. The benchmarks we are running are our HECBioSim benchmark suite here. Our benchmarks aren't too heavily tuned as these are designed to be used by our community to work out how much HPC time to ask for on this resource, so are set at a level that would be reasonable for any normal biomolecular MD simulation. More information on the benchmarks themselves can be found at the above link.
Wilkes3 is one of the UK Tier2 facilities based at Cambridge and is comprised of 80 main GPU nodes each of which is equipped with two 64 core AMD EPYC 7763 CPUs with 1000GB of RAM. Each node has four NVidia A100 80GB GPUs and is connected to a high performance Lustre parallel filesystem, internode communication is handled by a high speed and low latency infiniband interconnect.
AMBER 20 - single node, single GPU
20k atom - 430 ns/day
61k atom - 288 ns/day
465k atom - 63 ns/day
1400k atom - 19 ns/day
3000k atom - 9.4 ns/day
Gromacs 2022.2 - single node, single GPU
20k atom - 533 ns/day
61k atom - 210 ns/day
465k atom - 37.5 ns/day
1400k atom - 12.2 ns/day
3000k atom - 6.2 ns/day
NAMD 3-alpha12 - single node, single GPU - do note, this is an alpha release of 3.0
20k atom - 168 ns/day
61k atom - 158 ns/day
465k atom - 26.0 ns/day
1400k atom - 8.6 ns/day
3000k atom - 3.7 ns/day
In our initial benchmarking run we focused on single node with single GPUs on Bede. We will be adding more stable codes and starting to look at how to get the best out of single node multiple GPU and also multinode GPU simulations soon.