dgx server imageHere we present the results of our benchmarking of JADE2. 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.

JADE2 is the second iteration of the UK Tier2 service JADE and is hosted by STFC on behalf of Oxford University. JADE2 comprises of 63 nodes of the NVidia DGX1-MaxQ where each node contains two Intel 20 core E5-2698v4 (broadwell) processors and eight NVidia V100 32GB SXM2 GPUs. Each node has onboard NVLink for high speed inter GPU communications and has a fast infiniband connection to storage. Each node has access to high speed Lustre parallel filesystem.


AMBER18 - single node, single GPU
20k atom - 355 ns/day
61k atom - 202 ns/day
465k atom - 27 ns/day
1400k atom - 8.42 ns/day
3000k atom - 3.98 ns/day

AMBER 20 - single node, single GPU
20k atom - 368 ns/day
61k atom - 211 ns/day
465k atom - 30 ns/day
1400k atom - 9 ns/day
3000k atom - 4.1 ns/day

Gromacs 2020.3 - single node, single GPU
20k atom - 372 ns/day                     
61k atom - 132 ns/day                     
465k atom - 19.1 ns/day                   
1400k atom - 6.2 ns/day                  
3000k atom - 3.2 ns/day

Gromacs 2020.4 - with plumed 2.6.2 - single node, single GPU
20k atom - 372 ns/day
61k atom - 136 ns/day
465k atom - 19 ns/day
1400k atom - 6.2 ns/day
3000k atom - 3.2 ns/day

NAMD 2.14
20k atom - 48.3 ns/day
61k atom - 20 ns/day
465k atom - 2.26 ns/day
1400k atom - 0.76 ns/day
3000k atom - 0.4 ns/day

NAMD 3-alpha7 - single node, single GPU - do note, this is currently an early alpha release of 3.0
20k atom - 158 ns/day
61k atom - 112 ns/day
465k atom - 12 ns/day
1400k atom - 4 ns/day
3000k atom - 1.8 ns/day


In our initial benchmarking run we focused on single node with single GPUs on JADE2. We will be added more stable codes and starting to look at how to get the best out of single node multiple GPU soon.