On today’s episode of “The Interview” with The Next Platform, we take a look at the evolution of the NAMD molecular dynamics and how the introduction of GPU computing upended performance expectations and set the stage for new metrics now that the Volta GPU architecture will be available on large supercomputers like the Summit machine coming to Oak Ridge National Lab.
Our guest is well known for being part of the team that won a Gordon Bell Prize in 2002 for work on scaling NAMD. Dr. Jim Phillips is a Senior Research Programmer in the NCSA Blue Waters Project Office at the University of Illinois at Urbana-Champaign and has been the lead developer of the highly scalable parallel NAMD code since 1999.
In the audio interview in the player below, we take a look at the evolution of the code itself across many architectures in HPC over the years, leading to the present where GPUs dominate to speed the code. Much of what Phillips describes in the first part of the podcast is more thoroughly detailed in this lengthy story of NAMD and the systems it exploited on its journey from small project to well-known molecular dynamics tool.
Aside from presenting a background on the early code and systems, we talk about choosing the right architecture for a complex HPC code like NAMD, including a discussion on where other approaches fit, including the use of FPGAs for molecular dynamics and of course, where the custom supercomputer from D.E. Shaw “Anton” fits in the MD ecosystem.
“The highly parallel molecular dynamics code NAMD has been long used on the GPU-accelerated Cray XK7 Blue Waters and ORNL Titan machines to perform petascale biomolecular simulations, including a 64-million-atom model of the HIV virus capsid. In 2007 NAMD was was one of the first codes to run on a GPU cluster, and it is now one of the first on the new ORNL Summit supercomputer, which features IBM POWER9 CPUs, NVIDIA Volta GPUs, and the NVLink CPU-GPU interconnect.”
In terms of the latest GPUs and how they scale and perform on NAMD, Phillips is looking ahead to early results on Summit with Volta GPUs, Power 9 host CPUs and the NVlink interconnect. The first thousand or so nodes are on the floor and running for a select set of HPC applications with their team securing 64 nodes for early testing. Eventually, the NAMD team wants to take 20% or so of the machine for their work, but for now, they are working out what few driver and compiler issues they have hit along the way. The team expects good performance but of course, that is less about the GPU and more about the CPU memory and bandwidth and how much noise is on the system when it runs at scale.
For those planning on going to the GPU Technology Conference in late March, be sure to catch the talk “ORNL Summit: Petascale Molecular Dynamics Simulations on the Summit Power9/Colta Supercomputer” with Dr. Phillips—more info on that here.