Petascale powerhouse cracks important HIV code
Cray's Blue Waters simulates 64 million atoms
The Blue Waters petascale computer at the University of Illinois' National Centre for Supercomputing Applications is being credited with cracking part of the code of HIV – and possibly helping point the way to new treatments.
Simulations carried out on Blue Waters allowed researchers to determine the precise structure of the HIV capsid – the protein shell protecting the virus, which is an important part of its ability to attack the human immune system.
A sample of 64 million atoms was analysed, and according to the researchers, that's what required the petaflop capability of the supercomputer. What scientists were looking for was a detailed atomic-level analysis of the 1,300 identical proteins that form a cone-like structure in the capsid. Previously, attempts to analyse the capsid couldn't capture the whole structure in one simulation.
Preparing the ground for the experiment, the researchers, led by Peijun Zhang of the University of Pittsburgh had first conducted various physical experiments including cryo-electron tomography to slice the proteins up and gain a broad understanding of their shape. They then used Blue Waters to conduct a simulation, run by University of Illinois physics professor Klaus Schulten and postdoctoral researcher Juan Perilla, called “molecular dynamic flexible fitting”.
Schulten says the capsid is “one of the biggest structures ever solved”, with further explanation in the video below.
Blue Waters was originally going to be constructed by IBM, but Big Blue withdrew in 2011, realising it couldn't put the machine together on budget. Cray picked up the build, assembling XE6 Opteron blade server nodes and XK6 CPU-GPU nodes, linked with Cray's Gemini XE interconnect.
The 237 XE cabinets house more than 22,000 compute nodes, while the 32 XK cabinets are home to 3,072 compute nodes. There's 26.4 PB of storage with an aggregate I/O of more than a terabit per second. ®
Sponsored: Benefits from the lessons learned in HPC