CSIRO lights up cluster, drops in GPUs

This is being reported in a few places, but I found mine at NetworkWorld.com. We write about the Commonwealth Scientific and Industrial Research Organisation (CSIRO) in Canberra from time to time. This time the occasion is the launching of their new system, a 61,440 core system (this count includes GPU cores, which we haven’t exactly agreed should be counted in the core count, but whatever)

CSIRO superCSIRO Computational and Simulation Science leader, Dr John Taylor, said the Nvidia-based GPU technology gives the organisation the potential to deliver huge gains in computational efficiency, energy efficiency and scientific research.

…Under the hood:

  • 128 Intel Dual Xeon E5462 compute nodes (a total of 1024 2.8GHz compute cores) with 16GB or 32GB of RAM, 500GB SATA storage and DDR InfiniBand interconnect
  • 64 Tesla S1070 (200 GPUs with a total of 48,000 streaming processor cores)
  • 128 port DDR InfiniBand Switch expandable to 144 ports
  • 80 Terabyte Hitachi NAS file system

It looks like they are approaching the project with a good bit of realism though, which is good

A major barrier to be overcome, however, is moving the CSIRO scientists’ code over onto the new GPU cluster, and having the approach to coding changed.

“The key issue at the moment is you can’t take an application that’s running on a multi-CPU core system and run it on the GPU,” Taylor said. “You’ve got to change the code, which is very typical.”

Resource Links: