PSD IT has deployed a small High Performance Computational resource to allow our users the opportunity to evaluate GPU Computation.
The server, bangcompute.uchicago.edu, is currently configured to allow all Particle Theory users access by way of their Theory server accounts.
- Dual-Socket Quad-Core nehalem XEON processors @ 2.4GHz [16 virtual cores with HyperThreading]
- 24 GB of RAM
- FLASH drive boot disk with a 1.5TB scratch space. Your home directory will be your NFS-mounted Theory home.
- One Nvidia c1060 TESLA computation card. 240 compute cores w/ 4GB RAM
- One Nvidia GTX 260 video card for compuation. 192 compute cores w/ 896MB RAM
Software, aside from the base Ubuntu install
- CUDA libraries
- MatLab [ask about Jacket]
- Intel Fortran suite
- g77 (pending)
Requests for additional software should be addressed to Tom Indelli. Tom's office is at Hinds B-033.
Restrictions on Use
As stated above, BangCompute's main duty is as a testbed for evaluating CUDA code [Graphics Processing Unit code]. We are opening this up to general use but we may have to block out time to allow individual researchers unrestriced access while benchmarking performance.
If this becomes nescessary, we'll email all of our users and kill jobs as required.