Forgot your password?
New user?
Document Actions
benchmarks
A short compilation of benchmarks for various WN candidates
All of the fully loaded tests were done by loading up the CPU, Network, RAM, and harddrives. All nodes except the Sun had 4GB of RAM. The Sun node had 8GB of RAM.
| Box Manufacturer | CPUS | Idle Power | Loaded Power |
| Sun Sunfire X2200 | 2 X Opteron 2218 (2.6 GHz) | 1.6 Amps | 2.1 Amps |
| Team HPC | 2 X Opteron 242 (1.6 GHz) | 1.2 Amps | 1.4 Amps |
| Team HPC | 2 X Opteron 248 (2.2 GHz) | 1.4 Amps | 1.6 Amps |
| Team HPC | 2 X Opteron 270 (2.0 GHz) | 1.2 Amps | 1.7 Amps |
| Team HPC | 2 X Opteron 275 (2.2 GHz) | 1.4 Amps | 1.8 Amps |
| Dell | 2 X Irwindale 533 (3.0 GHz) | 1.7 Amps | 2.6 Amps |
| Dell | 2 X Woodcrest 5130 (2.0 GHz) | 1.8 Amps | 2.4 Amps |
| Silicon Mechanics | 2 X Woodcrest 5130 (2.0 GHz) | 1.3 Amps | 1.7 Amps |
| Ace | 2 X Opteron 2214 (2.2 GHz) | 1.5 Amps | 2.5 Amps |
I did some CMSSW runs on several of the machines we have. I did not run more than a single process on any of them since I could not fully occupy some of them (one was running dcache--the woodcrest). The results may be tainted a little by this, but they still provide a good feel for various CPU performance of the different machines.
I ran CMSSW_1_0_0 and generated 1000 Higgs->4 mu events with tracking. The configuration file I used I copied from CMS tutorials, higgs.cfg.
| Box Maker | CPU | Elapsed time from CMSSW and time | Speed Ratio wrt red WN | CPU Clock Ratio |
| Sun Sunfire X2200 | 2x2 Opteron 2218 (2.6 GHz) | 346871u (RT: 3474 sec) | 1.17 | 1.18 |
| Team HPC | 2x2 Opteron 275 (2.2 GHz) | 397969u (RT: 4067 sec) | 1.0 | 1.0 |
| Silicon Mechanics | 2x2 Woodcrest 5130 (2.0 GHz) | 449751u (RT: 4503 sec) | .90 | .91 |
Mako also did some benchmarks using Gaussian. Those results just need to be transcribed over.