Author: Michel Langeveld
Date: 22:40:01 01/14/04
Go up one level in this thread
Translation of Text to English: No rights can be taken from this Translation :-) Vincent Diepeveen the brain behind the chessprogram Diep. NUMA-systems, running on Itaniums, Opterons, Xeons of SGI MIPS-processors, all did run Vincent's program. DIEP is a extreme complex and calculating intensive application that is based completely on integer instructions. The program is only 550Kb in size and strongly improved since the first benchmark tests of Ace's Hardware. Although the benchmark is dependend on a good memory-subsystem, calculation is the main issue here. In our opinion this benchmark is a welcome addition to our benchmarksuite, because branch predition is important for the next generation of software that will be based on advanced AI-techniques. DIEP is tested with a depth of 13 ply. A friend gave us the possibility to run the benchmark on the SGI Origin 3800 supercomputer. One Origin 3800 rack contains 128 SGI MIPS R14000K processors with 8MB L2 cache and a processorspeed of 500Mhz. This configuration gave a performance of 1681Knps. We have to admit the Origin 3800 is a bit older system, but the results illustrate that the Opteron is a well balanced HPC- and server processor. The processor has no weak points, beside the theoretical low SSE2 FLOPS peak performance what is caused by the low clockfrequency of the Opteron. The dual Itanium 2 1,3Ghz 3M brings DIEP to performance of 180Knps. We are quite sure that even a quad Itanium 2 1.5Ghz is not enough to outperform the quad Opteron 848 in this benchmark. We can conclude that the Opteron really shines with this kind of AI (Artificial Intelligence) workloads. The scalibility is also really good. Doubling processors results in 90% better results. Cheers, Michel
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.