Processor benchmarking

Our first question was what type of processors to use. At the time there were three options: a Pentium 3 (P3), Pentium 4 (P4), and AMD Athlon. The P3 and AMD could be bought in a dual configuration for a reasonable price, while the P4 dual Xeon was prohibitively expensive. The cost of disk and memory along with space considerations made a dual system an attractive option. To test which configuration was the best for our current needs we built a single node of the P4 and AMD and compared it to our current P2 and P3 machines. For the P4 we used a 1.7 Ghz processor and Rambus memory. For the AMD we used a 1.2 GHz processor and DDR memory. We built a dual AMD (1.2 GHz) node and compared it to our current P2 and P3 configurations. Figure 1 shows the comparison of running a small wave equation migration program on each configuration. Assuming a little less than linear speed up with clock speed for the P3, we estimated that we would get approximately a 10% speed up with a 1.2 Ghz P3.

 

speed Figure 1 Benchmarking wave equation migration on quad Pentium 2, dual Pentium 3, Pentium 4, and dual AMD. The single `*' is for the Pentium 4.

The end result of this initial testing was that we felt that the high price of Rambus memory did not justify the speed improvement, therefore, we abandoned that option. The AMD and P3 were approximately the same cost (the P3 processor was more expensive but DDR memory was more expensive than SDRAM). We were leaning more towards the AMD build with the assumption that DDR memory prices would decrease as it became more standard when Intel made an announcement that they were going to release a new compiler for linux.