[Bioclusters] Call for information.

[Ivo Grosse]

Hi all,

did somebody recently do a Blast benchmark of dual-P3s versus dual-P4s 
versus dual-Athlon-MPs?  For example blast human against mouse, or 
human against pufferfish.  Is it correct that, say, 1.6 GHz (1900+) 
dual-Athlon-MPs are faster than, say, 2.2 GHz dual-P4s?  How much 
faster?

We did some really extensive benchmarks last fall, and we obtained 
quite consistently the following results:

1.1 GHz dual-Tualatins (with SDRAM) were slightly faster than 1.2 GHz 
dual-Athlon-MPs (with DDR RAM) were slightly faster than dual-P4-xeons 
(with RDRAM at that time) ... perhaps because of the following reason: 
Tualatins/Athlons/P4-xeons have (or had) 512/384/256 kB L2 cache, and 
apparently cache size is (or was) crucial for Blast.

Hence, we decided against dual-P4-xeons (obvious because of the above 
and because they were really expensive last year and because they 
required RDRAM, which was really really expensive last year, 
particularly when at least 2 GB RAM per node are needed) and also 
against dual-Athlon-MPs.  The choice of Tualatin versus Athlon was 
harder, and what counted in our case was not only the slightly superior 
performance, but the simple fact that we had (and still have) extremely 
limited space.

I wonder how this comparison has changed after the P4s are now 

- produced on a smaller die, hence requiring less power and producing 
less heat,

- clocked at 2.2 GHz,

- equipped with 512 kB L2 cache?

Also, could anyone send a comparison of RDRAM versus DDR RAM (with 
respect to the performance of Blast)?

And, could anyone comment on the thermal stability 1U dual-Athlon-MP 
nodes (if, say, 40 of them are stacked in a single 40U rack and if all 
of the CPUs are running full speed pretty much 24 hours a day and 7 
days a week)?

Thanks a lot!

Ivo