We do alot of turbomachinery CFD analysis, mainly for centrifugal compressors, using Tascflow and Star-CD. Mostly we use Dec Alpha's for this work and we are planning to get some new 833 MHz Alpha's in the near future. However, I have seen performance numbers for the newest PC's (P4 1.5 GHz) which seem to rival the Alpha performance even in numerically intense applications. I'm looking at the SpecCFP numbers here (www.spec.org/osg/cpu2000/results/cfp2000.html). The PC's cost less than half what the Alpha's cost (single CPU), so if the performance numbers are true, we should really change over to PC's. Does anyone have any comments on PC performance vs. Alpha's (or other workstations) for CFD apps? Can anyone out there confirm or refute what I have said? Any comments, general or specific, would be appreciated.

I've been researching this area for the last few weeks to determine which new computer to get. Here are my general thoughts. The Alpha manufacturer (and others) have focused much of their efforts on making one very fast processor. Intel and AMD have focused on mass producing processors. Years ago, comparing a single processor Intel with an Alpha system gave a very favorable price/performance ratio to the Alphas. However, now that Intel and AMD can mass produce the processors AND now that these can be made into multiple processor workstations and servers, the Alpha benefit has eroded. Compare the price/performance of an 8 processor P3 Xeon with any Alpha and you will see my point.

Intel won't say for sure, but the street estimate is that multi-processor P4's will be available in June/July 2001. Buy a computer then, I bet a dual processor P4 at 2.0 GHz will meet most peoples needs. As for AMD, the rumor has it that a multi-processor version might come out on May 15th.

Intel's is making a great effort for very inexpensive 4,000 processor servers. I hope that CFD codes will be written to make efficient use of this high number of processors. Who knows if Alpha machines will be able to catch up?