I need to assemble a workstation to be used with fluent in vehicle external aerodynamics. I have P4 1.5 256 Mb RAM at 600 $ for processor, board and RAM against dual P3 800EB on Gigabyte dual board with 256 Mb RAM at 500$ for processor, board and rams which to buy based on a price/preformance analysis..?

The two systems should give you similar performance running Fluent (although you need to run in parallel on the PIII - extra licence and extra work partitioning etc). The P4 runs fluent very well - on other applications the P4 will not be much faster than a single CPU PIII. Make sure the P4 has dual-channel (2*128) PC800 RDRAM though, that is a must to get good performance with it. The 1.8 GHz chip has come down quite a bit in price recently btw.

I'd get the P4.

Thanks for your help. Here is another question What about Athlon 1.4 or 1.2? Thanks

Do not use an AMD to run Fluent. The performance is rather poor if you compare it with a P4 especially for large cases. The RDRAM has the advantage of very fast communication between the processor and the memory.

Eric

Here is what I emailed directly to Danial yesterday, I think it could be of some help to others on this board:

"Hello Danial,

Sorry if this is short or doesn't cover all the topics you mentioned, today is a difficult day for us all. I never mind email, so please respond if you need more info.

My personal experience is with dual P2 450 MHz, single P3 600 MHz, and dual Xeon 1.7 GHz. All 3 machines run Win NT or Win2k. I have never tried Fluent on an Athlon machine. Normally I would say the Athlons have much better performance than the MHz rating shows (a 1.4 GHz Athlon is roughly the speed of a 1.8 GHz P4 for most programs). However Fluent seems to really like the P4 compared to the Athlons (according to the cfd-online creator). Thus I would make a direct chip speed comparison for Fluent uses.

Dual machines are at least as fast as a single processor machine. The speed boost depends on the size of the Fluent case. There are two steps: (1) calculating the governing equations for half of the model with each processor then (2) sending information between processors. In small Fluent cases, each iteration is fast and you would see a negligible difference between a single P3 800 MHz or a dual P3 800 MHz. This is since step (1) becomes small and step (2) is rate limiting. As you work with larger/more complicated models, each iteration takes more time. Thus step (1) increases its time required, while step (2) stays relatively constant. The result is about a 50% speed boost. Eventully with very large models step (1) dominates and you frequently see a 80%-110% speed boost. (Note: you get more than a 100% boost since only one processor has to share resources to run the computer operating system).

When you move to clusters of machines, you can dramatically decrease the time required for step (1). However step (2) increases with each additional processor. In a theoretical machine with infinite processors you will get step (1) finished in an instant but step (2) will take forever to complete. Thus each additional processor gives you less in return. See the Fluent website for some good benchmarks showing the effect of 4, 8, 16, or 64 processors.

Conclusion: a) For small models I'd rank the speed in this order (from slow to fast): single 800 MHz P3, dual 800 MHz P3, single 1.2 GHz Athlon, single 1.4 GHz Athlon, single 1.5 GHz P4, dual 1.5 GHz P4. However the prices vary drastically so here is my opinion of price/performance ratio (from worst to best): dual 800 MHz P3, dual 1.5 GHz P4, single 800 MHz P3, single 1.2 GHz Athlon, single 1.5 GHz P4, single 1.4 GHz Athlon. b) For large models I'd alter these rankings to this speed order: single 800 MHz P3, single 1.2 GHz Athlon, single 1.4 GHz Athlon, dual 800 MHz P3, single 1.5 GHz P4, dual 1.5 GHz P4. Here is my opinion of price/performance ratio: single 800 MHz P3, dual 800 MHz P3, single 1.2 GHz Athlon, single 1.5 GHz P4, single 1.4 GHz Athlon, dual 1.5 GHz P4.

Other answers and thoughts: 1) I don't know the speed difference with various dual boards including the Gigaboard, but usually these motherboards only affect speed by a few percent. 2) If you use small models and small problems you should go with the single 1.4 GHz Athlon. 3) If you use large models, 256 MB is far too small. You should look at a minimum of 768 MB. If you are on a limited budget this will alter things dramatically. 4) The NT version of Fluent does NOT allow clusters as far as I know, you will need to contact Fluent to check. The unix versions of Fluent DO allow clusters. 5) You must specifically install the parallel Fluent version to run Fluent on more than one processor (the single processor version will have the grayed out menu options you see). I have a university license and Fluent did not originally send the parallel Fluent version to me. After 2 years they sent a CD with the parallel version, this may have been a mistake that benefits me or Fluent may have changed policies. On the Fluent CD it specifically lists 'serial + parallel' if you have the parallel version (under the Fluent version), Fluent CDs without parallel capability are blank in that location. 6) The current P4 chipset has 428 pins on the processor. Intel just released a 478 pin P4 chip (at 1.7 GHz and above) that will replace the older 428 pin chips. Thus your computer will not be easily upgraded if you buy a 1.5 GHz machine today. Soon Intel will release 478 pin versions at all speeds, I would highly suggest getting one of these if you choose to go the P4 route. There are much faster P4's available than the 1.5 GHz model (up to 2.0 GHz) which may be worth considering if you have the money. Or if you can wait a couple of months, the next generation P4 will be out which should give a good speed boost (at a premium Intel price of course). 7) One option you didn't mention was dual 1.2 GHz Athlon MP. They cost a bit more than the single Athlons but probably have the best price/performance ratio of all non-clustered machines (although a tiny bit slower than dual P4 Xeons)."