Hello, for running my CFD codes I am about to bye a new computer and I am wondering, can I use the memory of my old computer (512 MB RAM-400 MHz) to my new one (1 GB- 667 MHz)? Is the difference in speed of these RAMs going to decrease the overral speed (if I was using only the 1 GB RAM)? My old computer is a Pentium 4- 2.4 MHz and the new one a Pentium dual core 2- 2.4 MHz. Also how many times (approximately) is the second faster than the first? According to my calculations it will be about 4 times faster, but I would appreciate your opinion!

Thank you in advance.

George

PS: I don't trust the benchmarks of the various sites-companies and that is why I need the opinion of CFD users.

Are you sure that both machines use the same type of memory module? My guess is that the P-4 would have been using DDR1, and the dual-core machine would need DDR2. The slots for these two types of memory are not even the same, you physically wouldn't be able to fit the modules into the slots on the motherboard. They are just completely incompatible. You might get a factor 4 speedup if your code can run parallel, although I suspect that is being just a bit optimistic.

Thanks for your response.

Yes, P-4 has DDR1 slots. But aren't there motherboards that have slots for both memories (DDR1,DDR2)?

My code can't run on parallel. Has Pentium dual core 2 hyper-threading technology (meaning to divide the processor into two sub-processors) and be able to take advantage only the one sub-processor?

I remember reading of one board that could take either type (may have been for AMD though?), but not at the same time. If your code can't run in parallel you won't benefit much from the dual core, which acts like two separate CPU's, and I doubt that you will see much better than 50% improvement.

Are you sure about the factor 50%? Because it must be strange from 2002 (when P-4 2.4 MHz was released) until 2007 the CPU technology to have improved only by this factor. Even for not parallel use...

At work there is a P-4 3.2 MHz hyper-threading computer (bought in 2005) and compared with the P-4 2.4 MHz, the first is 10-15% faster, even if the CFD code takes advantage only the 50% of the CPU.

Waiting for your reply.

No, of course I'm not sure. That's why you test these things. The Core 2 CPU does much more work per clock cycle than the P-4, which was a very different chip design. According to SPEC CPU2000, the 2.4 P4 scores about 1092, compared to 2497 for a Core 2 Duo E6600. So that is a massive advantage (more than I expected, to be sure!) to the newer CPU. Looking more closely at the individual components of the benchmark, the "CFD" part is a code called galgel. Here the P4 returns a runtime of 157, vs. 50.5 for the Core 2 Duo. So on the face of it, it looks like you could get your three times speed-up. Just bear in mind that these benchmark times are done with codes that are heavily optimised to make the best use of SSE type instructions, and if your code isn't, you may not see the same performance trends.

I'm also not sure that these figures quite stack up when you look at SpecFPRate. This would indicate a score of about 10.8 for a single 2.4 P4, vs. 44 for dual-core E6600. So this is a 4:1 advantage for the new chip with parallel processing, which would suggest a 2:1 advantage without parallel, compared to the 3:1 from the standard benchmark.

SPEC is artificial, of course, which is why you should test your own code. Having said all that, this is not something worth agonizing over. The E-6600 CPU is an amazing bargain. Just get it.

Thank you for your quick response and for all the information that you provided me. When I test the new CPU I will let you know.

Also, a suggestion for the forum: It would be very helpful for all who want to purchase a new computer, if anyone has information about the speed-up of the new processors to make his or her suggestions- conclusions. As Charles said, no benchmark can be more reliable than people who test it on their computers and their codes (commercial or not)

"I doubt that you will see much better than 50% improvement."I do not agree with you unless you use the parallelel program.

A side note:

If you do mix different memory (assuming both are of the same type, i.e. DDR2) you will limit the system to the slower memory bounds. This in turn might force you to use unwanted dividers to be able to use the full potential of the CPU. I would not recommend it.

As for the benchmarks. It is really difficult to compare the two systems since one is x86 and the other x86_64. Obviously you would want to use the 64 bit capability with your new processor and this includes using a 64 bit OS. This means that you will not really compare the two CPUs against each other. The choice of simulation size will also affect the results since the E6600 has a much larger cache. You can of course just ignore the above and see what speedup you get compared to your previous setup.

Finally, IMHO, the E6600 is one of the most cost efficient CPUs today, especially considering it's overclocking potential.