Tweaking for GIMPS

[magnav0x]

I was checking the "Per iteration time:" line in the client. Is a higher or lower time better here? I made some changes to memory timing, and my Per iteration time appeared to go up by .020. I thought at first a higher number would be better, but now that I think of it, I'm thinking a lower iteration time would be better.

P4 1.6 (mobile processor) at stock speed is pulling a Per iteration time of 0.040 sec

AthalonXP 1800 (clocked at 1650 mhz, DDR Ram fsb clocked at 300mhz and all memory latency's set as low as possible) is pulling a Per iteration ime of 0.065 sec

[magnav0x]

Well after messing with cpu/memory fsb's, memory latencies, and multiplier on my Athalon XP, I got it about as quick as I could with gimps. I tryed raising the memory FSB to 333mhz, but I was unable to use lower memory latencies (unstable). I finaly settled with the following configuration.

AthalonXP 1800 (clocked at 1725 mhz, multiplier of 11.5, CPU FSB of 150 , DDR Ram FSB at 300mhz and all memory latency's set as low as possible) is pulling a Per iteration ime of 0.059 sec

The same configuration above using default latencies yielded a Per iteration time of 0.072 sec

I tryed various multipliers and cpu FSB's to get my memory stable with the lowest possible latencies. It appears, tweaking the memory yields much better results than the actual overclocking of the CPU. Just a thought that may help others. Share your results!

[outlnderFDC]

There are many things you need to watch when pushing your computer to it's fastest for Gimps.

Memory settings should be as fast as possible, as lomg as they pass the Prime95 stress test for 12 hours. Many results are failing due to too high overclocked machines. I have personally over 40 P90 CPU years because of bad chipset to memory errors.

Motherboard chipsets are a vital part of the Prime95 equation. Sis chipsets for P4s tend to be error prone. The internal error-checking of Prime95 will not deteck these errors, only when 2 others have matching residues and yours is wrong will you discover this. The CPU years will be deducted from you total.

Even though asynchronous FSB/memory timings are the nowm, they may be a cause of result failures. Some individuals have had problems with bad results with asynchronous timings but get good results when the timings are set synchronous. Just think, you may be losing 2, 3 or even 4 weeks worth of work when trying to get iteration timings down a couple of miliseconds. Is it worth it??