Lowered My WU Time by 10 to 12 minutes Per Unit

[llubon]

Hello

I have been playing around with OC'ing my machine and tinkering with my FSB to try and find the SWEET spot for crunching... here what I found.

The System: XP2100, Epox 8K3A+, 512 of OCZ pc3000, Innovatek watercooling, Ennermax 650 PS.

1st Set-up: Xp2100@2030mghz 184x11 Vcor 2.0 & Dimm 3.2
Results: 416Mflops Duration: 2.48hrs

2nd Set-up: XP2100@1896mghz 145x13 Vcor 1.85 & Dimm 2.8
Results: 358Mflops Duration: 2.36hrs

Now it seems to me that set-up number (2) would be more productive than set-up number (1) Even though I truned less Mflops my times decreased by 11min.

Now your probley thinking "HOW THE HELL" is that possible, As I watched through both WU's I noticed in the 1st set-up my Mflops started very high then dropped down around 25% and ran level till around 80% and steadily went up to completion.

On the 2nd set-up I ran a more consistant level through out the whole WU. Stayed around 350Mflops most of the time.

Can someone please chime in on this and give me some of your wisdom or conjecture ???

Thx llubon

 

[LandShark]

the one and only thing to tell the difference between hardware/performance is to use the same WU (like our benchie WU). 'cos every WU doesn't crunch the same! and it could vary a LOT!!

on my P4, a 0.417AR could be done in 'bout 2:40ish, a 0.0xxAR needs 3:0x mins, a 3.xxx+AR can finish in 2:10mins!!

 

[llubon]

Thanks LandShark... You are correct I did not use the same WU's.
Damm I thought I was on to somthing thier !!!! I guess I need to run bench (.417) and see what I find, but what you just said make perfect sence... I let you know what the out come is...

My thought process was that a higher multi'er -V- higher FSB would yeild lower times.

Thx for the input. llubon

 

[LandShark]

i'll put my $$ on high FSB setting

i'm in planning to do something like that in the future when i have the hardware. i'm planning to do a P4 test to find out is RDRAM really helps or DDR still hold or after all GHz win.

like a 1.8a clocked at 2.5G, 2 mobo. one RDRAM, one DDR. now that we could see which is better and by how much 'cos we would rule out the GHz part. then try to o/c the CPU as high as possible but most like wil be lowering memory speed (either x3 in RDRAM or slower timing in DDR), and compare again.

it should be interesting. (if/when i have time...... )

btw, Welcome come to the forum & the team!!

 

[macklin01]

Hmm,
I'd have to wonder if the differing memory speeds (due to FSB) or the differing WU's caused the change in results. Honestly, as I've seen that the console-based version only takes like 8,000 KB - 9,000 KB tops (on a 613-atom protein on my machine, for example), I would doubt that memory is as big of a player, so I would lean towards WU variation.

I seem to remember reading at the Stanford F@H site (I just joined yesterday, so it's fresh in memory ) that right now, faster CPU speeds translate to faster calculations, although in the future, more complicated proteins could tilt that more towards memory speeds.

Great idea for a comparison, and welcome to teh forums! -- Paul

 

[LandShark]

well, don't know 'bout Folding. but on SETI, memory is the KEY (King). and different WU WILL also have different finishing time simply b/c some WU requirer less work to be done than others.

 

[macklin01]

Just curious -- how much memory does the SETI application take when it's up and moving?

This is very interesting -- it says that SETI is a good memory benchmark, and Folding is a good CPU benchmark.

-- Paul