Comparison CPUs' With Same WU

mas92264 · Post by **mas92264** » Mon Dec 08, 2003 6:26 pm

By chance, I had WU #683 running on 5 machines this weekend. The results are as follows

CPU	  Ghz	  Ram	  Time	 Ratio

1600P	1.40	512.00	18.5	1.09
2000P	1.67	512.00	17.0	1.00
2500B	1.83	512.00	12.7	0.75
2600T	2.13	512.00	11.8	0.69
2800B	2.17	1024.00	10.4	0.61

Where P=Palomino, B=Barton and T=Thoroughbred. The 2800B is at 167*13. Time is minutes per "frame" in the log file.

Edit:

What's interesting is that the 2000 Palomino is way slower than the Barton/T'Bred and that while the 2800 clock is only slightly higher than the 2600 (about 2%,) the production is about 13% higher.

I double checked the production (time) #'s and re-booted the Palomino 2000 and re checked its time - was the same. All computers running 4.00 PRE 1 or 2, with forceSSE switch on.

The 2500, 2600 and 2800 are all 166 fsb - 1600 and 2000 are 133fsb. All mobo's are via chipset except 2800 which is nForce2 (Shuttle SN45G.) The only other potential glitch is that the 2000P is using on board graphics, all others have agp cards.

I think this pretty much refutes the "only clock speed matters" concept.

FWIW. (with thanks to Zyzzyx)

Edit: The 2600T should be 2.08 ghz (roughly 166*12.5)

DryFire · Post by **DryFire** » Mon Dec 08, 2003 7:24 pm

hmmm, time to overclock my 2500+ barton a tad bit more, thank god it's winter time.

Zyzzyx · Post by **Zyzzyx** » Mon Dec 08, 2003 7:32 pm

Just a curious thing for ya to check in the log file:

I thought I remembered reading that even same named WUs can have a different 'build', possibly affecting compute time.

I too have a 683 on my secondary system right now. Looking at the log file at the beginning of the WU it says:

Project: 683 (Run 53, Clone 78, Gen 19)
Protein: p683_TZ2_NAT_EXP

I expect the Protein name to be the same, but I'm curious about the Project numbers.

mas92264 · Post by **mas92264** » Mon Dec 08, 2003 7:57 pm

No doubt the different run/clone/gen could have an effect on the outcome and I'm sure that there were differences in the 5 683's that I listed.

I guess a more accurate test would be to "capture" a wu and copy it to several computers and run it - if that could be done.

M

Zyzzyx · Post by **Zyzzyx** » Mon Dec 08, 2003 8:58 pm

thats' what I've done when I compared my 1600 TBird and 1700 TBred chips. Copied over the Work directory, along with unitinfo.txt and queue.dat. After backing up my originals as it was in the middle of a WU.

Also have the same version of FahCore_78.exe on both systems as well.

ColdFlame · Post by **ColdFlame** » Thu Dec 11, 2003 6:02 pm

Okay 3 of my machines got the same WU. Here are the frame times and PPW:

Protein name p256_p341_gnra_den credit 53.77

CPU - frame time - PPW
Celeron 2.9 Ghz - 13 mins, 23 sec - 404.98
AthlonXP 2100+ - 12 mins, 27 sec - 435.34
Duron 1.85 Ghz - 14 mins, 11 sec - 382.14

Note: I expect Celeron numbers to change a bit because I'm running a lot of other stuff on that Celeron including a domain controller and all stuff that comes with it, IRC client and I'm term served into it, chatting in mIRC, etc.

One conclusion I can draw from this: L2 cache does matter. My AthlonXP runs at 1.75 Ghz and Duron at 1.85 Ghz. You can see the results. When I get the "clean" Celeron frame times I will be able to say about whether Celeron sucks or not.

BTW, this new protein sucks. My Duron was giving me 450-500 PPW and now with this one only 380 PPW. Same PPW degradation on other machines.

dukla2000 · Post by **dukla2000** » Fri Dec 12, 2003 2:14 am

ColdFlame wrote:One conclusion I can draw from this: L2 cache does matter.

Seems to tally with mas92264 numbers where his Bartons are quicker than the rest (clock for clock).

ColdFlame wrote:BTW, this new protein sucks. My Duron was giving me 450-500 PPW and now with this one only 380 PPW. Same PPW degradation on other machines.

Certainly I agree it is difficult to benchmark processors using Folding. From my observations it seems you need (as this thread has swung towards) the identical WU and measure the same frame(s).

To show some of the variations: I record WU times occasionally for my boxen, and I always record when the boxen are otherwise idle. My P3 box averages around 315 ppw, but depending on the WU the expectation ranges from 280-510. Also my 2 tbreds are both clocked @ 2GHz. For protein 682 my ppw is 315 on box A, 336 on box B. For protein 344 my ppw is 497 on Box A, 486 on Box B.

BTW mas92264 you show your 2600T at 2.13GHz: I presume that is 16*133 and not 166fsb? I re-jigged your table with the 1600P as the baseline. Ignoring the 2000P (cause you say it does other bits) the 2600T scores 0.66 on the GHz ratio, 0.64 on the time ratio. i.e it just scaled up. The 2 Bartons have significantly better time ratios than their GHz ratio.

mas92264 · Post by **mas92264** » Fri Dec 12, 2003 7:24 am

BTW mas92264 you show your 2600T at 2.13GHz: I presume that is 16*133 and not 166fsb?

The 2600 T'bred is 166 fsb, so it's 12.5. Must've picked up the # from the wrong table. 2600/2083/166 (not 2.17 that I put in the table above) is correct (from the AMD tech doc.) Good catch!