hey fellas, heres an interesting one for ya..
ive got 4 machines, all identical in hardware, all 4 have quadro 1700s (identical)
2 finish all their wus on time, 2 are always overdue to report. (honestly about 40% slower)
all are winxp x32, all using same latest nvidia drivers and bionc versions and using same boinc settings.
only 2 differences i can see
1) the two slowies had sp3 rc2 instead of the official sp3 release
ive done a full windows update so now they should be on par service wise with the other two, however ive got a feeling they are still slower.
2) the slowies are using 0.02% cpu+cuda and the fasties using 0.03%+cuda.
any way of forcing 0.03% on the slowies?
*they are all dedicated crunchers and all running wcg only but its not project specific as it happened with milky and couple otheres too.
any help would be appreciated :)
Odd indeed. Have you done BIOS setting comparisons (screen for screen) between a slow and fast box?
I can't be sure, but didn't SP3 rc2 require uninstall before you could install the full release SP3? You can use MBSA to check release level. You could also look in Add/Remove Programs to check.
Are there ANY differences at all in BOINC settings, preferences etc on each machine? Local preferences override global ones. Also, do you have an app_info.xml file or a cc_config.xml file that might be different between them?
And are you certain that the cards, MOBOs etc are all identical? Power supplies?? RAM??
all 4 machines are HP workstations, same skeleton in every way, except the cpu varies between them..
the cards are 100% identical in every way, same for ram, bios etc.. (hp bios isnt very flexible)
as for local settings, they are all 100% the same as well
the only opt appz i have on them is the milkyway and seti, but all 4 have the same..
id be willing to reimage both, but not sure if its hardware related in any way.
surely there cant be ThAT much difference between 2 winxp systems, one with rc2, one with official sp3, and both updated to latest updates. i mean were looking at more then 40% efficiency difference :pcwhack:
How different are the CPUs?
the two slowies are:
GenuineIntel
Intel(R) Core(TM)2 Duo CPU E8500 @ 3.16GHz [x86 Family 6 Model 23 Stepping 6]
(2 processors)
GenuineIntel
Intel(R) Core(TM)2 Duo CPU E6750 @ 2.66GHz [x86 Family 6 Model 15 Stepping 11]
(2 processors)
two fasties are:
GenuineIntel
Intel(R) Core(TM)2 CPU 6400 @ 2.13GHz [x86 Family 6 Model 15 Stepping 6]
(2 processors)
GenuineIntel
Intel(R) Core(TM)2 CPU 6400 @ 2.13GHz [x86 Family 6 Model 15 Stepping 2]
(2 processors)
different CPU's
doesnt that answer your question???
am I missing something here Bashhead Bashhead Bashhead
Quote from: ONYX on March 21, 2009, 07:24:09 AM
different CPU's
doesnt that answer your question???
am I missing something here Bashhead Bashhead Bashhead
Yes and No. Did you notice that the two fastest CPUs in fact are running at a lower speed.?
Yes, but we are talking about GPU work units, not CPU ones. So it appears that the faster CPUs are taking work time away from the GPUs, and the slower CPUs are allowing more efficient use of the GPU.
I also wonder if it may not have something to do with the WUs that are being crunched on the CPU.
Went to bed last night, expecting to find a GPU unit done and dusted, but awoke to find it has barely moved.... and TheGnat was crunching WCG, NOT MW... I guess I goofed before crashing in not updating. But are the boxes all crunching the same projects??
Yes, there are different types of GPU Grid work units, and it is possible that bigger units are being sent to the machines with the faster CPUs, though this would also be reflected in the credit granted for those units (the longer ones get roughly 50% more credit than the shorter ones).
The "Average credit per CPU second" shown by boincstats is the same 0.070000 for all 4 computers.
However if the wallclock credit rate is better for the E6400s this means the graphics cards are all working at the same efficiency but the combination with the 2 E6400s with a lower speed, less cache and a lower FSB is faster overall which is the reverse of what would be expected.
Only thing I can think of is the lower FSB means that lower overall memory latencies are being used. Memory strap issues are sometimes weird, sometimes a lower MHz memory speed works better for some applications than a higher one and this differs with different brands of memory and different motherboards due to the difference in where the northbridge latencies change. It's related to the tRD setting I think but most of this stuff is beyond my level of knowledge. Also sometimes memory runs at Command Rate 1 at lower speeds and Command Rate 2 at higher speeds and this can also make a difference. If most of the processing is done on the GPU, then memory performance may have a greater impact than CPU speed or amount of cache.
May be something else completely, differences in memory performance are just a possibility for consideration.