Two identical servers, different performance

[bruce]

It's normal for each computer to have a client with Machine ID set to 1. If you install a second client on the SAME machine, you have to use different MachineIDs.

[7im]

When the PPD numbers are being quoted here, are these numbers from a 24 hour period?

What is the PPD from each machine when you look at a week's worth of folding?

PPD can very quite a bit if one machine gets a -bigadv work unit, and the second machine does not and folds a regular SMP work unit instead.

[Tim_H]

do you have any frame times on the same Project for both machines?

[spitter3]

When the PPD numbers are being quoted here, are these numbers from a 24 hour period?

What is the PPD from each machine when you look at a week's worth of folding?

PPD can very quite a bit if one machine gets a -bigadv work unit, and the second machine does not and folds a regular SMP work unit instead.

http://folding.extremeoverclocking.com/ ... =&u=550074

[7im]

PPD per machine, not total user PPD.

[Grandpa_01]

There appears to be way to big of a difference between your frame times on those 2 WU's for being run on Identical rigs. 2686 and 6900 run very close to the same frame times on my rigs with the 6900 being the faster. You have a 18 min difference with the 2686 being the faster of the 2. I would say it is most likely a hardware problem have you benchmarked these machines to see if there is a performance difference between them. If you have what did you bench them with.

[grizli]

When the PPD numbers are being quoted here, are these numbers from a 24 hour period?

What is the PPD from each machine when you look at a week's worth of folding?

PPD can very quite a bit if one machine gets a -bigadv work unit, and the second machine does not and folds a regular SMP work unit instead.

I've looked at PPD over a long period of time. I check on boxes daily with FahSpy and they have been running for over a month, it has been like that since day one.

[grizli]

do you have any frame times on the same Project for both machines?

Right now both are working on 6900 project.
Machine A:

Code: Select all

[08:20:47] Folding@Home Gromacs SMP Core
[08:20:47] Version 2.27 (Mar 12, 2010)
[08:20:47] 
[08:20:47] Preparing to commence simulation
[08:20:47] - Looking at optimizations...
[08:20:47] - Created dyn
[08:20:47] - Files status OK
[08:20:54] - Expanded 24875224 -> 30796292 (decompressed 123.8 percent)
[08:20:54] Called DecompressByteArray: compressed_data_size=24875224 data_size=30796292, decompressed_data_size=30796292 diff=0
[08:20:54] - Digital signature verified
[08:20:54] 
[08:20:54] Project: 6900 (Run 32, Clone 20, Gen 59)
[08:20:54] 
[08:20:54] Assembly optimizations on if available.
[08:20:54] Entering M.D.
[08:21:00] Mapping NT from 16 to 16 
[08:21:05] Completed 0 out of 250000 steps  (0%)
[08:47:47] Completed 2500 out of 250000 steps  (1%)
[09:14:25] Completed 5000 out of 250000 steps  (2%)
[09:40:58] Completed 7500 out of 250000 steps  (3%)
[10:07:45] Completed 10000 out of 250000 steps  (4%)
[10:34:24] Completed 12500 out of 250000 steps  (5%)

Machine B:

Code: Select all

[10:02:14] Folding@Home Gromacs SMP Core
[10:02:14] Version 2.27 (Mar 12, 2010)
[10:02:14] 
[10:02:14] Preparing to commence simulation
[10:02:14] - Looking at optimizations...
[10:02:14] - Created dyn
[10:02:14] - Files status OK
[10:02:20] - Expanded 24872697 -> 30796292 (decompressed 123.8 percent)
[10:02:20] Called DecompressByteArray: compressed_data_size=24872697 data_size=30796292, decompressed_data_size=30796292 diff=0
[10:02:20] - Digital signature verified
[10:02:20] 
[10:02:20] Project: 6900 (Run 42, Clone 20, Gen 56)
[10:02:20] 
[10:02:20] Assembly optimizations on if available.
[10:02:20] Entering M.D.
[10:02:27] Mapping NT from 16 to 16 
[10:02:31] Completed 0 out of 250000 steps  (0%)
[10:35:59] Completed 2500 out of 250000 steps  (1%)
[11:09:25] Completed 5000 out of 250000 steps  (2%)
[11:42:54] Completed 7500 out of 250000 steps  (3%)

[spitter3]

Have you ran Linx on both machines and compared gflops , have you ran memtest and checked that all the memory settings are the same, timings/triple channel/ghzs??

[grizli]

OK, here are benchmarking results (used Performance Test 7.0)
Machine A (faster one):

Code: Select all

CPU Mark: 10382.6/11000
CPU Integer Math: 3878.7/4000
CPU Floating Point Math: 4490.9/5000
CPU Prime Numbers: 1374/2000
CPU SSE: 29.3/30
CPU Compression: 13335.7/14000
CPU Encryption: 42.6/50
CPU Physics: 708.6/800
CPU String Sorting: 7559.1/8000

Memory Mark: 2393.5/3000
Memory Small Block: 5071.9
Memory Read Cached: 2149.2/3000
Memory Read Uncached:2039.2/3000
Memory Write: 2257.1/3000
Memory Large RAM: 6707.2/7000

Disk Mark: 1377/2000
Disk Sequential Read: 140.1/200
Disk Sequential Write: 111/200
Disk Random Seek +RW: 129.7/200

Machine B:

Code: Select all

CPU Mark: 10308.8/11000
CPU Integer Math: 3837.2/4000
CPU Floating Point Math: 4477.6/5000
CPU Prime Numbers: 1358.9/2000
CPU SSE: 29.1/30
CPU Compression: 13353.4/14000
CPU Encryption: 42.3/50
CPU Physics: 682.8/700
CPU String Sorting: 7835/8000

Memory Mark: 2323.9/3000
Memory Small Block: 5086.3/6000
Memory Read Cached: 2152.1/3000
Memory Read Uncached:2037.8/3000
Memory Write: 2258.6/3000
Memory Large RAM: 6429.8/7000

Disk Mark: 1219.9/2000
Disk Sequential Read: 129.6/200
Disk Sequential Write: 104.7/200
Disk Random Seek +RW: 103/200

[7im]

Identical hardware should fold at the same speed, assuming you are getting the same types of work units on both systems.

Then something else is slowing down one of the systems, like an AV scan on higher priority robbing CPU cycles, or a higher temp in one system is causing CPU throttling, or a power saving mode is kicking in from the OS or from the BIOS. It will be one of those "smacks self in forehead" moments when you find it.

[spitter3]

If it was my machine, I would bump up the vcore a little bit on machine b and run the benchmark again, see if the number get a little closer, some cpu's just need a little more volts.

[grizli]

Starting to think it is temperature related... Machine B is surrounded by other boxes and gets pretty hot.
Windows/Bios both set in Performance mode. Though, can temperature really cause such big throttle?

[7im]

Server hardware should have come with some type of monitoring software. Check it for throttling.

Or load a monitoring app of your own, like CPU-Z, and watch the GHz of the chips in real time. Box A should match Box B. If not, it's throttling...