2665 units drastically underutilizing my cpu's

[stanmc]

I have two AMD 4-core processors running @ 2.9 and 3.2 Ghz. When they are processing 2671 and 2677 units they are close to 100% usage on all four cores. Under FAHcore_a1 with a project 2665 unit they have one core at 20-30%, one near 50%, one in the 70% range and one near 90+%. The performance on the 2671/2677 units is approximately 6 minutes per % of work. On the 2665 they are 15 minutes per % of work. What a waste of my machines, especially since they are dedicated to FAH. I do two units per day per machine normally. At the current rate on 2665 I will complete one a day per machine. Damn. To make matters worse, the points awarded are the same. My progress in points is cut by 50%.


Not that I'm complaining.

The first 2665 I downloaded for my 3.2GHz machine failed at 0% with a NaN. (P2665/1/888/117) Miraculously, it cleaned up after itself and downloaded a new unit.

[toTOW]

Search on the forum ... it's a normal behavior for the A1 core.

[stanmc]

Search on the forum ... it's a normal behavior for the A1 core.

Thanks for your advice on doing a search. I did search and didn't find anything that confirmed the poor performance of the A1 core in a manner similar to what I find on my machines. The A1 core really sucks.

[stanmc]

The A1 core really sucks.

After spending 1 day 0 hours 39 minutes and 17 seconds on one 2665 unit and 21 hours 57 minutes 52 seconds on the other, both of them aborted in writing the file to disk.

[12:38:17] Leaving Run
[12:38:21] - Writing 22081679 bytes of core data to disk...
[12:38:21] ... Done.
[12:41:40]
[12:41:40] Folding@home Core Shutdown: INTERRUPTED
[0]0:Return code = 0, signaled with Quit
[0]1:Return code = 0, signaled with Quit
[0]2:Return code = 0, signaled with Quit
[0]3:Return code = 18
[12:41:45] CoreStatus = 12 (18)
[12:41:45] Client-core communications error: ERROR 0x12
[12:41:45] Deleting current work unit & continuing...

This was a complete waste of my computers time.

[7im]

Nothing fah related is a waste of computer time. A1's came first, and still need to be completed to finish the scientific work. A2s are an improvement. Instead of bashing A1s for going slower, you should, IMO, be thanking the project for making improvements in the A2 to better use your computing resources/donation.

[stanmc]

Nothing fah related is a waste of computer time. A1's came first, and still need to be completed to finish the scientific work. A2s are an improvement. Instead of bashing A1s for going slower, you should, IMO, be thanking the project for making improvements in the A2 to better use your computing resources/donation.

I think you've missed the point. After spending a considerable amount of time on processing the A1 units, they both failed to complete writing the results to disk and thus could not be uploaded. So at the minimum spending 21 hours in one instance and 1d and several minutes in another and not getting to upload the units and also not getting points for them seems to me to be a MONUMENTAL waste. I would love to know how else to categorize spending that much time for nothing. Thanks for your view on this.

[7im]

I agree, WUs that fail, whatever the core version, are a waste of resources in some regard. But a few bad WUs are always expected with any core type. Part of the game.

On the bright side, you probably moved on to some A2 work units after these A1s failed, and you're machines are back to using 100% CPU sooner than if you had finished those WUs normally.

Then again, we have to ask if these were actually bad WUs, or if too much overclocking a CPU or memory is the cause? NaNs are most often caused by bad or unstable hardware configurations, but not always.

[stanmc]

I agree, WUs that fail, whatever the core version, are a waste of resources in some regard. But a few bad WUs are always expected with any core type. Part of the game.

On the bright side, you probably moved on to some A2 work units after these A1s failed, and you're machines are back to using 100% CPU sooner than if you had finished those WUs normally.

Then again, we have to ask if these were actually bad WUs, or if too much overclocking a CPU or memory is the cause? NaNs are most often caused by bad or unstable hardware configurations, but not always.

Indeed I have moved on to A2 units. In the amount of time expended on the A1 units which reached 100% in two instances, I could have done two A2 units per processor and completed them successfully most likely as have the previous 96 A2 units. My systems check out on every stability check I can find. I think 96 successful A2s proves that. Once again, thanks for your comments and ideas.

[7im]

Prime95 is not a good test. OCCT is good. StressCPU v2.0 is very good, but doesn't stress the memory, so use MemTest86 or the MS Mem Tester as well. The Intel Burn In utility is also very good.

But again with A1s vs. A2s, A1s are like standard CPU work units for SMP pointwise, and A2s are like bonus work units. You get what you get. They all need to be processed. Getting upset about WU assingments is like getting upset about the weather. Not much we can about it.

[shdbcamping]

I believe the complaint would be relating to the extremely high failure rate of the 2665 class WU's, whether under VM or SMP folding. They are very problematic and will even crash a system. I have less than 15% success with these on 3 different machines. Sometimes it just feels good to complain about the weather

[stanmc]

I believe the complaint would be relating to the extremely high failure rate of the 2665 class WU's, whether under VM or SMP folding. They are very problematic and will even crash a system. I have less than 15% success with these on 3 different machines. Sometimes it just feels good to complain about the weather

Well somebody got the message. I don't mind a little underutilization of my machines. I have recently spent $800 building machines to fold for FAH. But when programs complete slowly and then don't write the data to the disk correctly thus not affording an upload, I think I have a valid objection. As for the weather, I live in Chicago and we always say just wait a few minutes it will change (not always for the better). I think I can beat your 15% sucdess rate. Mine is zero. Thanks for your comment. As for the other dude, I've run the standard mem checks and stressed the machines. They check out. The best test I've found is consistent performance on FAH (especially on smp with the A2). 97 good ones. My machines are correctly set up. They work correctly. They do what they are supposed to do the way they are supposed to do it. Now I'm going to go find one of my friends and complain about the weather and politics, past current & future.

[Flathead74]

I have four systems running these WUs, and since September 3rd, I have had only two E_U_Es, both of which were p2665's.
These were on two separate machines.
One of them is my responsibility because the heatsink was dirty.

I did not count the number of completed p2665 WUs, but I'm certain that the success rate is well better than 15%.

The failure (E_U_E) rate for these systems is more along the lines of 1% - 1.5%

I don't think there is anything wrong with the p2665 WUs.

[Mactin]

A little bit of fah history:
Remember that the 1920 points for a p2665 includes a substantial bonus. When they where first issued, in May 2008, the benchmarking process gave them only 1275 points. After much protest was the bonus applied.