project:6892 run:921 clone:0 gen:42 - BAD_WORK_UNIT

[Napoleon]

Seems clear enough, date is 2011-12-24 UTC.

Code: Select all

00:04:14:Unit 00:Completed 52500 out of 250000 steps  (21%)
00:12:00:FahCore, running Unit 00, returned: BAD_WORK_UNIT (114 = 0x72)
00:12:00:Sending unit results: id:00 state:SEND error:FAULTY project:6892 run:921 clone:0 gen:42 core:0x78 unit:0x0000002f6652edc54e25d8dcd41703f8
00:12:00:Unit 00: Uploading 581B to 171.67.108.53
00:12:00:Connecting to 171.67.108.53:8080
00:12:01:Connecting to assign3.stanford.edu:8080
00:12:01:Unit 00: Upload complete
00:12:01:Server responded WORK_ACK (400)
00:12:01:Cleaning up Unit 00

However, this is my second uniproc WU failure recently, previous report is in viewtopic.php?f=19&t=20240. I'm not willing to blame these on my CPU OC yet, but certainly will be watching it, just in case the OC is getting unstable after about one year of almost nonstop folding @2.1GHz. I'd appreciate follow-up info on these two WUs.

[bruce]

Hi Napoleon (team 191980),
Your WU (P6892 R921 C0 G42) was added to the stats database on 2011-12-23 17:04:09 for 0 points of credit.
Hi Anonymous (team 0),
Your WU (P6892 R921 C0 G42) was added to the stats database on 2011-12-25 13:04:17 for 136 points of credit.

I'd vote for it being your overclock since somebody else completed it successfully.

(Still no data back on the other one, but it has deadlines of 27 - 39 days.)

[Napoleon]

I have a list of suspects for this one, ATM:

1. Flaky power. It was quite stormy over here during Christmas, and based on news I was fully expecting to see my rig to have lost power and stopped folding while I was traveling. It was still folding, but elsewhere I could see that there were problems and actual blackouts
2. CPU is getting "tired" after 1.6GHz ==> 2.1GHz OC plus practically 24/7 folding for about a year
3. Borderline memory stability

I let the BIOS handle the OC automatically. I noticed that the automatic OC drops effective memory clock from 1067MHz to 1048MHz. For a tiny (1067MHz - 1048MHz) / 1067MHz * 100% == 1.78% drop in memory clock, the BIOS apparently sees fit to tighten one memory timing parameter by 1. Don't remember which one, but I think BIOS reduced the value from 8 to 7 (12.5% drop).

Not the first time I've seen a bit-too-tight memory timings causing subtle problems with folding on an older setup, viewtopic.php?f=19&t=15821#p156883. If only I had ECC memory to aid bug hunting this time around. Anyway, I'll let the current uniprocessor WUs finish and do some stress testing on the CPU and memories while the GPUs are folding as usual. Thanks for the follow-up(s), much appreciated.

[Napoleon]

While I'm still quite convinced that this failure as well as the previous one mentioned in the OP were freak occurrences instead of a CPU OC becoming only marginally stable after a year of 24/7 folding, I decided to reduce the CPU OC from 2.1GHz to 2.055GHz (3x 685MHz FSB). CPU stock clock would be 1.6GHz (3x 533MHz FSB), BTW. I also bumped up the core voltage a bit, I'm staying slightly below rated maximum core voltage, though.

Further study of various finer points of BIOS actually paid off otherwise, too. I found out I can run my memory at its full rated speed instead of the stupid 1048MHz effective clock from automatic OC. I'm able to bump up the memory voltage a bit as well, while staying slightly below its rated maximum voltage. I'm now running the memories at 2x 685MHz == 1370 and BIOS now applies the JEDEC #4 stock timings from SPD for 2x 685MHz == 1370MHz clock.

Updates for burn-in and folding details linked in my sig, in case anyone is interested. Hoping for a rock stable folding year 2012 with this rig!