all wu's crashing on HD5670
7 Jun 2012 23:12:01 UTC
Topic 196382
(moderation:
Good day.
I've recently installed and tested a 1GB HD5670 on computer 5393203/Draft-5.
After discovering from the interwebs that Catalyst v12.4 has no functioning openCL (? BOINC would not report any usable GPUs) I downgraded to v11.12 and was able to complete a few WU's, albeit with lots of failures.
Upon further reading in the msg boards i upgraded to v12.1 driver and tried again.
Most of my WU's are failing - it looks like I have a total of 6 completed units, 2 of which have validated.
The only other project is WUProp@Home, which should max at 2.5% of one core.
Does anyone have any ideas? I want to contribute to Einstein with my GPU but not at the expense of all these failed WU's.
Thanks!
Dell Precision 390 / C2D6700(2.66GHz)
WinXP SP3 32bit / 4GB DDR2
Sapphire HD5670 1GB / Catalyst 12.1
all wu's crashing on HD5670
)
Hi!
The good news is that when those WUs fail, they seem to fail quickly afetr starting, so there is no big loss of runtime (would be much worse if they failed near the end). Still, of course I understand your frustration.
The log output doesn't give me a clue what exactly is going wrong here. I would check the following issues just to make sure, tho:
- is your card ord PC overclocked? if so, try running at stock specs.
- is there any other application running in parallel with BOINC/E@H that is allocating excessive memory on the graphics card?
- I did not quite understand (some stuff I found was in French only) what WUProp@Home is doing and HOW it is doing it. I would suggest suspending WUprop@Home and see if this has an influence on the error rate. Just to make sure.
Cheers
HB
thanks for the ideas,
)
thanks for the ideas, but:
-card is run at stock speed
-nothing else is running (currently no other projects)
-i have disabled wu@home and will check again, but i think that this has nothing to do with it as i had the same errors before adding wu@home
here's a sample error output, it looks like "there was an error while deleting the color transform:"
7.0.28
There was an error while deleting the color transform. (0x7e3) - exit code 2019 (0x7e3)
Activated exception handling...
[11:22:04][3556][INFO ] Starting data processing...
[11:22:04][3556][INFO ] Using OpenCL platform provided by: Advanced Micro Devices, Inc.
[11:22:04][3556][INFO ] Using OpenCL device "Redwood" by: Advanced Micro Devices, Inc.
[11:22:05][3556][INFO ] Checkpoint file unavailable: status.cpt (No such file or directory).
------> Starting from scratch...
[11:22:05][3556][INFO ] Header contents:
------> Original WAPP file: ./p2030.20100902.G43.62-00.47.N.b4s0g0.00000_DM112.80
------> Sample time in microseconds: 65.4762
------> Observation time in seconds: 274.62705
------> Time stamp (MJD): 55442.037711691861
------> Number of samples/record: 0
------> Center freq in MHz: 1214.289551
------> Channel band in MHz: 0.33605957
------> Number of channels/record: 960
------> Nifs: 1
------> RA (J2000): 191254.517101
------> DEC (J2000): 91420.7707996
------> Galactic l: 0
------> Galactic b: 0
------> Name: G43.62-00.47.N
------> Lagformat: 0
------> Sum: 1
------> Level: 3
------> AZ at start: 0
------> ZA at start: 0
------> AST at start: 0
------> LST at start: 0
------> Project ID: --
------> Observers: --
------> File size (bytes): 0
------> Data size (bytes): 0
------> Number of samples: 4194304
------> Trial dispersion measure: 112.8 cm^-3 pc
------> Scale factor: 0.0065445
[11:22:07][3556][INFO ] Seed for random number generator is 1158627328.
[11:22:15][3556][INFO ] Derived global search parameters:
------> f_A probability = 0.08
------> single bin prob(P_noise > P_thr) = 1.32531e-008
------> thr1 = 18.139
------> thr2 = 21.241
------> thr4 = 26.2686
------> thr8 = 34.6478
------> thr16 = 48.9581
[11:22:22][3556][ERROR] Error during OpenCL kernel setup: HSFFB (error: -5)
[11:22:22][3556][ERROR] Demodulation failed (error: 2019)!
11:22:22 (3556): called boinc_finish
]]>
I also have ATI WUs crashing
)
I also have ATI WUs crashing instantly on start-up. HD 5670 and HD 6950, Catalyst 12.1 (last OpenCL release for WinXP). Like enginerd, I'm also using WinXP - I wonder if there's been much E@H/ATI testing on that OS? I know AMD isn't really testing drivers on WinXP much any more - I plan to move to at least Win7 eventually, but possibly not for a few months or so.
Soli Deo Gloria
I encountered the same
)
I encountered the same problem with an HD 5830 on XP. For some reason BOINC 7.0.28 and Einstein don't play well together on XP 32 bit no matter which driver you use. I uninstalled 7.0.28 and downgraded to 7.0.27 and the problem went away. 7.0.28 on Win7 64 bit works fine.
Oh wow, now that's
)
Oh wow, now that's interesting. I can't recall what BOINC version I tested on the HD 6950. But it's possible it was also 7.0.28.
Soli Deo Gloria
It worked for a week or so
)
It worked for a week or so now all mine are erroring out again after trying to update the drivers. Should have left it alone.
RE: [11:22:22][3556][ERROR
)
FYI, this simply means that the application couldn't acquire the resources (i.e. memory) it needs. This shouldn't be a problem, however, given your GPU sports 1 GB of RAM - as long a you don't too many tasks in parallel.
Oliver
Einstein@Home Project
RE: I wonder if there's
)
Well, we do have two dedicated Windows (XP32, Vista64) test systems with NVIDIA and ATI GPUs. We can't of course test all possible hardware/software permutations. You're invited to join our test project at Albert@Home and help us testing.
Best,
Oliver
Einstein@Home Project
RE: I uninstalled 7.0.28
)
I can't reproduce the problem using 7.0.28 myself on Windows Vista 64-bit using two HD 6970 GPUs and Catalyst 12.2. All WUs crunching happily...
It seems that the BOINC client up-/downgrade didn't really have an impact right? If it all, this appears to be a driver related issue, which is in fact also much more likely...
Cheers,
Oliver
Einstein@Home Project
nanoprobe suggested
)
nanoprobe suggested downgrading to boinc 7.0.27.
i tried this with no luck.
guess i'll wait for the official 12.6 ati drivers and try again.
:)