einsteinbinary_Errors_v1.39 BRP4G-cuda32-nv301
22 Feb 2015 0:01:33 UTC
Topic 197984
(moderation:
A Little Help please?
(I did just yesterday upgrade to a GTX 750Ti SC- hoping it will provide better Validations)
=Here's my full error log and system specs:
= My Host 11709374
=
Task 484538967
validate error(8:00001000) - Invalid
=
Task 484538963
validate error/Invalid
=
Task 484538965
(same)
=
my setup in Boinc Manager reads: (0.2 CPUs + 1 NVIDIA GPU)
my Pref settings for all the GPU apps is 1.0
=
App is v1.39 BRP4G-cuda32-nv301
using cuda API 3020|(0.2 CPUs + 1 NVIDIA GPU)
=
This is w: Intel DZ75ML-45K Mboard|chipset 9.4.0.1027 - i5 2500k
Intel HD 2000 iGD (disabled)
i5-2500k-3300 mhz clock
OpenCL 1.1 (Build 37149.37214) - 11 extensions
=
Nvidia GT 640 2048MB 128 bit DDR3 PCIe 2.0 - driver 347.09 WHQL
Ramdac 400mhz
2 Kepler cores @ 901mhz (overclocked the core from 901 to 1094)
Mem from 800 to 1002 Mhz bandwidth is 32.1 GB/s
Fan runs @65% - Temp never over 49 even at 97-99% load(GPU-Z)
384 CUDA cores
CUDA 7 driver
OpenGL 4.5 (GeForce GT 640/PCIe/SSE2 with 342 ext.)
OpenCL 1.1 CUDA 7.0.23 FULL_PROFILE - 16 extensions
Compute capability 3.0
=
Exit BOINC and Updated to driver 347.52 | (2 releases past 347.09)
Feb 13 9:15p CST | 3:15 UST
Reset Both Clocks to Default: 901Mhz Cores |800 Mhz Mem|25GB/s
=
Feb 14 12:30p CST= Currently crunching a BRP4G; one more left in my queue.
Waiting for this one to finish to check for error status prior to posting
this
=
Task 484529674 = Validated
==stderr output "looks" the same (to me) - I'm unable to determine the meanings of them yet ie-error hunting
the only persistent "error" I see is something about "dirty SumSpec pages" and "Checkpoint file unavailable: status.cpt (No such file or directory)."
=Is there a diag Flag(s) to set in BMgr to correct or debug these errors?
Thanks in advance
JBird
einsteinbinary_Errors_v1.39 BRP4G-cuda32-nv301
)
It appears that you were overclocking your card. That is probably the cause of the errors.
RE: A Little Help
)
Are you saying you have a 750Ti installed?
Because it still says you have a 640 with the driver: 34752
I have heard people saying that driver causes them problems so they switched back to to 34725 or 34709
You can look at all of my GPU's and I don't use that driver you have.....and ALL of mine are overclocked and *superclocked* and never fail.
Run at an average of 58C running BRP PAS X3 24/7 for a couple years now.
RE: A Little Help
)
You don't have any tasks for the 750Ti at the moment. With BRP4G now out of work for the time being, the only way to get work for your new card is to enable either the BRP5 or the BRP6 (or both) searches. I'm sure your 750Ti would do very well with either of those searches.
....
If you had reset the clocks to default before doing that last task, that seems to be pretty conclusive evidence that the overclocking was to blame for the previous validate errors.
There are no errors to correct or debug.
The apps from the project that volunteers run are designed to return certain information to the Devs. These programs can receive input on a channel called "standard input" (stdin for short) and send results type output to a channel called "standard output" (stdout). There is another output channel available called "standard error" (stderr). Stderr is typically also used for non-result and non-error type ancilliary output such as diagnostics and statistical information. In other words, anything that is not precisely the 'answers' being sought is likely to be directed to stderr.
When you review the stderr output by clicking on the task ID link on the website, you should expect to see mainly diagnostics and statistics. You can distinguish these by noticing the [INFO] tag near the start of these lines. Notice that the particular messages you quote from the stderr output all have the [INFO] tag prepended to them.
If there is an actual error detected, there will be an output line with the [ERROR] tag near the start. I wasn't able to find any such lines in the stderr output of any of your tasks. This is not surprising as overclocking may tend to give rubbish answers rather than create an error type that the app can detect and report.
Cheers,
Gary.
Thanks Magic for the Reply
)
Thanks Magic for the Reply and Notes.
Yes I *do have a EVGA 750Ti SC 2048 Maxwell NOW; but haven't tested it against any jobs(BRP4G) at Einstein yet.(Just finished up my queue there the other day and haven't reloaded yet)
But the last BRP4G *did Validate and 1000 credits; and yes, it was on the GT 640 2048 Kepler at stock clocks(901mhz Cores and 800mhz Memory) and the 347.52 Driver and PrecisionX-set Fan at 75% with core temp hovering at 50 thruout the run. (BOINC screensaver disabled too--I know Bernd said it's not *supposed to matter, but it does on my Build/experience here).
=
So, logical conclusions at this point: GT 640 Kepler works *best at stock clocks
on these cuda32 fueled apps even with the cuda7 driver loaded (The GT 640 Kepler comes with)
JBird
Thank you Gary for Joining
)
Thank you Gary for Joining in; was hoping you would.
Yes I'm still *after it - that is, figuring out how everything works.
Honestly, hoping to find *where to look to uncover whatever causes these fails and troubleshoot the problems if I can; other than my rote experiment of resetting clocks to stock revealed.
Thanks for the nudge to BRP5 and 6. Think I'll do just that!
Gotta reset/turn down cache first while I'm there ordering apps.
I'll let y'all know how it goes.
=
PS - The 750Ti SC has almost doubled my RAC at Seti; running cuda50s 2up
(0.04 CPUs + 0.5 NVIDIA) Anonymous platform app with mbcuda.cfg priority set to High
=
JBird
Hey Gary, What is the "other
)
Hey Gary,
What is the "other name" for the BRP6? Is that Perseus Arm?
I don't see it in Einstein Apps nor Prefs "as such"
I'm getting ready to load a minimum cache, nothing but GPU apps and see how the 750Ti does with them.
PS- I did see a cuda55 app setup, but it's for Mac.
It's my perception that the cuda32 library/SDK could be the holdup in my case at least, with these task errors.
Thanks
JBird
Binary Radio Pulsar Search
)
Binary Radio Pulsar Search (Parkes PMPS XT) is the name of the new application.
Thanks robl, I did get those
)
Thanks robl, I did get those but they "show" as BRP5-cuda32-nv301 in my Tasks sheet.
I'm hunting BRP6 which one is that?
JBird
Binary Radio Pulsar Search
)
Binary Radio Pulsar Search (Parkes PMPS XT) = BRP6
See this thread in tech news and this message in science for more info about the new search and why the plan class show as BRP5 although they belong to the BRP6 search.
BRP5 --> Binary Radio Pulsar
)
BRP5 --> Binary Radio Pulsar Search (Perseus Arm Survey)
BRP6 --> Binary Radio Pulsar Search (Parkes PMPS XT)
Binary Radio Pulsar Search (Perseus Arm Survey) "BRP5" - transition to "BRP6"