Advanced search

Message boards : Graphics cards (GPUs) : a lot of errors on a new gtx660 with 306.97

Author Message
Doc
Send message
Joined: 27 Nov 10
Posts: 6
Credit: 4,969,648
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwatwatwatwatwat
Message 27206 - Posted: 3 Nov 2012 | 17:20:41 UTC
Last modified: 3 Nov 2012 | 17:21:24 UTC

hello all

recently i got myself a gigabyte gtx660, an oc'd version

i was eager to run gpugrid apps on it, and it runs like hell...
...but errors a lot of wu's

anything i should check, correct, or do to make it run good?

...it's a complete fail, when it errors a longrun wu on a 88th percent

i'll appreciate any help on this

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 27208 - Posted: 3 Nov 2012 | 21:14:28 UTC - in response to Message 27206.

A few things to try:

- switch PC off, remove power cord (or switch PSU physically off) and wait for ~15 minutes, then try again (solves some random issues which persist after a normal reboot)
- lower the GPU clock by ~100 MHz - does it work now?
- do other projects run without errors?
- what about normal 3D, e.g. 3D Mark (whatever is the current one)

MrS
____________
Scanning for our furry friends since Jan 2002

Doc
Send message
Joined: 27 Nov 10
Posts: 6
Credit: 4,969,648
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwatwatwatwatwat
Message 27209 - Posted: 3 Nov 2012 | 21:51:02 UTC - in response to Message 27208.
Last modified: 3 Nov 2012 | 21:51:45 UTC

1 - I did that, not intentionally though, did not help

2 - will try that, it will take at least 24hrs of testing but worth trying

what surprises me (if that works), is that factory set card would not withstand a constant 100% workload (in that case it is not even exactly 100% > rather around 93-97%)

3 - SETI@home CUDA application by Lunatic (for Fermi chips) works almost fine
it does not produce any errors, but makes everything stutter and the load varies greatly between 15-90% all the time

4 - other 3d stuff (3dMark, UnigineHeaven, Furmark, OCCT, my games do not stress the card enough) works fine, absolutely no complaints, the card even overclocks a little bit, both core and memory

anyway, I will try downclocking the card, though it's a bit tricky - when downclocked it also drops the voltage (something to do with TDP offset, or sth...), which usually would be a nice thing, but not exactly in this case

thanks for your advice

Doc
Send message
Joined: 27 Nov 10
Posts: 6
Credit: 4,969,648
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwatwatwatwatwat
Message 27212 - Posted: 4 Nov 2012 | 13:09:24 UTC - in response to Message 27209.
Last modified: 4 Nov 2012 | 13:11:59 UTC

first conclusions:

I made a terrible mistake and forgot about a pci-e bus 1MHz overclock,
which I made while running my GTS250.
now it is set back to 100MHz... and this was probably the reason

anyway, I got a 1 whole wu completed and validated using 100% factory settings
the card was running at 1136MHz all the way, at 63degC also all the way
the fans were set to auto, which made them run at 63% > ca. 2000RPM (pretty loud)

now I am testing with the card underclocked (-117MHz) and undervolted (-0,063V),
which makes the card run at 62degC while the fans are running at 1500RPM (45%) and the noise/sound is really comfortable. minimum setting of the fan speed is 1380RPM > 40%

like I said before, underclocking and undervolting this card is a bit tricky
the card model is: GV-N660OC-2GD
tweaking software: Gigabyte OC Guru II

factory settings (effective under load): core 1136MHz, voltage 1,175V
both are lowered to: core 1019MHz, voltage 1,112V

by setting the following:
core down by 59MHz
power target down by 30% (do not be mislead, it runs at 82-84% TDP with this setting)
minimum voltage down by 0,05V (but I am not sure if it had any effect under load)
fan speed is set manually to 45%

the performance of the card should be down by roughly 10%, but now crunching does not cause any discomfort, plus I surely made the life span of the card noticeably longer

all of the above is still in stability testing period (I have a gut feeling it will be OK), I will confirm in ca. 48hrs

Doc
Send message
Joined: 27 Nov 10
Posts: 6
Credit: 4,969,648
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwatwatwatwatwat
Message 27216 - Posted: 4 Nov 2012 | 16:24:53 UTC - in response to Message 27212.

soooo far, so good.
it must have been that pci-e clock issue.

plus I found (genius...!) the cuda 4.2 overclocking thread...
...it would be a terrible misfortune if this card could not overclock at all,
seeing what could be done...

over and out

werdwerdus
Send message
Joined: 15 Apr 10
Posts: 123
Credit: 1,004,473,861
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 27223 - Posted: 4 Nov 2012 | 23:53:52 UTC

you could let it get up to 70c if you just want to lower the fan a bit. shouldn't hurt it at all.
____________
XtremeSystems.org - #1 Team in GPUGrid

Doc
Send message
Joined: 27 Nov 10
Posts: 6
Credit: 4,969,648
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwatwatwatwatwat
Message 27224 - Posted: 5 Nov 2012 | 0:09:28 UTC - in response to Message 27223.

yes, I think so too

but if it is possible to run in cool, quiet and fast - I will go for that

the present state is as follows:
overclocked core by 55MHz
TDP set to -12%
fans set to 45%

this resulted in factory clocks of 1136MHz under crunching load, but with the voltage dropped to 1.112V, plus if the GPU load goes down, the clocks are going for that moment up, until 1189MHz (as far as I remember)

this way the card runs at ca. 85-88%TDP, so the temperature does not exceed 63degC with fans set to 1500RPM, which is perfectly fine, and all that without loosing any performance (maybe even in some cases, with a performance gain)

at the same time, any memory overclock (even a mere 200MHz) results in an error

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 27225 - Posted: 5 Nov 2012 | 10:36:22 UTC - in response to Message 27224.

Sounds like a pretty good setup :)

MrS
____________
Scanning for our furry friends since Jan 2002

Doc
Send message
Joined: 27 Nov 10
Posts: 6
Credit: 4,969,648
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwatwatwatwatwat
Message 27227 - Posted: 5 Nov 2012 | 12:39:23 UTC - in response to Message 27225.

thanks :)

so, after 8h41m we have a longrun wu completed and validated, with 76k points,
without any anomalies during the process, I call that a success

next couple of days will show if I am right though

Post to thread

Message boards : Graphics cards (GPUs) : a lot of errors on a new gtx660 with 306.97

//