GPU OCing Help Thread

[Zloyd]

Damn ive noticed that i lose work when i stop boinc and re-enable sli to game. I lose 1 gpu thread it errors out and starts on a new work unit

I usually right click on boinc and hit exit before re-enabling sli and I disable it again before i start boinc.

But the last 2 times its been crapping out. I think I will try a system shutdown and then restart to enable sli . Anyone else notice this issue ?

[Snow Crash]

Damn ive noticed that i lose work when i stop boinc and re-enable sli to game. I lose 1 gpu thread it errors out and starts on a new work unit

I usually right click on boinc and hit exit before re-enabling sli and I disable it again before i start boinc.

But the last 2 times its been crapping out. I think I will try a system shutdown and then restart to enable sli . Anyone else notice this issue ?

Are you getting the exit confirm dialog, are you choosing "Stop running etc"?
Sometimes it takes a few for the drivers to reset so maybe wait a minute or two between switching back and forth.

[Zloyd]

I was not getting the exit confirm dialogue have now enabled it.

[Zloyd]

Still getting errors , also dont seem to be getting a improvement in performance while crunching. Am I running a bad OC ? I am comparing to Snowcrash's and Whitefiredragon's 2xx cards but they are on xp64. Wonder if this is just win7 32 holding me back.

[WhiteFireDragon]

yeah your errors are most likely because of the overclocks because the error happens after a few minutes of the WU. if you get an error in under 15 seconds, the it's most likely because of the WU, but this doesn't seem like the case. lol it's a bad idea to clock your cards to try and match my numbers. my 295 is most likely a better clocker than the average 295. i'm running it at 1585mhz shaders with this co-op version, and XP also gives a good boost to the GPU utilization. if you switch to XP, you WU completion time will closer to mine. i have a regualr dual pcb 295 coming in soon, so i'll try to max out the clocks to see how it compares to the co-op version. maybe that one will be a better comparison to yours. what's your max stable OC?

btw, what's the difference between the gtx285 vs gtx275 in terms of crunching? both have same shader cores, with the 285 having barely faster shader clocks. any kind of shader overclocking will cancel out this advantage, and memory bandwidth and capacity does not matter, so is it safe to say a 275 = 285 when crunching?

[Zloyd]

Im thinking there is no way of really telling if I can come close to those numbers as long as I run 32 bit. Dont think im going to get close to those numbers with 32 , irrespective of the oc the numbers that are being churned out are more or less similar. Im not sure how much moving to win 7 64 will help this.

[Zloyd]

Where can I see projected gpu stats ?

Sorry White I didn't answer your question last time round but my stable gaming oc is 626c1100m corresponding shader for crunching I'm playing with the shader now but before I try something blindly I want to check my oc against a projected stat or something. I saw a page in the 480 thread or somewhere cant find it.

[Zloyd]

Have run into some driver issues. I cant get my 295 to sit at 3d clocks when I disable SLI , it used to happen before quite easily but now its a major pain in the arse and if I try to set the power management mode to max performance it goes to SLI on its own it seems.

I have tried driver sweeper but with no success , I have retried it last night and installed the latest whql drivers but the same symptoms remain. Any suggestions short of a complete reinstall ?

[Otis11]

When the clocks drop does the WU error? - Might be clocked to high...

That's the only thing I can think of... Not familiar with 295s

[Zloyd]

Nope its happening at stock clocks. Its dropping even before i can start boinc. And once i start boinc only 1 gpu clocks up and accepts work the other is just idling.

[WhiteFireDragon]

hmm weird problem. try enabling SLI and run a few stress test (3dmark and furmark) while monitoring the clocks and GPU usage with evga precision. if this seems fine, then disable SLI and crunch on it to see what happens.

[Otis11]

Hey guys, was just updating this and needed some information.

Since points seem to change quickly as they update the client, I axed that portion of the entries, but I'd really appreciate some more numbers for the cards we're running here at XS. (Especially the 240, 250, 270, 280 and all 4xx)

I also need a stock baseline of PPD for every card if possible and a PPD for the Strongest OC you've got (trying to get a range of what can be expected - but please don't exaggerate, this is all to get max ppd/$ for any given set up. People may use the info you submit to choice what to buy)

This is all the info I need:

Core/Shader/Mem - Your Name.......... PPD
.......... Brand: xxxx.......... Cooling: Stock fan @ xx% .......... Voltage: Stock .......... Temp: ???

Finally, does anyone know the steppings for OCing the 4 series?

Thanks for the help guys!

[WhiteFireDragon]

i'll try to posting results on all my cards, but i don't have a lot of results for baseling PPD, none of my CPU's and GPU (except one) are running stock . here's the first results on the minimum at maximum for my gt240's:

this is the worst case scenario for a gt240:
- Galaxy gt240 ddr3 512mb
- all stock clocks and voltages
- fan auto (40%) at 58C load
- 15.7k PPD (taken from RAC over a couple months of WU)
- notes: win7, WCG also loaded, machine used for daily tasks also

best case scenario for gt240:
- EVGA gt240 ddr5 512mb
- 405/1595/1701 clocks on stock voltages
- fan 85% at 67C load (temps a little unfair, it's sandwiched between two gtx295's with almost no room to breathe)
- 20.1K PPD (point projection using many different WU's types)
- notes: winXP, WCG not loaded, but it's sharing resources with 4 other GPU's in the system. you may get a little more if the GPU was by iteself

will look at stats for my other cards tonight

[Otis11]

Updated. - also added OS for reference.

Thanks for the info.

[PoppaGeek]

GTX 460 MSI 1gig DDR
800/1600/1800(mem stock)(volts stock)
Vista 64, Fan on auto - 59% with GPU @ 65% usage temp 48c room at 25.5c
Almost 30k a day but some days I play LOTRO for 1-3 hours.

Not using SWAN_SYNC, 6 core running WCG.

[Otis11]

Thanks

[PoppaGeek]

Thanks

Thanks for all you do for the team.

[WhiteFireDragon]

ok, i put one of my gtx275's at stock clocks for a few days just to test for the lower end of PPD for a gtx275.

this is the worst case scenario for a gtx275:
- evga gtx275 ddr3 896mb
- all stock clocks and voltages (648/1458/1188mhz)
- fan auto (40%) at 71C load
- 34.7k PPD (taken from PPD projection averaged from several types of WU)
- notes: win7, WCG also loaded, machine used for daily tasks also

here's the upper end for my other gtx275, but not maxed out yet:
- EVGA gtx275 ddr3 896mb
- 468/1620/1134mhz, on stock voltages
- fan 80% at 69C load, in a small mATX case
- 42.0K PPD (point projection using many different WU's types)
- notes: winXP, WCG also loaded, sharing resources with another crunching gts250 in the system. PPD will be a little higher with HT off, no WCG, and single GPU

will look at stats for my gtx295's, gts250, and gtx260 later. btw Otis11, you have a typo on the first page. the gts240 you labeled is not retail parts you can buy, and it's based on the older g92 chip. i think you meant gt240. and brand for my second upper end gt240 is evga, not galaxy. there's also no such thing is a gtx270 . i know these aren't important typos, but still might confuse some people trying to pick up a gts240 or gtx270 and can't find anywhere selling them

[Otis11]

there's also no such thing is a gtx270 . i know these aren't important typos, but still might confuse some people trying to pick up a gts240 or gtx270 and can't find anywhere selling them

I was wonderin' why everyone was doing such a poor job of gettin stats for those.

Thanks. Anything I can do to make this better is always appreciated.

[Kryja]

I've clocked my 275 a bit. Just upped the shaders and downed the core clock. I haven't touched the memory - there is no point right?

How do I find out expected ppd?

485/1630/1152 - ??? PPD
OS : xp 32bit .......... Brand: BFG.......... Cooling: Stock fan @ 65% .......... Voltage: Stock .......... Temp: 70

[Snow Crash]

I've clocked my 275 a bit. Just upped the shaders and downed the core clock. I haven't touched the memory - there is no point right?

How do I find out expected ppd?

485/1630/1152 - ??? PPD
OS : xp 32bit .......... Brand: BFG.......... Cooling: Stock fan @ 65% .......... Voltage: Stock .......... Temp: 70

Good work ... and you're correct, there is not much gain (maybe 1% at best) for OCing memory and you're running the best OS for GPUGrid
As for expected PPD ... that's kind of tricky, some WUs point better than others and some (although not recently) are double length so run a really long time and don't point particularly well, how stable is the card long term, it this a dedicated cruncher, etc. A ballpark firgure based on your last returned WU (first one with the new shader setting) gets to about 43k per day, however those WU types point the best so you might want to lower your expectations to about 40k per day.

[Kryja]

Cool thanks Snow. I've upped the fan to 70% now, gpu core is sat on 67. Nice little drop - just depends if I can sleep with it.

You think upwards of 1650 would be achievable?

[Snow Crash]

2XX shaders move in straps so your 1630 is really running at 1620.
(see the first post in this thread)
The next strap is at 1656 which, if you are not volting up will be pretty tight.
Then after that move on to 1674 and kill em
Here is how I test ....
Suspend Network Activity in BOINC.
Suspend Activity in BOINC.
Stop BOINC.
Make a copy of your BOINC data directory.
Mess with your OC and Restart BOINC.
If you get a Comp Error stop BOINC.
Delete BOINC data dir
Rename your backup to BOINC
MAKE ANOTHER COPY (can you guess I've forgotten this step more than once )

Repeat steps 4 and 5 until you are stable.
Once you are good then turn Network Activity back on.

This prevents you from pulling too many WUs and GPUGrid shutting you off for the day

Let us know how high you can go
KASHIF_HIV are the most demanding and TONI_CAP is about the average.

[Kryja]

Cheers Snow

I'll get to work after dinner and report back

[Kryja]

1656 shader is safe. Ran a KASHIF_HIV for a couple of hours and no errors. Halted everything, copied my BOINC folder and upped to 1669 and it errored my wu after a while. Wiped my boinc, coppied my backup in its place, still had the errored wu

I'm happy sticking at 1656 as I don't really want to start pumping volts into it. Not exactly going to get pie with just a 275 now am I Just happy to be putting some work out. Temps were @ ~70 with 70% fan. Good enough me thinks.