Low memory clock on Maxwell2 cards (960/970/980, probably Titan X)

James
James
Joined: 11 Jul 13
Posts: 17
Credit: 347,034,869
RAC: 162,472

I am not trying to boost

I am not trying to boost speed of WU's just reduce my BRP4 and 6 WU invalids. With a brand new GTX980Ti I am seeing 50% invalids.

I am working that out in another thread so no need to rehash over drivers and such.

Reading this thread and using inspector I had similar P2 issues and made reasonable Memory speed increases to 3505 which seems to be stable with no ill effects.

According to NVI and PrecisionX, my GPU clock is running at 1404mhz but CPU-Z, graphics section and the P2 overclocking section of NVI says my GPU clock is 595mhz. So my question is, what am I to believe is the actual GPU Clock speed? 1405 or 595?

Thanks, and awesome thread!

Jim

Zalster
Zalster
Joined: 26 Nov 13
Posts: 2,934
Credit: 3,071,507,649
RAC: 729

I'm going to hazard a guess

I'm going to hazard a guess and say you are OC your 980Ti?

If so, might want to go back to stock speed and see if the Invalids go away.

If they do, then you have to incrementally increase the speed until you find the point invalids start to pop up again.

Also, how many work units at a time are you running? Have you been seeing the driver crash and recover? Do you reboot the entire system when it does this?

Gamboleer
Gamboleer
Joined: 5 Dec 10
Posts: 173
Credit: 166,790,515
RAC: 0

RE: I am not trying to

Quote:

I am not trying to boost speed of WU's just reduce my BRP4 and 6 WU invalids. With a brand new GTX980Ti I am seeing 50% invalids.

I am working that out in another thread so no need to rehash over drivers and such.

Reading this thread and using inspector I had similar P2 issues and made reasonable Memory speed increases to 3505 which seems to be stable with no ill effects.

According to NVI and PrecisionX, my GPU clock is running at 1404mhz but CPU-Z, graphics section and the P2 overclocking section of NVI says my GPU clock is 595mhz. So my question is, what am I to believe is the actual GPU Clock speed? 1405 or 595?

Thanks, and awesome thread!

Jim

Hi Jim,

I'm not an expert on this, but:

For NVI, if you make the changes described in this thread and click the button to apply in the lower right, you should see the "Current Clock" value increase on the lower left side of NVI once you start work units. I believe this is your true value.

The clock shown in CPU-Z seems to be the base, non-multiplied, non-overclocked clock rate, which on my 960 is 540 MHz, when the card runs at 1078. CPU-Z does this with memory as well, which on my machines always reports at exactly half its clock speed. (Oddly, HWMonitor on my AMD setups shows the true, current rates as set by the clocking software I use on those machines, MSI Afterburner).

For the invalids, I had a factory-overclocked AMD 7970 that was doing exactly the same thing, though it would run benchmarking software (Furmark) for hours without any problems. I was able to get it to nearly 100% valid results by underclocking the core clock to 925 from 1000 (which set it back to the normal rate for the card) and memory from 1300 to 1200 (which was a fairly substantial underclock).

James
James
Joined: 11 Jul 13
Posts: 17
Credit: 347,034,869
RAC: 162,472

RE: I'm going to hazard a

Quote:

I'm going to hazard a guess and say you are OC your 980Ti?

If so, might want to go back to stock speed and see if the Invalids go away.

If they do, then you have to incrementally increase the speed until you find the point invalids start to pop up again.

Also, how many work units at a time are you running? Have you been seeing the driver crash and recover? Do you reboot the entire system when it does this?

No OC, not the CPU, RAM or the GPU. Completely stock out of the box I was getting around 70-90% invalids, after updating drivers I was able to get it around 30-50% invalid rate. I then read through this thread and changed the P2 setting but didn't see any change in WU completion or invalids.

The current driver, 361.43 is stable, not crashes , and seems to have the least amount of invalids.

I spent about a week and a half playing with 1 GPU per WU then 2, then finally 3. Didn't have any effect on the ratio of invalids.

Thanks for providing feedback.

Jim

James
James
Joined: 11 Jul 13
Posts: 17
Credit: 347,034,869
RAC: 162,472

RE: Hi Jim, I'm not an

Quote:

Hi Jim,

I'm not an expert on this, but:

For NVI, if you make the changes described in this thread and click the button to apply in the lower right, you should see the "Current Clock" value increase on the lower left side of NVI once you start work units. I believe this is your true value.

The clock shown in CPU-Z seems to be the base, non-multiplied, non-overclocked clock rate, which on my 960 is 540 MHz, when the card runs at 1078. CPU-Z does this with memory as well, which on my machines always reports at exactly half its clock speed. (Oddly, HWMonitor on my AMD setups shows the true, current rates as set by the clocking software I use on those machines, MSI Afterburner).

For the invalids, I had a factory-overclocked AMD 7970 that was doing exactly the same thing, though it would run benchmarking software (Furmark) for hours without any problems. I was able to get it to nearly 100% valid results by underclocking the core clock to 925 from 1000 (which set it back to the normal rate for the card) and memory from 1300 to 1200 (which was a fairly substantial underclock).

How would I know if its factory overclocked? Nothing on the box nor the manual says its factory OC'd. Its the FTW (for the win) version so maybe it is?

Interesting idea about de-rating the card, I hadn't thought about that, may as well give it a shot.

Thanks
Jim

Jacob Klein
Jacob Klein
Joined: 22 Jun 11
Posts: 42
Credit: 66,715,711
RAC: 41,338

FTW is EVGA's motto for "For

FTW is EVGA's motto for "For the Win", which is generally their most overclocked version of a GPU.

To see if you're GPU is factory-overclocked, open GPU-Z, and compare:
GPU-Z "Default Clock" (which is the factory default non-boost clock of your GPU)
versus
"Base core clock (MHz)" on the following wiki pages (which is the reference stock non-boost clock)

https://en.wikipedia.org/wiki/GeForce_600_series
https://en.wikipedia.org/wiki/GeForce_700_series
https://en.wikipedia.org/wiki/GeForce_900_series

For your GTX980 Ti, for instance, the reference stock clock is 1000 MHz ... but I'm betting yours is overclocked to 1190 MHz, eh? That's a 190 MHz overclock! Time to downclock (with NVIDIA Inspector or MSI Afterburner) until it's stable!

PS: I just tonight purchased the same GPU you have, I think.
EVGA GeForce GTX 980 Ti FTW GAMING ACX 2.0+
http://www.evga.com/Products/Product.aspx?pn=06G-P4-4996-KR
... and I can't wait to clock it up or down, to find its sweet spot.

PS: If ever you see a ~550 MHz clock when you're expecting full power, it generally means that your GPU had a TDR (with a log in C:\Windows\LiveKernelReports\WATCHDOG\), and if so, it's either a driver problem or an overclock problem. Either way, restarting the PC should restore normal clocks, but if the problem is an overclock, you must downclock a bit. I'd recommend downclocking in 30 MHz intervals, to hone in on the sweet spot.

Zalster
Zalster
Joined: 26 Nov 13
Posts: 2,934
Credit: 3,071,507,649
RAC: 729

I agree with Gamboleer's

I agree with Gamboleer's recommendation.

Try underclocking it some and see if the Invalids go away.

My 980Ti run between 1250 and 1350 so you don't have to bring it down much to check

James
James
Joined: 11 Jul 13
Posts: 17
Credit: 347,034,869
RAC: 162,472

Man I am very impressed with

Man I am very impressed with the level of help I get from these forums. Everyone has offered good sound advice and been patient with my....uh.... learning curve. :)

I used NVI to declock by -90Mhz which is the lowest it will let me go....that was the same with PrecisionX not sure if I can go below that but I guess its a start.

Current clock is 1315Mhz with 93% GPU load and 74% MCU.

I stopped BOINC completely before making these changes then restarted it.

I am still running 3 x WU and the estimated runtime jumped to 6+ hours but that usually drops significantly after a few WU's are uploaded.

I will let it run as is through the night and probably tomorrow and see what we get.

Jacob, it does appear we got the same card, I am liking it so far, outside of BOINC it runs excellent although I haven't tried any really GPU intensive games....pretty much Division only. If you get your card dialed in I would love it if you could let me know what settings you ended up with, if I get it dialed in I will do the same.

Zalster, did I read your setup's correctly? your running 4 Titan's and 9 980Ti's?? Isn't that like almost $10K worth of GPU's? that is some real dedication man!

Again, thanks everyone for your help.

Jim

Mumak
Joined: 26 Feb 13
Posts: 312
Credit: 1,663,030,699
RAC: 690

If that won't help then my

If that won't help then my guess is that there might be something wrong with the card (memory perhaps?). Something that didn't reveal the manufacturer's validation process.
I had a few different cards (AMD and NV) that after some time started to produce invalid results here. When there's some problem with the GPU (like memory) this might not be noticed in games, but computing tasks which validate results will show that pretty well.

-----

James
James
Joined: 11 Jul 13
Posts: 17
Credit: 347,034,869
RAC: 162,472

Man, I think we got it. Out

Man, I think we got it. Out of 22 WU's only 2 were invalid but I think they were uploaded before I made the changes above.

If it remains the same I will leave it for the weekend and see what I get before pushing the settings further and seeing what the card can tolerate.

Again, thanks for the help, greatly appreciated!

Jim

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.