GPU voltage question

Alpha Gamer

Posts: 357   +119
My main gpu is a gigabyte gtx 670 (GV-N670OC-2GD) and I also have a gtx 560 ti for physx. I downloaded evga precision tool so that I could overclock the 560 ti and squeeze a little bit more performance when using physx. I started adjusting the voltage for the 560 ti and after pressing "apply", I realized that the software does not have "per gpu" voltage adjustment. It looks like it's some kind of master voltage adjustment, instead. I'm not even sure, I would like someone to explain me what it really is.

On the other hand, I'm quite sure I read on many articles about kepler based gpus that it's voltage adjustment is totally different from what we've seem so far. Based on that, it's quite impossible I may have tempered with it's voltage, meaning that the voltage I changed could only be the 560 ti's one.

My intention is to leave the 670 at stock settings and do some overclocking on the 560 ti alone.

Here are some images

http://imageshack.us/photo/my-images/217/gtx560tiprecisonvoltage.png/

http://imageshack.us/f/528/gtx670precison.png/

In this video here, they also talk about the new voltage system.


Thanks in advance
 
have you tried overclocking without a voltage bump? Is it also a stock cooler or a custom cooler on the 560ti?
 
Yeah, I did. It's ECS stock cooler. My only objective right now is to roll back the 670 to it's stock settings, if I really did mess them up, of course.
 
I would roll back both to stock voltages just hit the default button it should reset both cards, I'm not sure if it changes voltages for one card at a time, I think it does though. I can't use evga precision with my GPU or else I would check. I also don;t see what kind of benefits you will see from over clocking the 560ti more for physx, it's alredy a good card for it, and the physx engine does have it's limitation when it comes to using the graphics cards full potential. But if you do, can you upload a picture of of the voltage tuner with the 670 selected and then the 560ti selected?
 
As a matter of fact, the default button only changes the three settings right above it, that is:
1- power target (not available for 560 ti)
2- gpu clock offset
3- mem clock offset
In order to change voltages, you must press the reset button on the voltage tuner. And there's no way to see if it really changes voltages for one card at a time, once the tab is the same for either card ("both cards" ?). Sorry, English is not my native language.

After pressing both the reset and default buttons, I have this with the 670 selected
http://imageshack.us/f/85/670voltagetuner.png/
After pressing both the reset and default buttons, I have this with the 560 ti selected
http://img209.imageshack.us/img209/2532/560tidefaultreset.png

Now, about the overclock and reasons for doing it. After running Batman Arkham City's benchmark a dozen times, I realized that the 670 alone offers higher fps than when paired with the 560 ti (which is doing only physx). The overclock is an attempt to stop the 560 ti from bottlenecking the 670.

Edit - Even if the voltage tuner is only one, we can see different values in the center of both pictures. Therefore, assuming it does change voltages for one card at a time, rendering this whole thread useless... or maybe not, who knows?
 
Can I see the gpu-z sensor tab you can download it here if you don't have it already. I want to see the voltage in particular becasue I don;t know if both those cards have the same voltage, I don;t think they are supposed to. I find it so strange that the 560ti is not powerful enough for physx, double check and make sure it is the enabled phyxs gpu in the nvidia control panel setting. Also how many stream processors are on your Ti? the 670 has 896, you would need at least half that to avoid a bottleneck between the two cards. Unless you have the 448 core version you might be out of luck :/
 
I woulds say try it, but only a little mv at a time and make sure you're stressing the card out. Keep a really close eye on the voltages in gpuz that should pick up the cards voltages very well. and make sure you are only selecting the 560ti for the voltage changes and see if it;s the only card changing, the 988mv might only be the default setting, I would but it up on the 560ti and see if the 670 changed and if it didn't but the 560ti voltages change you might be good to go. The edit you posted does show the cards are being seen at different voltages so I don;t think all hope is lost with the effort.Keep me updated on the testing the individual volt changes, just be aware of both cards, keep gpuz running for monitoring! also how on earth are your idle temps so low, I mean like what do you do to get them that low?
 
Yeah, I'll keep trying. I tested Batman Arkham Asylum and had same results, with and without the 560 ti. On Metro 2033, the results seemed to be better with the 560 ti. I'll see if I can borrow other games that allow physx and see what happens.

About the temps,
1- my case is well cooled, or so I believe
2-Although it's still Fall, it's very cold in Sao Paulo right now. I'm wearing two pairs of pants and gloves indoors.

Thanks for all the help.
 
No problem, and thats why they are so cold, it's summer here in the US so it's hot :/. If you need any more help just ask.
 
Interesting Update:
I just ran three different benchmarks:
1- Cryostasis Tech Demo
2- Mafia II Benchmark
3- PLA Benchmark

And I had significantly higher results with both cards than with just the gtx 670. I don't know the exactly numbers, but believe me: what happened in Batman Arkham City can be considered an anomaly.
 
Interesting, I know cryotasis is a heavily physx geared game, so the extra card will help a lot. It is strange that it only impacts BAC in terms of performance, but I guess with physx cards you win some you lose some.
 
Batman:AC is somewhat notorious for producing lower framerates with a dedicated PhysX card regardless of mix.
BatmanAC_PhysX_DSOGaming.png

And with everything maxed at 19x10 (8xMSAA)
BatmanAC.png

A lot has to do with the the GPU / CPU workload and the nature of the physics effects being used. PhysX rendering fluid simulation and very heavy particle effects would probably net the benefit of a dedicated PhysX card - rippling (dynamic) cloth/paper, destructible enviroments and bog standard smoke/fog aren't going to benefit much (if at all).
 
Batman:AC is somewhat notorious for producing lower framerates with a dedicated PhysX card regardless of mix.

And with everything maxed at 19x10 (8xMSAA)

A lot has to do with the the GPU / CPU workload and the nature of the physics effects being used. PhysX rendering fluid simulation and very heavy particle effects would probably net the benefit of a dedicated PhysX card - rippling (dynamic) cloth/paper, destructible enviroments and bog standard smoke/fog aren't going to benefit much (if at all).
Very interesting! Where did you get those pictures from?
I would like to read the whole article.
 
MSI Afterburner is really easy to use.

Also, you aren't really supposed to fiddle with the voltages until your core/memory clocks become unstable.

I've kept my 6850 on stock voltages and got this far:

6850 stable.png
 
Yup, LNCPapa on the ball as per usual.
I didn't add the source since DSO had an ad/redirect unless I opened in a new window.
There are plenty of similar links available -most can be sourced via the PhysXInfo.com site.
Here's a vid from Linus ( Mafia II bench) for example:
 
Hoho,
Thanks for the additional info on the subject. The graphs and the video were very helpful. As I said earlier, I ran a couple of tests here, let's take a look at the numbers?

Passions Leads Army Benchmark
single gtx 670 = 75.0
dedicated gtx 560 ti = 91.6

Cryostasis
single gtx 670 = 84.1
dedicated gtx 560 ti = 89.7

Mafia II Benchmark
single gtx 670 = 40.3
dedicated gtx 560 ti = 59.9

All tests were done on 1920x1080.

If I happen to run any other tests, I'll just keep editing this post. Unfortunately, I won't be very precise on the tests, as I will probably be changing drivers between them.
 
By all means, keep the benchmarks coming in- PhysX benchmarks with dedicated PPU's aren't that common, and once you take into account how many possible card combinations can be used those results look very sparce if you are looking for one particular setup.
Driver revisions going forward shouldn't be a problem since people are more looking at percentage differences ( GPU+PPU vs GPU) than absolute frames per second.
I've used a dedicated PPU (GTX 280) until recently with my 580's, and while the framerate increases were in the "ok but nothing exceptional" bracket, I did notice that gameplay was generally smoother (and raising the minimum framerate portions) and suffered lower instances of framerate tanking during sudden use of dynamic processing effects
 
MSI Afterburner is really easy to use.

Also, you aren't really supposed to fiddle with the voltages until your core/memory clocks become unstable.

I've kept my 6850 on stock voltages and got this far:

View attachment 72902
Just for the record,
unless I missed something, MSI Afterburner does not provide options for overclocking a 2nd GPU (the gtx 560 ti in my case). Messing around with it, I could only overclock my main card.
 
More numbers:

Mars Benchmark
single gtx 670 = 67.51
dedicated gtx 560 ti = 46.57

Funny thing is that although the single card was faster, I felt that the rendering and animations were smoother with the 2nd card. Perhaps my eyes fooled me...

Isn't it possible to just edit my above so that all the information could be kept in one place?
 
I was afraid someone would say that... after a while the "edit" button disappears and only the "report" one remains.
 
That's weird. I just called up an earlier post from a few months back and the edit button is still there...
GkhRE.jpg

Maybe you're on some kind of double secret probation !
 
Back