metallicamaster3

To be fair, being this an Overclocking forum, it was pretty easy for this to be stickied since there is no current thread on OC'ing nVidia cards -- whereas if someone makes a thread explaining how power works inside GPUs, it probably won't get stickied as easily unless it was really popular.

__________________

PizzaMan

Great thread!

rep'd

__________________

____________________________
.:|Official|Volt|Mod|Squad|Member|:.

GTL testing and tweaking for all 7 series nVidia boards

TurboTurtle

Updated slightly. Made easier to read, added TwoCables' comments and common problems/solutions section.

Also, MM3 - you definitely need to make that Zeus picture your avvy.

__________________

metallicamaster3

Hahahaha, maybe my profile pic ! Only because my avatar is forever reserved by Cliff <3.

__________________

TwoCables

Update: 720/1440/999 was unstable, so I'm now trying the aforementioned 684/1476/999 that I accidentally tested when I was actually shooting for 684/1368/999.

So far, so good. But as before, time will tell. As soon as I submit this reply, I'm going to see if I can get it to pass 30 minutes of ATITool's artifact scanning, and maybe 10 minutes of Furmark at 1440x900 Windowed with 4x MSAA again. I like Windowed, but I also like to make it as big as possible without hiding too much of the big furry donut. Anything bigger cuts off too much of the top or bottom of it.

So, here I go! Oh, and by the way: 684/1476 is what I get when I link the 2 clocks (but it actually choose 1474, even though the actual is 1476).

Edit: I just learned that this is confusing (link), so here's what I meant: when I have it Linked and enter "684", it wants to use 1474 even though the actual clock ends up being 1476 anyway. So, tries to use 1474, but it ends up being 1476 instead due to the strap.

__________________

TurboTurtle

Interesting. I myself have not yet found a well-founded answer as to why precision sometimes set/reports either the core or shader clock 2-5MHz below the requested, even if it falls within the correct parameters. Still searching on that end.

Does it change for you if you unlink them? I've always had them unlinked for my overclocking just to have the ability to test one clock at a time.

__________________

metallicamaster3

This is why http://www.overclock.net/nvidia/4976...er-straps.html

__________________

TurboTurtle

Right, and as the straps show it should be set to 1476 as requested, but is being set to 1474 according to Cables. 1474 is not the strap-point.

Same happens with my core clock at certain speeds - even stock. The stock is supposed to be 590, but as you can see in my first picture in my original post it's set to 586.

__________________

TwoCables

Doh. I haven't tested it yet. I got caught up with other things and I forgot all about it. I'll begin testing as soon as I post this as I will have nothing to do afterward.
But what I said was that the Linked option wants to use 684/1474, but the actual shader clock ends up being 1476 due to the strap. I mean, for some reason, it thinks that 684/1474 is linked (it wants to use 1474 as the shader), but it ends up being 684/1476 instead. So, in other words, the actual clock is not ending up at 1474. That's what I meant when I said "...even though the actual is 1476". I re-read what I posted, and even I got confused. I'll edit it in order to clear up the confusion. I'm sorry about that.

Here's a screenshot to provide a visual demonstration:



This is the same for both RivaTuner and Precision. After all, it's the same program, but Precision has a different GUI and less options.


UPDATE:

I just finished testing 684/1476/999 for 30 minutes with ATITool's artifact scanning, and then exactly 20 minutes with Furmark's Stability Testing mode using 1440x900 Windowed with 4x MSAA. The maximum temperature during artifact scanning was about 67.5°C, and then 78°C with Furmark. The FPS readings in Furmark for the Min, Ave and Max were 49, 61 and 87 respectively.

Now I'm going to play Crysis with Precision open so I can monitor the temperature and hopefully the framerate as well. I changed its polling period to 2 seconds. Here I go!


UPDATE #2: The 684/1476/999 overclock failed in Crysis about 5 minutes ago. It took almost 3 hours and 20 minutes for it to happen, but Crysis simply quit on me, so I assume that the overclock was unstable. But now I'm experimenting with 666/1440/999.

__________________

TurboTurtle

Ooooh!

Ok, just misunderstood what you were saying. My bad.


What temps were you pushing in Crysis? I imagine it'd be rather high after 3 straight hours.

__________________