Overclocking on the new Kepler architecture (on which the nVidia 600 series is based) is a little different to how GPU overclocking has been done in the past. With previous nVidia cards you could set it, get it stable and then forget it, you also need to factor in the boost clock.
To overclock the card I decided to use MSI Afterburner; it’s a tool that I’ve found effective and feature packed with previous cards, so the choice, for me at least, is easy. There are other software options you can use that may do the job as well as afterburner, such as EVGA’s Precision X and Sapphire’s TriXX.
Once open, I enabled voltage control and monitoring in the settings menu.
Then the fun began.
Knowing that the GTX 680 has its voltage mostly locked, I maxed out the voltage setting and TDP overhead (power limit.) If you’re using this as a guide, please know about your card before you decide to throw the maximum voltage limit at it; it’s very likely you’ll cook your GPU chip – and no, they don’t cover frying it in their warranties.
I was able to throw 176mhz into the core which brought the core up to 1300Mhz, which is awesome. However, the moment I opened MSI Kombustor to test the overclock, once the boost clock kicked in, the display driver had a bit of a heart attack and crashed the software. So it was time to draw the overclock back. At +113 Mhz Kombustor would run fine, boost clock and all. I attempted to run 3DMark11. No go. In fact, I had to scale back to +61 on the core clock in order to get 3DMark to run and clear.
Initially I was disappointed that I couldn’t add much to the core at all; but then I remembered this card is already overclocked. It’s stock at 1120 Mhz on the core. So really I was bringing the core to 1185 Mhz.. which made the boost clock 1250 Mhz. For a card that isn’t actually designed for overclocking, this is a nice speed.
So we’ve hit our “sweet spot” for core speed, what about the memory?
I put 125Mhz in it for an initial bump. 3DMark11 cleared fine. 250Mhz? Also cleared fine. 350Mhz!? You bet. From this point I went in 25 Mhz increments on the memory clock and didn’t get an artifact until 477 Mhz. So I brought the memory clock down to +452 Mhz. The effective memory clock at stock is 6.2 GBPS. At this speed, the effective memory clock is 7.2 GBPS. That’s a 1 GBPS increase. Excellent!
How do these numbers translate into extra performance? Lets find out on the next page..
After we completed our overclocking, we decided to run some benchmarks at the overclocked speeds to translate the figures into performance. We’ve provided overclocked results side-by-side with the stock results, to show just how much performance you can squeeze out of the inno3d iChill Black Series GTX 680.
The inno3d iChiLL Black Series GTX 680 has a rather special cooler on it from Arctic Cooling. It’s a closed loop cooler and it really does perform as well as the packaging and marketing media claim. At stock settings, we never once broke 44C. Not once.
So, with 100mV thrown into the core and a 135% TDP, did the cooler continue to perform? You bet. During all of our testing, stability checks and benchmarking, we never once topped out over 56C.
..and the whole time I couldn’t hear a thing. It was very pleasant.