20% More Performance? Overclocking the NVIDIA GeForce GTX 980 Ti Under HG10

By Dustin Sklavos, on July 20, 2015

It goes without saying that NVIDIA’s Maxwell architecture is incredibly efficient and extremely powerful, and that this performance is realized most effectively in the GeForce GTX TITAN X and the newly released GeForce GTX 980 Ti. We’ve already overclocked and tested the GM200-based TITAN X and the GM204-based GTX 980, and now it’s time to take a look at what happens when we strap an HG10-N980 and Hydro Series H75 cooler to a reference 980 Ti.

People who remember the GeForce GTX 780 are going to have a big of déjà vu with the 980 Ti. The 780 was a cut down GeForce GTX TITAN that was able to meet and beat the better-enabled chip when overclocked, and the same is true of the 980 Ti and TITAN X. The GM200 powering the TITAN X is a monstrous chip, with 3072 CUDA cores, 96 ROPs, and 12GB of GDDR5 on a 384-bit memory bus benefiting from the substantial improvements in compression that GM204 and the GeForce GTX 980 brought with it. Modern graphics cards are intensely heavy with shader hardware; AMD’s recently released Fury and Fury X have freakishly high shader counts and correspondingly high shader performance.


Source: AnandTech

Historically, though, cut-down variants like the 980 Ti can have at least as much going for them if not more than the fully enabled parts. Modern graphics cards are intensely heavy with shader hardware; AMD’s recently released Fury and Fury X have freakishly high shader counts and correspondingly high shader performance. The same is true of the GM200, so when the 980 Ti gets two of the GM200’s 24 shader clusters fused off, it’s an extremely safe cut to make. Going down from 12GB of GDDR5 to 6GB is an equally smart move. Maybe the unkindest cut is the 16 texturing units that get disabled with the two shader clusters, but that’s substantially ameliorated by the 980 Ti’s higher stock clock speed.

The flipside of a cut-down chip is that it can use less power and typically will generate less heat. This was part of what made the GTX 780 such a great overclocker; under water it had a very hard time actually hitting its TDP limit. My pair actually hit the limits of the voltage and silicon before the TDP. What makes the 980 Ti such a compelling overclocker is that it’s easier to get 6GB of GDDR5 to run at a high speed than it is to get 12GB (especially when cooling with an HG10-N980), and it’s easier to get 2,816 CUDA cores to run at a higher speed than 3,072.

I found in our testing that unlike the 780 but like the GM204-based 980, you’re largely limited by the 980 Ti’s TDP limit, and that makes liquid cooling it that much more important. We’ve established that more efficient cooling and the resulting lower temperatures mean the hardware requires less power to operate.

The testbed used will be featured in an upcoming build log, but the core components you need to be aware of to examine these test results are:

  • CPU: Intel Core i7-5960X @ 4.4GHz, cooled with a Hydro Series H100i GTX
  • DRAM: 16GB (2x8GB) Dominator Platinum DDR4-2666
  • Motherboard: ASRock X99E-ITX
  • OS: Windows 10 Insider Preview Build 10165

The GeForce GTX 980 Ti we’re using is a reference model from Gigabyte, and we’re using a prototype HG10-N980 to cool it along with an H75 cooler.

 

Stock

Overclocked

Peak GPU Clock

1215MHz

~1500MHz

GDDR5 Clock

7GHz

7.9GHz

When overclocking, I raised the TDP limit to 110% and the voltage by the maximum allowed 87mV. You can see these are pretty massive overclocks and I’m not convinced the 980 Ti can’t go higher still. The GPU core clock is up 23.5% while the memory gets another 12.9%. It’s important to note that the TDP limit has consistently been the bottleneck on our overclock, though; we peak at 1.5GHz but clocks typically hover around one or two bins down.

So what does our beastly overclock earn us? We tested four demanding games at two resolutions: 3440x1440 for comparison purposes in a future blog, and 4K as the target for single GPU performance. Games had all of their settings but anti-aliasing maxed out, excepting Grand Theft Auto V, which we trimmed back slightly.

At 3440x1440, the performance differences are huge. The 980 Ti essentially jumps another class. Grand Theft Auto V’s average sails past 60fps, while Shadow of Mordor becomes much more playable.

When we move to 4K, we find the extra performance even more vital, and every game gets a healthy performance jump. BioShock: Infinite breaks 60fps and Shadow of Mordor easily breaks 40fps.

If we break down the performance difference between a stock 980 Ti and one heavily overclocked under an HG10-N980, it’s pretty staggering. As I mentioned before, this is more or less the kind of performance difference that characterizes the gulfs between classes of cards. And it’s worth mentioning that during the entire duration of the testing, the 980 Ti’s GPU temperature never exceeded 61C.

The fact is, NVIDIA’s GeForce 980 Ti is a beast with a tremendous amount of gas left in the tank. More than either TITAN X or the GTX 980, the 980 Ti seems to be the ideal mate for our upcoming HG10-N980 bracket.


Comments