The NVIDIA GeForce GTX 980 Ti Review
by Ryan Smith on May 31, 2015 6:00 PM ESTOverclocking
Finally, no review of a high-end video card would be complete without a look at overclocking performance.
From a design standpoint, GTX 980 Ti already ships close to its power limits. NVIDIA’s 250W TDP can only be raised another 10% – to 275W – meaning that in TDP limited scenarios there’s not much headroom to play with. On the other hand with the stock voltage being relatively low, in clockspeed limited scenarios there’s still some room for pushing the performance envelope through overvolting. And neither of these options addresses the most potent aspect of overclocking, which is pushing the entire clockspeed curve higher at the same voltages by increasing the clockspeed offsets.
GTX Titan X by comparison ended up being a good overclocker, and while we'd expect GTX 980 Tis to use slightly lower quality chips as part of the binning process, it should still overclock rather well.
| GeForce GTX 980 Ti Overclocking | ||||
| Stock | Overclocked | |||
| Core Clock | 1000MHz | 1250MHz | ||
| Boost Clock | 1075Mhz | 1326MHz | ||
| Max Boost Clock | 1202MHz | 1477MHz | ||
| Memory Clock | 7GHz | 8GHz | ||
| Max Voltage | 1.187v | 1.23v | ||
Overall we're able to get another 250MHz (25%) out of the GTX 980 Ti's GPU, and another 1GHz (14%) out of its VRAM. This pushes the GTX 980 Ti's clockspeeds up to 1326MHz for the standard boost clock, and 1477MHz for the maximum boost clock. The card is heavily TDP limited at this point, so it's unlikely to sustain clockspeeds over 1400MHz, but working clockspeeds in the 1300MHz range are certainly sustainable. Meanwhile interestingly enough, this is actually a slightly better overclock than what we saw with the GTX Titan X; the Titan was only able to get another 200MHz out of its GPU and 800MHz out of its memory. So GTX 980 Ti ends up being the better overclocker by 50MHz.
The gains from this overclock are a very consistent across all 5 of our sample games at 4K, with the average performance increase being 20%. Though not quite enough to push the GTX 980 Ti above 60fps in Shadow of Mordor or Crysis 3, it is enough to crack 60fps on Battlefield 4 and The Talos Principle.
The cost of that 20% overclock in terms of power and noise is similarly straightforward. You're looking at an increased power cost of 30W or so at the wall – in-line with the 25W increase in the card’s TDP – while on the noise front the GTX 980 Ti is pushed out of its sweet spot. Card noise levels will increase by around 4.5dB(A).
Facebook
Twitter
RSS
290 Comments
View All Comments
Klimax - Tuesday, June 2, 2015 - link
Just small bug in your article:Page "GRID Autosport" has one paragraph from previous page.
"Switching out to another strategy game, even given Attila’s significant GPU requirements at higher settings, GTX 980 Ti still doesn’t falter. It trails GTX Titan X by just 2% at all settings."
As for theoretical pixel test with anomalous 15% drop from Titan X, there is ready explanation:
Under specific conditions there won't be enough power to push those two Raster engines with cut down blocks. (also only three paths instead of four)
Ryan Smith - Wednesday, June 3, 2015 - link
Fixed. Thanks for pointing that out.bdiddytampa - Tuesday, June 2, 2015 - link
Really great and thorough review as usual :-) Thanks Ryan!Hrobertgar - Tuesday, June 2, 2015 - link
Today, Alienware is offering 15" laptops with an option for an R9-390x. Their spec sheet isn't updated, nor could I find updated specs for anything other than R9-370 on AMD's own website. Are you going to review some of these R9-300 series cards anytime soon?Hrobertgar - Tuesday, June 2, 2015 - link
When I went to checkout (didn't actually buy - just checking schedule) it indicated 6-8 day shipping with the R9-390X.3DJF - Tuesday, June 2, 2015 - link
Ummm....$599 for R9 295X2?......where exactly? every search i have done for that card over the last 4 months up to today shows a LOWEST price of $619.Ryan Smith - Wednesday, June 3, 2015 - link
It is currently $599 after rebate over at Newegg.Casecutter - Tuesday, June 2, 2015 - link
From most result the 980Ti offer 20% more @1440p than a 980 (GM204) and given the 980Ti cost like 18-19% more that the orginal MSRP of the 980 ($550) It's really not any big thing.Given GM200 a 38% larger die, and 38% more SU's over a GM204 and you get 20% increase? It worse when a full TitanX is considered, that has 50% more SU's and the TitanX get perhaps 4% more in FpS over the 980Ti. This points to the fact that Maxwell doesn't scale. Looking at power the 980Ti is needing approx. 28% more power, which is not the worst but is starting to indicate there a losses as Nvidia scaled it up.
chizow - Tuesday, June 2, 2015 - link
Well, I guess its a good thing 980Ti isn't just 20% faster than the 980 then lol.CiccioB - Thursday, June 4, 2015 - link
This is obviously a comment by a frustrated AMD fan.Maxwell scales perfectly as you didn't consider the frequency it runs.
GM200 is 50% more than a GM204 in all resources. But those GPU run at about 0.86% of GM204 frequency (1250 vs 1075). If you can do simple math, you'll see that for any 980 results, if you multiply it by 1.5 and then for 0.86 (or directly for 1.3, that means 30% more) you'll find almost exactly the numbers the 980Ti bench shows.
Now that the new 980 $500 price, do the same and... yes, it is $650 for 980Ti.
Oh, the die size... let's see... 398mm^2of GM204 * 1.5 = 597mm^2 which compares almost exactly with the calculated 601m^2 of GM200.
Pretty simply. It shows everything scales perfectly in nvidia house. Seen custom cards are coming, we'll see GM200 going to 50% more than GM204 at same frequency. Yet these cards will consume a bit more, as expected.
You cannot say the same for AMD architecture though, as with smaller chips GCN is somewhat on par or even better with respect to nvidia for perf/mm^2, but as soon as real crunching power is requested GCN becomes extremely inefficient under the point of both perf/Watt or perf/mm^2.
If you tried to plant a doubt about the quality of this GM200 or Maxwell architecture in general, sorry, you choose the wrong architecture/chip/method. You simply failed.