The NVIDIA GeForce GTX 980 Ti Review
by Ryan Smith on May 31, 2015 6:00 PM ESTGrand Theft Auto V
The final game in our review of the GTX 980 Ti is our most recent addition, Grand Theft Auto V. The latest edition of Rockstar’s venerable series of open world action games, Grand Theft Auto V was originally released to the last-gen consoles back in 2013. However thanks to a rather significant facelift for the current-gen consoles and PCs, along with the ability to greatly turn up rendering distances and add other features like MSAA and more realistic shadows, the end result is a game that is still among the most stressful of our benchmarks when all of its features are turned up. Furthermore, in a move rather uncharacteristic of most open world action games, Grand Theft Auto also includes a very comprehensive benchmark mode, giving us a great chance to look into the performance of an open world action game.
On a quick note about settings, as Grand Theft Auto V doesn't have pre-defined settings tiers, I want to quickly note what settings we're using. For "Very High" quality we have all of the primary graphics settings turned up to their highest setting, with the exception of grass, which is at its own very high setting. Meanwhile 4x MSAA is enabled for direct views and reflections. This setting also involves turning on some of the advanced redering features - the game's long shadows, high resolution shadows, and high definition flight streaming - but it not increasing the view distance any further.
Otherwise for "High" quality we take the same basic settings but turn off all MSAA, which significantly reduces the GPU rendering and VRAM requirements.
After initially expecting Grand Theft Auto to be a walk in the park performance wise, the PC version of the game has instead turned out to be a very demanding games for our GPUs. Even at 1440p we can’t have very high quality with MSAA and still crack 60fps, though we can get very close.
Ultimately GTA doesn’t do any better than any other game in setting apart our GM200 cards. GTX 980 Ti trails GTX Titan by 4% or less, essentially the average outcome at this point. Also average is the GTX 980 Ti’s lead over the GTX 980, with the newest card beating the older GTX 980 by 29-31% across our three settings. Finally, against the GTX 780 the GTX 980 Ti has another strong showing, with a 69-79% lead.
On an absolute basis we can see that at 4K we can’t have 4x MSAA and even crack 30fps on a single-GPU card, with GTX 980 Ti topping out at 27.8 fps. Taking out MSAA brings us up to 46.2fps, which is still well off 60fps, but also well over the 30fps cap that this game was originally designed against on the last-generation consoles.
Along with an all-around solid benchmark scene, the other interesting benchmarking feature of GTA is that it also generates frame percentiles on its own, allowing us to see the percentiles without going back and recording the game with FRAPS. Taking a look at the 99th percentile in this case, what we find is that at each setting GTA crushes some group of cards due to a lack of VRAM.
At 4K very high quality, 4GB cards have just enough VRAM to stay alive, with the multi-GPU R9 295X2 getting crushed due to the additional VRAM requirements of AFR pushing it over the edge. Not plotted here are the 3GB cards, which saw their framerates plummet to the low single-digits, essentially struggling to complete this benchmark. Meanwhile 1440p at high quality crushes our 2GB cards, with less VRAM than a Radeon HD 7970 falling off of the cliff.
As for what this means for the GTX 980 Ti, the situation finds the GTX 980 Ti trailing the GTX Titan X in 99th percentile framerates by anywhere between 3% and 10%. This test is not designed to push more than 6GB of VRAM, so I’m not entirely convinced this isn’t a wider than normal variance (especially at the low framerates for 4K), though the significant and rapid asset streaming this benchmark requires may be taking its toll on the GTX 980 Ti, which has less VRAM for additional caching.
Facebook
Twitter
RSS
290 Comments
View All Comments
FlushedBubblyJock - Wednesday, June 10, 2015 - link
I bought a bunch of G80 G92 G92b and G94 nvidia cards because you could purchase memory size, bandwidth, bit width, power connector config, essentially any speed at any price point for a gamers rig, install the same driver, change the cards easily, upgrade for your customers without hassles...IT WAS A GOLD MINE OF FLEXIBILITY
What happened was, the amd fanboys got very angry over the IMMENSE SUCCESS of the initial G80 and it's reworked cores and totally fluid memory, card size, bit width, and pricing configurations... so they HAD TO TRY TO BRING IT DOWN...
Thus AMD launched their PR war, and the clueless amd fan launched their endless lies.
I'll tell you this much, no on would trade me a 9800GTX for a 9800GT
I couldn't get the 92 bit width cards for the same price as the 128 bit
DDR2 and DDR3 also differentiated the stack massively.
What we had wasn't rebranding, but an amazingly flexible GPU core that stood roaring above at the top and could be CUT down to the middle and the low gaming end, an configured successfully with loads of different bit widths and memory configs....
64 bit width, 92, 128, 256, 384, 192, ETC...
That was an is a awesome core, period.
BillyONeal - Sunday, May 31, 2015 - link
And people have been bent out of shape about it. For "YEARS" :)dragonsqrrl - Sunday, May 31, 2015 - link
Their highest-end rebadge, the 390X, will likely compete with the 980, not the 980 Ti. The 980 Ti will be closer to Fiji's performance profile.austinsguitar - Sunday, May 31, 2015 - link
I dont think you realize how much more efficiant this card is even compared to past cards for its nm and performance. This is a feat. Just calm down and enjoy. I am very happy that the cards price us perfect. :) thanks nvidiaMapRef41N93W - Sunday, May 31, 2015 - link
Maybe you aren't aware of how silicon works, but this a 601mm^2 die which costs a boat load to produce especially with the rising costs of crystalline silicon dies. Being on 28nm this long just means the yields are higher (which is why a 601mm^2 is even possible).You aren't going to see a 14nm card that outperforms this by much till 2017 at the earliest which following the recent NVIDIA trends should see the Titan XYZ (whatever they want to call it) which should be a pretty huge jump at a pretty high price.
Thomas_K - Monday, June 1, 2015 - link
Actually AMD is doing 14nm starting next yearhttp://www.guru3d.com/news-story/it-is-official-am...
"Although this was a rumor for a long time now we now know that AMD skips 20nm and jumps onto a 14nm fabrication node for their 2016 GPUs."
dragonsqrrl - Sunday, May 31, 2015 - link
Not sure I understand your comment, 28nm is precisely why we're paying this much for this level of performance in 2015... But it's also pretty impressive for the same reason.Azix - Sunday, May 31, 2015 - link
14/16nm might cost more. 28nm should have better yields and lower cost. These chips do not cost much to make at all (retail price could be 2-3 times the chip cost)dragonsqrrl - Sunday, May 31, 2015 - link
I think you misinterpreted my comment. I was responding to someone who seemed shocked by the fact that price/performance ratios aren't improving dramatically despite the fact that we're on a very mature process. In response I said the fact that we're on the same process is precisely why we aren't seeing dramatic improvements in price/performance ratios."28nm should have better yields and lower cost. These chips do not cost much to make at all (retail price could be 2-3 times the chip cost)"
Yields are just one part of the equation. Die size also plays a significant role in manufacturing costs. The fact that your trying to say with a straight face that GM200 does not cost much to make says more than your written comment itself.
zepi - Monday, June 1, 2015 - link
Assuming perfect scaling 600mm2 28nm chip would shrink to 150mm2 at 14nm.GM107 is a 148mm2 chip, so basically this "monster" with just a dieshrink would find a nice place for itself at the bottom end of Nvidias lineup with after transition to 14nm.
This does not take into account the fact that at 14nm and 150mm2 they couldn't give it enough memory bandwidth so easily, but just tells you something about how significant the reduction in size and manifacturing cost is after the initial ramp-up of the yields.