In the last video I covered the single card performance of the Nvidia GeForce GTX 1080 Ti. I then compared the performance to a single Titan X, and then 2 Titan X's in SLI. I also found in each instance that a single 1080 Ti was better than Titan Xs in SLI, and in some cases a single Titan X was better than a 2-way Titan X SLI. So is having a second GTX 1080 Ti for 2-way SLI worth the additional money? How well does it scale in SLI? Let's find out!
If you would like more information on my testing methods, click the card in the top right corner, or use the link in the description below. Just to recap, here's the information on the computer I'm using to benchmark these graphics cards.
"With that said, let's start by recapping the specs of Sledgehammer. Sledgehammer is built on an Asus Z10PE-D16 Workstation motherboard, with two 2.3GHz Intel Xeon E5-2696v3 18 core processors, with hyperthreading. These processors have the potential to Turbo up to 3.3GHz. Just for clarification, the 2696 is the OEM version of the 2699. It also has 256GB of error correcting DD4 RAM, boots from a 950 Pro SSD, and runs dual Maxwell Titan X's. Nearly anything you would like to know about it, is in the prior 5 videos of this series."
The first of our benchmarks is Ghost Recon - Wildlands. If you recall from the first video, this game is extremely demanding and actually scaled well in 1080p on the Titan X 2-way SLI. Unfortunately, on the 1080 Ti, 1080p performance actually decreased when using SLI, but had healthy gains in 4K. We see an average frame rate increase of 11.96 fps, but at 46.82 frames per second, we're still falling well below the desired 60 fps. This game is still very new, so with updates and optimization, these numbers should increase. And for the record, all of these tests were done using driver 378.78.
In Rise of the Tomb Raider, a single 1080 Ti was already beasting to amazing performance levels. It had an average fps of 120.57 at 1080p, and 63.13 at 4K, absolutely simply amazing for a single card. However, SLI performance did not scale so well. We see a significant drop in performance in 1080p, but a less than stellar increase at 4K. Keep in mind, these results actually mirror the Titan X results, so I'm not sure if a driver update or game update (if they're still releasing them) would make any difference.
Grand Theft Auto 5 actually scales the right way with SLI, with every increase in card spec and card numbers actually resulting in an increase of performance. While that's true, if the only game you play is GTA 5, and you play in 1080p, a single Titan X comes in at only 3.07 fps lower than 1080 Ti's in SLI and only .63 fps lower than a single 1080 Ti. In the case of this game, you might actually get better performance from a CPU clock speed increase. A single 1080 Ti is amazing in 1080p and 4K, and the meager gains of SLI is definitely not worth the additional cost.
Moving on to the actual benchmark software. In 3D Mark's Fire Strike and Fire Strike Ultra, which is optimized to scale properly with SLI, we see the gains we were expecting. A single 1080 Ti performed at nearly double the levels of the Titan X. As expected, there is a significant increase in SLI performance, with almost a 200% increase in 4K performance. If you recall from the single card benchmark video, the scores of the 1080 Ti were nearly identical to the scores of the Titan X's in 2-way SLI. The thing to take away from this chart is there was a 64% increase in performance going from a single Titan X to SLI in 4K, but on the 1080 Ti the performance increase was about 75%. In a perfect world, we'd want an increase of 100% when adding a second identical GPU, but here in the real world, we will take 75%. Simply amazing!. Before leaving this screen, I also want to point out the 1080 Ti SLI setup actually broke the 20,000 point mile stone. Keep in mind, these results are the average of 3 tests.
In our DirectX 12 test, Time Spy, the SLI setup actually performed worse than a single 1080 Ti. I honestly have no idea why this may be, but historically Nvidia's GPU's have struggled with DirectX 12. While not impressive, even the Titan X's scaled properly. This is definitely disappointing for the 1080 Ti, and hopefully this can be resolved with a driver update. The Time Spy scores reflect the same disappointing results as the frame rates, with the 1080 Ti SLI coming in at 6,156.67, under the single card's 6,175.67. That's only a .3% decrease in performance, but when discussing SLI there should never be a decrease if things are implemented properly.
Unigine's Heaven 4.0 is showing the same decrease in 1080p performance as the others, but an increase in 4K performance. While the 1080p performance decreased 43%, the 4K performance increased 43%. The scores for Heaven reflect the same disappointing results for 1080p, but show a healthy gain in 4K.
Unsurprisingly, our final benchmark, Unigine's Valley 1.0, shows the same odd decrease in 1080p, but in this case there was also a drop in 4K performance. A single card was able to average 79.4 fps, but only 62.07 in SLI. For the first time, there was a drop in 4K performance, bringing 4K numbers below the 60 fps desired minimum when the single card averaged 70 fps. The scores reflect the same disappointing results, with the 1080 Ti 2-way SLI coming in BELOW the Titan X SLI in terms of both 1080p and 4K. Absolutely ridiculous.
If I had to guess, I would say these disappointing results are probably the result of a new graphics card with new drivers, and a lack of SLI optimization. We should also remember the fact these new cards are also attempting to address a lot of concerns with VR. This is probably why 4K accelled in areas where 1080p took a performance hit. However, based on the numbers I've seen, I can in no way form or fashion, recommend a 1080 Ti SLI setup. Not for gaming, not on a standard monitor. Nvidia has expressed their desire to kill off SLI for gaming, and if that is true, this is definitely a step in the right direction. Sure there are performance gains with SLI if you're gaming in 4K, but unless you just have money to throw at your frame rate, those gains typically aren't enough to justify twice the price. While I will continue to use multiple cards, as this is a production rig first, and gaming second, I cannot recommend SLI to anyone else. Is SLI dead?
Benchmark Method Video - https://youtu.be/a0JiX0cdLa0