NVIDIA's New DX11 Drivers Promise Huge SLI Performance Benefits, Assault AMD's Mantle

At GDC a few weeks ago, NVIDIA teased a new set of drivers that it promised would offer huge scaling benefits for both single-GPU and SLI configurations. The new driver, 337.50 Beta, drops today -- so what kind of gains can users expect? If you own a high-end NVIDIA card, apparently quite a bit, especially if you game with multiple GPUs in SLI. NVIDIA is predicting some huge jumps for its high-end range with scaling up 50-70% in some cases.

These are an impressive set of gains for NVIDIA's top-end cards, particularly the GTX 770, which remains rather more affordable than the top-end GTX 780 Ti. NVIDIA's own slide deck claims "up to 64% faster single GPU performance," but none of the slides the company included show gains anywhere near that high in actual game tests.

While the label claims that "GeForce + DirectX 11 is 12% faster", the actual benefit from the new driver in a single-GPU configuration is much smaller; NVIDIA shows the GTX 780 Ti moving from 1.16x the Radeon R9 290X's performance up to perhaps 1.19x. That's a gain of perhaps 2.5% -- the 12% is derived from the GTX 780 Ti's final score against the Radeon R9 290X's Mantle performance.

But if you're reading quickly all you see is "12%!  Awesome!"

Still, NVIDIA has obviously been able to wring additional performance out of DX11 titles with recent driver updates. To give you an example, when the GeForce GTX 780 Ti launched a few months back, the card put up 63.3 FPS in the Sleeping Dogs benchmark at a resolution of 1920x1200, when using the 331.70 drivers. With the 337.50 drives, our own internal testing shows the frame rate jumps to 67.8 FPS in Sleeping Dogs. At the same resolution Bioshock Infinite's performance increased from 97.88 FPS to 106.06 FPS. And Crysis 3's performance jumped from 40.96 FPS to 48.98--increases of 7.1%, 8.3%, and 19.6%, respectively.

Why Focus On SLI?

Despite the claims around huge single-GPU performance boosts, I'd have been honestly surprised if the company was able to deliver meteoric increases strictly from driver-level optimizations in shipping titles. First, if giving DX11 a 15-25% kick in the pants in most configurations was as simple as a driver update, both NVIDIA and AMD would've tapped that performance long ago. That's doubly true for AMD, which went to the trouble of creating its own API rather than simply iterating to improve the existing DX11 API.

Microsoft's own DX12 (which, according to our sources, is similar-to-but-distinct-from Mantle in multiple areas) isn't a project that the company threw together on a lark. Again, if squeezing DX12-like performance out of DX11 was realistically possible, then MS would've done that to start with.

The second reason for focusing on SLI improvements is that multi-GPU configurations are the area where Mantle shines the most. Unfortunately they're also a bit trickier to implement, but developers that put the time in to optimizing a title can deliver startling performance jumps between DX11 and Mantle. NVIDIA, therefore, has good reason to target them.

In fairness to AMD, NVIDIA's install base numbers here are slanted, either because the company compiled this list awhile back or because it missed a few entries. We spot-checked AMD's market share using the current version of the Steam Hardware Survey and found that an estimated 14.1% of the GPUs on the market can use Mantle. Of course, this is where we jump back in to "fun with percentages." AMD would argue that the Mantle market is 28% larger than NVIDIA made it look. NVIDIA would point out that 14% ain't that much bigger than 11% in the grand scheme of things.

Expect both companies to hammer home these points from now until DirectX 12 launches in late 2015. NVIDIA will point to unilateral DirectX 12 support across Kepler, Fermi, and Maxwell--and they should. AMD will argue that Mantle is actually available and shipping for games right now, while NVIDIA will claim that its fully committed to giving its customers DX12 on any modern GPU.  AMD will retort that come 2015, precious few gamers will actually be using 2010-era hardware. Given that we're going to see a mutual push for 4K from both companies over the next 12-18 months, I suspect backwards compatibility just won't be worth that much in the long run -- we're on the cusp of a new resolution, and the majority of gamers will want to have Kepler / GCN-hardware or better for making the jump.

Regardless of whether you're on Team Red or Green, you can look forward to a long discussion of which manufacturer is doing a better job of bringing value to gamers as more Mantle titles roll out this year and as Microsoft finalizes its own plans for whether DX12 will come to Windows 7 and Windows 8, or be reserved as a future Windows exclusive.