That's right, SLI — Nvidia's obsolete multi-GPU tech that Benchmark Labs managed to run using some unknown SLI-enablement magic. The channel was able to run SLI decently well in several DX11, DX12, and Vulkan titles, but ultimately the 3090s couldn't beat the RTX 4090 (though they got close).
Before we jump to the results, it's worth noting that all of this should be taken with a grain (or several) of salt. SLI has been unsupported for years at this point, and getting it to work with most modern titles — especially with DX12 and Vulkan — would require...unorthodox methods. We're not sure how Benchmark Lab managed to enable SLI in most of the games tested — again, specifically the DX12 versions — or if the results were doctored or extrapolated. That said, the results are pretty interesting (if they are, indeed, accurate).
Benchmark Lab tested 9 games: Spider-Man Remastered, Cyberpunk 2077, Watch Dogs Legion, Microsoft Flight Simulator 2020, Minecraft RTX, God of War, Witcher 3, Horizon Zero Dawn, and Read Dead Redemption 2. Testing was performed at 4k resolution using a variety of different quality settings and DLSS modes.
In Spider-Man Remastered, with maxed out settings (Ray Tracing and DLSS Balanced mode enabled), the RTX 3090s in SLI achieved 80 - 85 frames per second (fps) on average, while the RTX 4090 managed a noticeably higher 95 fps average.
In Cyberpunk 2077, with maxed out settings (Ray Tracing and DLSS Balanced mode enabled), the RTX 3090s in SLI managed 50 fps while the RTX 4090 averaged 70 fps.
In Watch Dogs: Legion, with maxed out settings (Ray Tracing and DLSS Quality mode enabled) the gap closed a little: the RTX 3090s in SLI averaged 73 fps, while the RTX 4090 averaged 80 fps.
The results were even closer in Microsoft Flight Simulator with maxed out settings (Ray Tracing and DLSS Quality mode enabled). The RTX 3090s in SLI averaged 80 fps, while the RTX 4090 averaged 83 - 85 fps. (However, note that Microsoft Flight Simulator is known to be CPU-bound, especially at higher frame rates.)
In Minecraft RTX, with max settings (Ray Tracing enabled), the RTX 3090s in SLI averaged 70 fps while the RTX 4090 hit a slightly higher average of 75 fps. And in God of War at max settings (DLSS Quality mode enabled), the RTX 3090s in SLI hit an average of 103 fps while the RTX 4090 did quite a bit better with an average of 120 fps.
For the rest of the titles, check out the video below:
RTX 3090 SLI Is The Only Setup That Can Get Close To The RTX 4090
With the exception of the Cyberpunk 2077 results, the RTX 3090s in SLI were around 8 - 15% slower than the RTX 4090. For an SLI implementation, this isn't bad at all — and it's the only GPU configuration we've seen so far that can get close to the RTX 4090's performance bracket. For some perspective, the next-closest GPU is the RTX 3090 Ti — which trails the 4090 by over 50% based on our tests.
Sadly, the 3090s in SLI setup wasn't being used to its full potential, as the secondary 3090 consistently maxed out at 45% utilization. This is one of the pitfalls of SLI setups, in which bad optimization leads to less-than-superb GPU utilization on the secondary card. In theory, if we could get anywhere close to 100% utilization on both GPUs, we would probably see the 3090s in SLI outperform the 4090 by a good margin.
But the chances of that happening are extremely unlikely, since Nvidia officially killed off SLI in the form of physical bridges with the RTX 40 series. SLI support has dwindled so much over the past several years that it's basically only useful in synthetic benchmarks at this point.
There are technically ways of enabling SLI (unofficially) in games that don't support it — like in the case of the benchmarks seen in this video. But these results are unpredictable at best, and usually result in system instability or severe micro-stuttering problems while gaming.
There is some hope for SLI, however. Multi-GPU workloads are very common in the enterprise space, and Nvidia even has multi-GPU technologies that don't require an NVLink or SLI bridge.
On the gaming side, modern APIs such as DX12 and Vulkan do have the capability to render frames to two completely different GPUs in tandem. So there's a chance multi-GPU tech might make its way back to the gaming space, eventually (but whether developers want to support multi-GPU technology for gaming is a different story).