After Nvidia drummed up a whole lot of hype for DLSS (Deep Learning Super Sampling) in the six months since the GeForce RTX 20 series launched, the first few games to support it are now with us. Both Battlefield V and Metro Exodus now support DLSS, although the reception has been mixed, to say the least. Both games exhibit fairly blurry image quality, particularly at lower resolutions.
DLSS is a method utilising the RTX technology and Tensor Cores on the Turing GPUs to improve frame rates while upsampling to higher resolutions, in theory allowing cards to use higher visual settings than they otherwise would. Nvidia DLSS works by training a supercomputer to analyse aliased images from a game, eventually learning where the jagged edges occur and then automatically generating the correct pixels rather than applying antialiasing techniques. The aim for Nvidia is to replicate the perfect image as closely as possible.
It’s all very good in theory, although the results aren’t exactly bearing great fruit. This is because the quality of DLSS will vary greatly depending on a number of factors, including the time Nvidia has spent training its AI, the complexity of the content, and the game engine itself. Now that Battlefield V and Metro Exodus have DLSS support, Nvidia’s supercomputer will continue to improve the image quality and any boosts will arrive as driver updates.
Andrew Edelsten, technical director of deep learning at Nvidia, has stepped in to try to answer a few questions about the tech. He led the team which helped to create DLSS in the first place.
First and foremost, DLSS will not benefit you if you already have high frame rates. We saw in our RTX benchmarks, for example, that DLSS provides no noticeable performance boost in Metro Exodus on the RTX 2060 when playing at the High graphics preset without ray-tracing. Enable ray-tracing though and we begin to see the frame rate gains. The benefits of DLSS become apparent at low frame rates and higher resolutions.
“DLSS requires a fixed amount of GPU time per frame to run the deep neural network”, says Edelsten. “Thus, games that run at lower frame rates (proportionally less fixed workload) or higher resolutions (greater pixel shading savings), benefit more from DLSS. For games running at high frame rates or low resolutions, DLSS may not boost performance. When your GPU’s frame rendering time is shorter than what it takes to execute the DLSS model, we don’t enable DLSS. We only enable DLSS for cases where you will receive a performance gain. DLSS availability is game-specific, and depends on your GPU and selected display resolution.”
Going forward, Edelsten explains that DLSS is an emerging technology and it will improve. At first, it’s being focused on high resolutions, with 4K resolution being the training target for its machine learning. If you play Metro Exodus or BFV at lower resolutions with DLSS enabled, this will mean more unreliable image quality. This is likely the primary reason why Metro Exodus looks very blurry in our comparison shots, for example, which were pulled at 1080p.
In terms of Battlefield V, Nvidia sounds as if it’s pretty happy with the 4K and 1440p performance of DLSS. They reckon it delivers up to 40% performance gains with good image quality. The next update to Battlefield V will be tailored towards image quality at 1080p and 3440x1440 ultrawide.
Metro Exodus is a little flakier. A day one patch was issued which improved DLSS sharpness, although we still found the image quality a little shaky in our testing. Nvidia will be training its DLSS across a larger section of the game next, which it hopes will result in a further boost to image quality. Fingers crossed it all works out.