introduced an impressive bit of new technology back in 2014
called DSR or Dynamic Super Resolution. Essentially it allowed players with 1080p or 1440p displays to take advantage of unused GPU power and enjoy the benefits of rendering games at a higher resolution, like 4K, without the need to invest in a new screen. The driver-based NVIDIA Control Panel feature lets you set a scale, like 4.0x, which then unlocks that higher resolution inside a game’s settings menu. Being a driver-based update there also wasn’t a need for game-specific integration, it worked (and still does) with everything.
But why would someone with a high refresh-rate 1080p display want to render a game at 1440p or 4K? Well, the answer to that comes down to the benefits of Downsampling or Super Sampling, which takes a more detailed image and shrinks that down to fit a lower resolution.
With DSR the results are immediate, rendering at a higher resolution offers up more detailed textures in characters and environments, less flickering, fewer jagged-edges on fine lines and smaller objects; even when displaying at 1080p. Better shadow quality too, as that’s an effect where quality is often dictated by a higher resolution. There’s increased detail in background objects and distant text, the list goes on. A game changer and something that has been a go-to for many GeForce GTX
and GeForce RTX
players for years.
Introducing DLDSR… It’s Like DLSS in Reverse
probably needs no introduction; it leverages the AI-based Tensor cores of the GeForce RTX range as a form of super-powered upscaling. Its full title being Deep Learning Super Sampling, where AI is used to render an image comparable to the outright native counterpart, with a massive boost to performance. And by massive we’re talking about a generational-style leap, the sort of thing that makes 4K ray-traced versions of Cyberpunk 2077
, Dying Light 2 Stay Human
, Metro Exodus
, and more only possible on a modern PC.
Comparing DLDSR to DLSS, and saying like it’s that but in reverse, is probably the easiest way to understand what it does.
It pulls this off by using advanced machine learning and a lower internal resolution to basically fill-in all of that extra detail based on an AI network's countless hours spent learning and understanding how a game should look in motion. Of course, there’s a lot more technical wizardry to it than that, but you get the idea.
And this brings us to the one draw-back that comes from using DSR in its current version, there’s a pretty sizable performance hit that comes from rendering a game in native 4K and then outputting that onto, say, a 1080p display. One that can diminish the benefit that comes from the improved image quality - or limiting the options of when to turn it on strictly to older games. Comparing DLDSR to DLSS, and saying like it’s that but in reverse, is probably the easiest way to understand what it does. It takes a higher resolution than that of 1080p and 1440p (2.25x being the DLDSR sweet spot) and then presents image quality on par with standard DSR at 4.0x (for 1080p output that’s 4K).
With less of a hit to performance, as seen with DLSS, the DL part of DLDSR is all about Deep Learning and taking advantage of the AI-based Tensor cores of the GeForce RTX range to do some of the heavy lifting.
With less of a hit to performance, as seen with DLSS, the DL part of DLDSR is all about Deep Learning and taking advantage of the AI-based Tensor cores of the GeForce RTX range to do some of the heavy lifting. Here the AI network for DLDSR was “trained to recognize lines and edges to discriminate between aliasing and texture detail”.
Where DSR was all about using all GPU power left unused to improve image quality, DLDSR is about leveraging cutting-edge RTX technologies to do the same but with less of a hit to performance. The results are impressive to say the least, with DLDSR offering 4K-like improvements without the giant hit to the frame-rate that comes from natively rendering a game in 4K. It also opens the possibility of enabling DSR for more games for GeForce RTX players.
From Cyberpunk to Rainbows
60 frames-per-second is pretty much the new baseline when it comes to in-game performance, and this means that DLDSR works great on titles that are less taxing on the GPU; older efforts like Prey
and Diablo III
are prime examples. But, if you’re someone that’s rocking a high-end GeForce RTX card and sporting a high refresh-rate 1080p or 1440p display, well, there’s a definite benefit too.
We decided to test DLDSR on two modern games on both sides of this spectrum - the visually stunning Cyberpunk 2077
and the competitive focused high performer that is Rainbow Six Extraction
. We also chose them as case studies partly because they fully support DLSS, so it’s interesting to see the overall difference that DLDSR brings when going the other way.
The good news is that DLDSR works pretty much exactly like DSR in that it’s accessed via the NVIDIA Control Panel (under Manage 3D Settings), where you can easily select DL scaling or Legacy scaling options and an overall image smoothing value. Tick a few boxes and these new resolutions will appear in-game, no matter the game.
If you’ve used a 1440p or 4K display, no doubt you’ve noticed the difference when jumping back to 1080p. This is not a slight on the resolution, a crisp 1080p display is still wonderful for smaller displays - but seeing a crisper image with more detail is something you immediately pick up on. DLDSR delivers that.
The immediate benefit of DLDSR can be seen across both Cyberpunk 2077 and Rainbow Six Extraction, in the former’s benchmark that takes place at the El Coyote Bar the bartender himself is more detailed and patrons feature noticeably less shimmer with DLDSR. The overhead lamp looks crisper too. All in all, it’s a game-changing transformation.
Best of all, for as demanding as Cyberpunk 2077 can be on modern hardware the performance improvement compared to standard brute-force DSR was over 25% in a like for like comparison. Impressive, because much like with Cyberpunk 2077’s excellent use of DLSS the image quality remained exceptional.
In Rainbow Six Extraction’s urban settings, hard edges, and fine texture detail make-up a lot of what you see that isn’t a creepy alien trying to eradicate humanity. One of the game’s co-op Incursions takes place in a museum and it’s here where the DLDSR benefits on a 1080p display are also readily apparent. Edges are no longer jaggy, textures have finer detail, and all manner of objects both large and small look crisper and more detailed. The effect, again, is impressive, especially in motion. And with Rainbow Six Extraction the performance difference between DLDSR and DSR sits in the 30-40% region in favour of Deep Learning.
The AI Future
Both DLSS and DLDSR solve problems, and both prove that raw-power isn’t the be all and end all when it comes to performance. And with the wide range of GeForce RTX GPUs and 1080p and 1440p displays out there, having a cutting-edge downsampling tool available makes sense. Using AI to improve performance is something that will continue to be a cornerstone of gaming tech moving forward, with DLDSR providing more proof of the benefits of AI.
And with the wide range of GeForce RTX GPUs and 1080p and 1440p displays out there, having a cutting-edge downsampling tool available makes sense.
In fact, it effectively opens the concept of Super Resolution scaling to a wider range of gamers, no matter if they’re rocking a GeForce RTX 2060
or a GeForce RTX 3080
. With performance and improved image quality going together, chalk it up as another great reason to go RTX On.