Monday Musing: DLSS 5 undoing somebody else work?

Back in March, Nvidia showed off DLSS 5. Time to muse about that. What is DLSS? It stands for Deep Learning Super Sampling. It’s a fancy way of saying machine learning based on high resolution images. It’s a bit of technology that is rendering frames in real time to complement GPU rendering real frames. These fake frames are added with true rendered frames. There is a little bit of overhead to boost performance too, along with the ability to upscale. Why would you want to do that? Well, it helps by boosting minimum, maximum, and average frames to be closer together. It just makes everything look smoother.

Yeah, this tech can do a decent amount. The most impressive thing is being able to upscale textures and the resolution of games. I need to explain the next part because it is important. At the moment, we need to get games playable first. Software can’t run on hardware if it just does not work. Some software is written with certain bits of hardware in mind and is trapped on it. Usually, solving that involves decompilation and recompilation. Programmers are doing this manually right now. They take old code, make a new source code, and port the game into it. It is time consuming and complex. Recompilation tries to avoid decompiling, but it still means making modern ports in C. You can still face issues trying to create a port that runs, even without reverse engineering everything. This is different from emulation, which tries to recreate original hardware and how it looked. But the downside there is the major performance hits and strange quirks it can or can’t recreate. The benefit of emulation is that you can make wide ranging systems, but if programmers did specific things to make it run well on the original hardware, that causes problems.

If you’re curious about remakes, they take the whole thing and start over. Remasters are usually decompiling along with recompiling things. Sometimes they take existing ports and update them to run on modern systems. It is why they use the same textures. Depending on the budget, a studio does a ton or not much. Look at the Metal Gear Solid Master Collection Vol. 1 and Vol. 2. Both used older remasters but ported them. For Vol. 2, they are actually making a new native port for the fourth game, along with an HD PSP port and a port of a Game Boy game. It requires time, energy, and a ton of skill to do. On the other hand, the Final Fantasy remakes were done from scratch and it shows, as they went in and made various changes. Two modern ones being 6 and 7.

That is something the Nvidia system does not usually do; it mostly just creates frames and can upscale. What makes this system so powerful is that unlike Nvidia’s lighting system RTX, which requires developers to opt in, this can be automatic and requires zero input. RTX is these days a rather performance heavy system. It still needs high end hardware, which leads to optimization struggles and different rendering solutions. Studios need to do the leg work to make it function, and when it is not being used by many, you can understand why they don’t.

What makes this system a bit problematic is the creation of frames causing things to change how they look, like changing the lighting or art style. This is what caused a fuss over DLSS 5. You can sort of understand why for retro games or newer games this may be a problem. All that work an artist has done is basically undone by a machine. That comes down to the sampling used and the failure of the system to understand exactly what is on screen. This new model moves beyond generating just frames. Now it adds lighting, textures, and materials. It redoes the whole thing. It is a bit like putting a filter on the Mona Lisa that makes it look like a cartoon.

Which means it is remaking not just frames but how something looks. It’s doing far more than the last system and offering a boost in performance. But the system is so demanding at the moment, requiring two GPUs to function in the demos. So it is not really a performance gain at the moment. That means it is not very optimized really, but it’s a look at where things are going. I would prefer the system to go in a different direction. Given how Nvidia rents out the ability to play games through the cloud, I get why they have done this. It’s a way for them to say, “Look, we can make games look better than you’re used to.” The direction I would like to see is them doing something about game preservation. I also understand that making older titles playable causes legal issues.

I did not fully understand why people had problems with this system until I bothered to read up on it. I would prefer if it made older titles more playable and could upscale or even provide textures. Right now we need volunteers to do the hard work, spending hours touching things up. I know Monday is meant to be for lighter posts but it is still worth reading about. The release date is sometime in Fall 2026 but it is limited to 5000 series cards. Making it more optimized is less flashy but honestly should have been the call. Considering the price, I have to wonder why are they doing this? The thing is, in theory, if Nvidia can reduce hardware overhead they can reduce the amount of resources a GPU needs, meaning they can sell more GPUs. But the thing is, it rarely works out like that.

Maybe DLSS 5 can do something, but right now it needs a ton of work. Applying a filter on games that changes how things look, well, that sounds awful.