Can next-gen Switch Pro really aim for 4K DLSS?

Can next-gen Switch Pro really aim for 4K DLSS?

Switch Pro, Switch 2, Super Switch, Switch Model S: it is not yet clear what the console will be called or when it will arrive, but rumors continue and a recurring topic associated with this possible new console is the use of DLSS technology by Nvidia, a deep learning-based super-sampling algorithm that allows you to combine traditional temporal sampling with a neural network to dramatically increase resolution and performance.

This technology has been used with great success on PC so far, and its job in this case would be to take the 720p resolution of the Switch tablet's screen and bring it up to 2160p to fit the now popular 4K screens in all our lounges. This at least in theory, but in practice could DLSS succeed in this feat on a device with mobile hardware? We have therefore decided to speak using numbers.

First of all, we have to talk about the plausibility or not that a processor with DLSS technology ends up on a portable console. Here are at least positive responses. The latest Nvidia Tegra SoC (codenamed Orin) is based on the latest Ampere architecture, and is designed for use primarily in the automotive sector. Here, however, the first problems arise, since the chip has an energy demand of 45W, when Switch has a power limit of 15W, of which 10-11W are used by the processor (probably halved in portable mode). But there is a solution to this problem: an ultra-low voltage version of the Orin chip that consumes only 5W, which should be able to scale up in performance, and therefore consumption, when connected to an external dock.

Could a Next Generation Switch Really Provide 4K DLSS? Here is our video analysis with Alex Battaglia.

Watch on YouTube. Even so, by integrating the tensor cores required to run the DLSS, we are still faced with a limited power reserve and AI upscaling is not 'free', hence our next step which is to measure the computational load required by the DLSS itself. In the video below you can see how we did it, but the calculation method is quite simple to explain. Using Doom Eternal as a basis, we employed an RTX 2060 to measure the time taken to process DLSS compared to native resolution rendering, concluding that the time cost for this process is 1.9ms. The RTX 2060 however has about 5.5x the deep learning power of the Orion chip operating at 10W, so assuming a linear proportion, DLSS would require 10.5ms on a probable Switch Pro. In a world where Doom Eternal is aiming for 16.7ms per frame, we are not within the time frame, it would strain. But if the game runs at 30fps with a frame-time of 33.3ms, then things get a lot more sustainable.

But it's important to point out that our measurements of DLSS cost performance are based on comparing inputs. and output, and we have no access to internal computing processes. As a result, our calculations are very approximate, but they help us get an idea of ​​the potential sustainability of this technology on portable hardware. And there are so many other variables to consider. For starters, we're taking 4K output for granted. There is nothing to prevent a developer from using DLSS to transform a 720p portable screen image into a 1440p one with DLSS, and then use the GPU scaler to achieve 4K output. In this scenario, the 10.5ms of time required for DLSS on Nvidia's Orion chip would take just half the time, or 5.2ms. There would undoubtedly be a degradation in quality, but it could be a great solution for some types of games.

And of course we are also basing our calculations on how DLSS works on PC, not necessarily Nvidia you do not implement a method of operation of its technology ad hoc for the console experience, just think that the company has even created low-level APIs specifically for Switch. But there is something even more important to ask ourselves: What is the visual result of a 720p image upscaled to 2160p via DLSS?

Compared with its internal resolution of 720p, 4K DLSS in ultra performance mode improves spatial resolution in still images. Compared to its internal resolution of 720p, 4K DLSS in ultra performance mode is less effective on the go, but convincing enough. Compared to native 4K, 4K DLSS in ultra performance mode loses detail and stability, but a possible next-gen Switch must offer decent images on a 4K screen. Compared to native 4K, 4K DLSS in ultra performance mode can exhibit temporal artifacts. It's not perfect, but the result is far better than any other upscaled 720p form. We hope the screenshots attached on this page perfectly illustrate why DLSS could change the game for a possible next-gen Switch and how an image made for a 720p tablet screen can transform and fit perfectly into a large 4K living room screen. . DLSS doesn't just upscale, it also applies anti-aliasing to the image. Basically it's a 2-in-1 solution. By combining the information extrapolated from the previous frames with the motion vectors that tell the algorithm where the pixels will possibly be positioned through a deep learning process, a transformative result is obtained.

A 4K DLSS image holds head to a native 4K presentation? In the PC universe, performance mode uses native 1080p input for scaling, which increases to 1440p in quality mode. By scaling from 720p there is less data to process and therefore there will be more inaccuracies and artifacts in the output image, but the point is that a probable next-gen Switch won't necessarily have to deliver 4K quality. The criteria for success are very different when it comes to comparing a PC experience viewed on a very close screen versus a console experience typically viewed on large TVs that are several feet away. The new Switch won't need to deliver the accuracy of 4K, it just needs to deliver an image that looks good on modern living room TVs.

In addition to DLSS itself, there are also several theories to explore. For one thing, why should Nintendo still aim for 720p for the screen of a possible next-gen Switch? From our point of view it represents the best ratio between pixel density and required GPU power, and in fact it is no coincidence that Valve has chosen 800p as the screen resolution of its Steam Deck. Going from 720p to 1080p would mean that much of the new Switch's power would be wasted on increasing resolution rather than increasing graphics quality.

Most Read Now

Magic The Gathering Innistrad: Midnight Hunt. Eurogamer.it presents you an exclusive card!

Humans take back Innistrad.

Tripwire in the blizzard: CEO John Gibson resigns after comments on Texas anti-abortion law

Tripwire Interactive boss leaves the company.

Alan Wake Remastered is reality and even has a launch window!

Alan Wake will also arrive on PlayStation consoles for the first time.

DLSS 2.2 is the latest version of Nvidia's machine learning-based upscaling technology. This video shows its application in Lego Builder's Journey and then discusses the latest innovations brought by DLSS 2.2 in other titles.

Watch on YouTube. The next big assumption we make is that Nintendo continues to rely on Nvidia for the next-gen Switch (but it would be madness not to, just for the compatibility factor) and that the company actually uses the Orin chip, or a variant thereof. On this last point, one of the most reliable leakers around, kopite7kimi, seems quite convinced. Even so, our calculations for a new Switch with a TDP of 10W seem plausible.

But the purpose of this test was basically centered on two points. The first was to figure out if an Nvidia chip based on the latest available architecture could actually handle DLSS, and the answer is positive. The doubt remains about the required power, since the tensor cores require power, but theoretically the console can increase the frequency of the chip in a docked configuration. So it is doable.

The second question concerns quality. Even with a simple 720p resolution to handle, DLSS produces good results at 4K, which surprise and enchant when placed side by side with the original input, even if they are not "as good" as native 4K.

The potential future applications of this technology are tempting, and not just for Nintendo consoles. Machine learning applications, which include super-sampling, are very sustainable going towards the enhanced mid-gen versions of the PS5 and Xbox Series which could be out in a few years, as long as Sony and Microsoft also choose this path for this new generation of console.







Powered by Blogger.