Nintendo has updated their patent on upscaling that names a specific technology that could either evoke excitement or shock, even fear, if it does get used on the Switch 2.
As Mike Odyssey shared on Twitter, the patent update was published on the United States Patent and Trademark Office just last March 13.
As such, a lot of the details on the patent corroborate what we already reported on earlier upscaling patents that we reported on this January, and that Digital Foundry commented on a few days after.
This patent still describes taking a video image output at 540p, and upscaling it to 1080p. But now, we’re quoting page 19 from the patent text to share the new technology that is described in this update:
“Machine learning can give computers the ability to “learn” a specific task without expressly programming the computer for that task. One type of machine learning system is called convolutional neural networks (CNNs) – a class of deep learning neural networks.
Such networks (and other forms of machine learning) can be used to, for example, help with automatically recognizing whether a cat is in a photograph.
The learning takes place by using thousands of millions of photos to “train” the model to recognize when a cat is in a photograph.”
We found a short video explanation of what a convolutional neural network is here. But what’s interesting, once again, is the patent shares possible uses for this technology, that goes beyond their stated purpose.
For example, the neural network described here could run from the cloud, but it could also be in the actual Switch 2 hardware. The network can make decisions if upscaling to a certain resolution will decrease the Switch 2’s battery, and choose not to upscale. It can even downscale, a choice that might be genuinely useful to increase battery life when playing a low demand game like Celeste.
The neural network can work in several different ways. As we mentioned above, it can stream data to the cloud, for a remote computer to upscale, and send it back to the Switch 2. But the game can also download the image data necessary, and it can even include the image date in the cartridge itself.
Interestingly, the neural network may take care of aliasing in images it generates so that an external anti-aliasing solution won’t be necessary. The current industry standard, TAA (temporal anti-aliasing) is often decried for reducing framerates, but its one developers still choose to use on PlayStation 5 and Xbox Series X consoles.
Since patents may not necessarily describe everything used in actual application, we don’t know how much of this technology will actually be used on the Switch 2. But if the console can decide to upscale or downscale your image on its own, that’s already a huge boon the average Steam Deck or ROG Ally user would envy.
So there’s a lot to be excited about this patent, if it turns out Nintendo will use it on the Switch 2.
