[In a recent online interview, Mark Papermaster was asked what kind of year it would be for AMD, and while the CTO's answer was naturally focused on AI, he also mentioned a curious little tidbit. He said that the company's gaming devices will have AI-based upscaling, and that 2024 will be the year we see it.

The discussion with No Priors (via Videocardz) on their YouTube channel is primarily about AMD's position and thoughts on machine learning, cloud computing, and other big money areas. If you're not interested in these particular topics, it's a good chat between the host and Papermaster, albeit a bit dry.

However, the CTO's final remarks on AMD's plans for 2024 were certainly interesting to gamers, thanks to his brief comments on how the chip giant will bring AI upscaling to gaming devices.

"Now, this is a very big year for us. Because we have spent so many years developing hardware and software capabilities for AI. We're just now getting our entire portfolio ready for AI: cloud, edge, PCs, embedded devices, gaming devices. We are using AI to enable the upscale of gaming devices, and 2024 will be a really big year of development for us"

. [At the moment, AI upscaling is the exclusive domain of Intel and Nvidia. Both companies employ a two-stage upscaling process: in the first stage, an image scaler is run through shaders to change the resolution of the rendered frame to the level of the monitor. In the second stage, a neural network is applied to tidy up the image and remove any defects caused by the upscaling process.

The mathematical operations to do this can be performed on the shaders, but in the case of DLSS and the Intel Arc-specific version of XeSS, they are processed in a dedicated matrix unit in the GPU for best performance and visual results. on AMD's Radeon models. The graphics chips used perform all calculations via the Compute Unit (CU), which is why FSR does not involve an AI step to organize the images.

However, one interpretation of Papermaster's comment is that the next graphics architecture, RDNA 4, may have a discrete matrix unit in each CU, and if so, it could be used to handle AI upscaling The idea is that it could be used to handle AI upscaling. Another view is that AMD could develop relatively lightweight neural networks processed by the CUs and use them as an additional step in the FSR pipeline to improve visual quality.

Currently, the performance of FSR and DLSS are about the same, but DLSS is better when it comes to graphic fidelity. It can be noticeably better in some games, but it is highly subjective during gameplay.

Once AI upscaling is implemented in FSR and can be performed on any GPU (as long as it supports the relevant matrix instructions), we can expect a definite improvement in quality, but performance will probably decrease.

It is the way Papermaster refers to "gaming devices" that makes it seem most likely that AMD's AI upscaling will be done this way. This is because if it is done just for RDNA 4, one would expect him to mention this, or at least talk specifically about the GPU.

However, the phrase "gaming device" covers everything that supports FSR, including current Radeon cards and the APUs in portable PCs and consoles. that a dedicated matrix unit like DLSS is needed. If found, it would only be supported on AMD's upcoming GPU architecture and would not be noteworthy on the Xbox Series X or PlayStation 5.

RDNA 4, whatever its AI capabilities, will compete with Intel's Battlemage and Nvidia's Blackwell at some point in 2024, but these companies are already upscaling with AI.

AMD is fashionably late to the party, but could steal the limelight if it can offer better upscaling that all GPUs can use. 2024 is a big year for AI and GPUs.

.