AMD

RDNA 4 should work on AI: versus Nvidia, where does it make sense?

We know well that AMD graphics cards while offering solid performance, lose to Nvidia when upscaling technologies are used. Your DLSS is way ahead of the curve in this regard, and it seems that even RDNA 4 would not bring any dramatic changes in this regard. While AMD already has dedicated accelerators for AI systems, it also says that while Nvidia's approach to accelerators is great, AMD doesn't want to go down the path that Nvidia did. At least not now. This was said by AMD representatives (David Wang and Rick Bergman) in an interview with the Japanese magazine 4gamer.

Indeed, AMD says its FidelityFX SuperResolution (FSR) doesn't need AI accelerators to achieve Nvidia DLSS-like performance and quality (this can certainly be argued when its competition is a much weaker Intel). According to him, Nvidia is pushing AI so vehemently because they're on the GPU and it's looking at ways to really use them. Otherwise, users would be paying for something they don't really need when buying cards (which is a very debatable claim). He wants to focus on what players really want, not what others tell them to want.

However, AMD believes that AI accelerators in GPUs have their place, but it's pointless to waste them on upscaling like Nvidia (general image processing) when it can be done without them. Instead, he says, they should be used for something that actually makes games more advanced and fun. As an example, he cites the movements or behavior of enemies and NPCs. But how this comes true is an unanswered question. How do you see it? Do you agree with AMD that AI accelerators should be used in this way, and not exclusively for upscaling and image processing?

From other information, we can point out that AMD should work on the Multi-Draw Indirect Accelerator technology, which should ensure that the rendering commands are handled more by the GPU, in order to limit the communication between the CPU and the GPU. This should bring an additional performance boost.

by: Milan Šurkala