AMD has just started marketing graphics cards based on the RDNA 3 architecture, but is already ready to talk about RDNA 4. David Wang, head of Radeon Technologies Group, gave information about the next generation in an interview with 4Gamer. Of course, there is quite a lot of information so far, and it is very superficial because it is unlikely that the first cards based on RDNA 4 will be released before the second half of 2024.

For example, Wang explained that AMD does not want to limit AI functions in GPUs to FSR-type scaling technologies. The company sees many other possibilities, but doesn’t want users to overpay for something they won’t use. Wang mentioned Nvidia here, which integrated large tensor blocks into their GPUs and is now trying to include them. AMD believes that such a solution is not the best for gaming graphics cards.

As an example of a possible application of an AI accelerator in the GPU, the head of Radeon Technologies Group named the behavior of opponents and non-player characters in games (NPCs). The AI-based learning model will then be more flexible, allowing developers to offer new opportunities to players. Here, by the way, we can talk about the experiments of enthusiasts on the implementation of ChatGPT in games specifically for NPCs, thanks to which you can communicate with them about everything, as in a regular chatbot in online format. It is possible that something similar awaits us in games in the near future. AI can also be used for image processing. Wang reminded that the popular neural network Stable Diffusion for GPU is based on the Vulkan ML API.