The RDNA3 architecture introduced Important news compared to RDNA2that is indisputable, and this has generated a lot of expectation about RDNA4, a next-generation graphics architecture that will be used in the Radeon RX 8000, and that according to AMD itself will focus on introducing “relevant news to improve performance in games”.
All the information that we are going to give you below about the RDNA4 architecture It is official, that is, it comes from AMD itself, specifically from an interview that David Wang, senior vice president of Radeon Technologies Group, and Rick Bergman, executive vice president of Computing and Graphics Business at AMD, gave to the Japanese outlet 4Gamers. You can consult the complete interview through the link that I have left you, although Google’s automatic translation is not entirely good, and it will force you to make an important interpretation effort at many times.
The interview is very extensive, but I have filtered the most important keys to share them with you in a summarized and easy-to-understand way. AMD has made it clear that RDNA4 will be an architecture that will be focused on gaming performancethat is, in raw power, and that the company does not plan to adopt a specialized and AI-based feature set to compete with NVIDIA.
What does this tell us? Well, in principle AMD will not launch a solution that is really up to NVIDIA’s DLSS, and that they will not shape a technology that rivals the generation of frames. In the interview Wang says that FSR is a technology that does the same as DLSS without resorting to AI, but this is not true. Both are image rescaling and reconstruction technologies, but they do not work in the same way, and the result they offer is not at the same level. DLSS uses AI and achieves much higher image quality.
RDNA4 will reduce CPU dependency
Wang believes that it is better to use the computing resources available at the AI level in improve gameplaybetting on things like the procedural generation of worlds, the improvement of NPCs and bots present in games and generally boosting the level of complexity of next-generation games, instead of devoting those resources to tasks processing, reconstruction and rescaling of the image.
This idea is very interestingbut I think Wang you forget that consoles are the basis of current video game developments, and that they do not have dedicated AI cores that can accelerate everything that the executive has raised in his comment. To take advantage of the AI at the level that Wang has said, developers would have to create two different versions of the same game, one for consoles and one for PC, and it goes without saying that this would not be profitable for them. If many already complain about having to adapt a game to Xbox Series S, imagine what would happen with that differentiated dual development.
Another important point that Wang has commented is the need to reduce the dependence of the GPU on the CPU. It is a subject that I have already mentioned to you on numerous occasions, and he has highlighted that they have already taken measures in this regard with RDNA3, among which we can mention the Multi-Drawing Indirect Accelerators (MDIAs), which allow you to generate various drawing commands that can be executed directly on the GPU. We can expect improvements in this regard in RDNA4, and also a significant leap in energy efficiency.
AMD has not given details about the presentation date of this new architecture, nor has it specified anything about its possible specifications, but I am convinced that they will also focus their efforts on improve performance with meshed and ray-traced shadersand that image rescaling and reconstruction will continue to play an important role in the RDNA4 architecture.
