Fourth generation RDNA architecture puts emphasis on Artificial Intelligence
In an interview with 4Gamer, AMD’s David Wang and Rick Bergman explain what will the next gaming graphics architecture be focusing on.
A Japanese media outlet published an interview made in late 2022 with AMD execs responsible for the company’s graphics business. They were asked about the current generation of RDNA architectures, as well as what is coming next.
The interview focuses on many important topics, as Primitive and Mesh shaders and how they are utilized by PS5/Xbox gaming consoles, but also on Multi Draw Indirect (MDI). AMD’s RDNA3 implementation of MDI acceleration is already 2.3x times more efficient than last generation.
The interview is not providing any statements that would declare RDNA4 ‘X times faster’ or more efficient than RDNA3. However, David Wang did emphasize Artificial Intelligence importance and how it correlates with AMD strategy in the future.
The RDNA3 architecture already features AI acceleration through WMMA Matrix Multiplications (a similar block to NVIDIA Tensor, Intel’s XMX Engine units). 4Gamer editor mentions that RDNA3 implementation is quite modest when it comes to FP16 compute performance of this unit (up to 123 TFLOPs), which is slower than RTX 4090’s 330 TFLOPs and even smaller than Intel Arc A770 at 138 TFLOPs. Needless to say, AMD has a lot of room for improvement here.
The reason why NVIDIA is actively trying to use AI technology even for applications that can be done without using AI technology is because NVIDIA has installed a large-scale inference accelerator in the GPU. In order to make effective use of it, it seems that they are working on a theme that needs to mobilize many inference accelerators. That’s their GPU strategy, which is great, but I don’t think we should have the same strategy. We are focused on including the specs that users want and need to give them enjoyment in consumer GPUs. Otherwise, users are paying for features they never use. We believe that inference accelerators that should be implemented in gamers’ GPUs should be used to make games more advanced and fun.
[..] The movement and behavior of enemy characters and NPCs are probably the most obvious examples [for AI cores]. Also, even if AI is used for image processing, AI should be in charge of more advanced processing. Specifically, a theme such as “neural graphics”, which is currently gaining momentum in the 3D graphics industry, may be appropriate.
— David Wang, AMD (machine translation)
Unlike NVIDIA’s inference processor focusing on image processing functions, such as super-sampling technologies, AMD can achieve similar results without specialized AI cores. That said, the power of AI cores can be put elsewhere. AMD sees the purpose of AI core going beyond image processing.
This type of core could just as well be used for making NPCs in games smarter. Wang claims that AI could make NCPs perform smarter tasks and perform more advanced tasks within the game mechanics. Although AI-acceleration for game characters is probably ambitious to implement than technologies like super resolution or ray tracing, it could be the most interesting field for GPUs to evolve in the future.
