In context: Team Red revealed its AI acceleration hardware alongside the Radeon RX 7000 graphics cards in December but hasn’t said much about what it plans to do with them – at least not in English. Japanese outlet 4Gamer spoke to AMD about how its AI ambitions differ from Nvidia.
In a recently published interview, senior figures from AMD expressed interest in AI applications in graphics cards beyond image reconstruction, which Nvidia has made popular. The machine-translated comments mark the first time the company has spoken on the subject at length since unveiling its first GPUs incorporating AI accelerators last year.
Nvidia has always been proud of its DLSS technology that relies on the tensor cores of RTX graphics cards to improve image quality and performance through machine learning. Team Red’s counterpart, FSR, strives for the same goal without hardware-based acceleration, which can be used on multiple GPUs but can be less effective depending on the situation.
When AMD unveiled and launched the RX 7900 XT and 7900 XTX with AI accelerators last December, onlookers thought they were the company’s answer to Team Green’s tensor cores. That may be the case, but AMD doesn’t want to stop there.
David Wang, senior vice president of engineering at Radeon Technologies, suggested using AI acceleration to improve AI and movement of NPCs in games. He also said that AI can speed up the graphics pipeline through neural techniques, mentioning Stable Diffusion among other technologies.
Wang said that AMD wants to ensure that AI acceleration in gaming GPUs mainly focuses on improving the gaming experience. His comments compared Team Red’s vision to Team Green’s various non-game applications that use tensor cores.
Nvidia revealed RTX-exclusive AI functions such as noise cancellation, video eye contact simulation, and web video upscaling. Wang indicated that some of these applications do not necessarily require AI and that Nvidia is using them to improve its proprietary technology. Wang did not criticize Nvidia’s stance but suggested that customers may be paying for features they don’t use.
Wang’s interview also covers topics such as the transition from primitive to mesh shaders. The company is also researching technology that can change the relationship between the CPU and GPU when drawing graphics.