The leader of AMD’s Radeon Technologies Group has revealed details about their plans for the next-generation RDNA 4 graphics architecture.
RDNA 4 is expected to build upon the current RDNA 2 architecture, which has received positive reviews for its performance and features and is likely to include advancements in performance-per-watt and ray tracing capabilities.
It’s important to note that Nvidia, a significant competitor to AMD in the GPU market, is also working on its next-generation graphics architecture, Ada Lovelace, which is expected to offer similar improvements. Nonetheless, competition between the two companies drives innovation and benefits consumers in the gaming and other graphics-intensive industries.
The introduction of Primitive Shaders by AMD during the Radeon Vega revealed it was a significant innovation in the graphics industry, allowing for more efficient rendering of complex scenes.
The fact that Microsoft chose to adopt Nvidia’s Mesh Shader technology as the standard for DirectX 12 effectively ended the race for standardization, as Mesh Shaders became the de facto standard for graphics rendering.While both Primitive Shaders and Mesh Shaders offer similar benefits in improved efficiency and performance, the decision to adopt one over the other ultimately came down to industry adoption and support.
However, the competitive landscape of the graphics card market is constantly evolving, and it’s possible that innovations could emerge in the future that challenge the dominance of Mesh Shaders. Ultimately, the goal of both AMD and Nvidia is to push the boundaries of what’s possible in graphics rendering and provide consumers with the best possible performance and experience.
RDNA 4 GPUs will feature even more advanced shaders and techniques as part of its new GPU programming model. These advancements are expected to improve performance and efficiency in graphics rendering further and provide a better overall experience for consumers.
In an interview with 4Gamer, senior executives, including David Wang, the head of Radeon Technologies Group at AMD, revealed that the company’s current RDNA 3 GPUs feature hardware-level Multi-Draw Indirect Accelerator (MDIA) technology, which allows for the processing of Multi-Draw Indirect (MDI). This innovation has led to a performance boost of up to 2.3 times compared to the previous generation of RDNA 2 GPUs.
MDI is a technique used to render graphics more efficiently by reducing the processing overhead on the CPU when dealing with large amounts of data. It has been suggested as a more advanced replacement for the commonly used Level of Detail (LoD) technique, which adjusts the level of detail in graphics rendering based on the distance between the camera and the object being rendered.
David Wang discussing the role of artificial intelligence (AI) in AMD’s Radeon-class GPUs for gamers, explained that while AI processing is an important aspect, it is mainly handled by CPUs since they can perform the necessary computations quickly. Wang stated that 95% of inference processing is done on CPUs. This is something that AMD, being a significant player in the server CPU industry, is well aware of.
He emphasized that while AI inferencing acceleration may not be a current priority for AMD, AI learning and training models remain a significant focus, primarily processed on the GPU side. He noted that AMD had made strides in this area by committing to the ROCm software suite, which he believes is now on par with NVIDIA’s CUDA platform.
This commitment by AMD is expected to help the company continue advancing its capabilities in AI processing, leading to better performance and efficiency for gamers in the future.
AMD is evaluating the possibility of integrating an AI inferencing accelerator in the Radeon RX 7900 “RDNA 3” GPUs, but the decision will depend on the users’ preferences and requirements. However, the company does not want to restrict the AI accelerator to features that only involve upscaling, like FSR, XeSS, and DLSS. They believe there are many other potential applications for the AI accelerator, but they want to ensure users pay for their desired features.
He also advised not to limit the AI Accelerator to image processing technologies like NVIDIA’s DLSS. He emphasized that NVIDIA has chosen this approach and that it works for them, but AMD intends to take a different route.
Rather than focusing solely on image processing and upscaling, AMD wants to use these AI engines to enhance gaming experiences in other ways. For example, AMD’s FidelityFX Super Resolution (FSR) technology does not rely on external hardware accelerators. The company’s core idea is to use AI to create more immersive and AI-driven gaming experiences, such as enhancing NPCs, enemy characters, and bosses in games.
In the end, Rick Bergman, the EVP at AMD, revealed the company’s future GPU strategy and ensured they would move to the RDNA 4 generation, which will provide an incredible performance than the RDNA 3 GPUs
In conclusion, Rick Bergman, the EVP of the Computing & Graphics Business Group at AMD, presented the company’s GPU strategy and confirmed that they would be moving to the RDNA 4 generation of GPUs, which will provide even more excellent performance than the RDNA 3 GPUs. We are excited to see what AMD’s RTG has in store for us shortly.
The upcoming AMD Radeon GPU generation will be available in 2024, based on RDNA 4, utilizing a cutting-edge process node.