AMD senior execs including head of Radeon Technologies Group, David Wang, recently sat down with 4Gamer to talk about their future GPU strategy including the RDNA 4 graphics architecture.
AMD Promises To Evolve To RDNA 4 GPUs With Even Higher Performance In The Near Future
AMD’s RDNA 3 graphics architecture was introduced and launched to the world in the Q4 of 2022. Since then, the graphics architecture has been featured in desktop, mobile & integrated graphics solutions from the Radeon RX 7900 (Desktop), Radeon RX 7000S/M (Laptop) & Radeon 700M series (iGPUs).
Starting with Mesh Shaders, David Wang states that AMD’s current RDDNA 3 GPUs feature MDIA, a Multi-Draw Indirect Accelerator that is capable to process MDI on a hardware level. This allows the company to extract up to 2.3x the performance versus RDNA 2 GPUs. It’s been proposed as a true replacement for the classic LoD (Level of Detail). AMD wants to see even more advanced shaders & techniques to be installed within their RDNA 4 GPUs as a part of its new GPU programming model.
“We would like to propose this as a new standard specification for the GPU programming model,” Wang said. Will this be installed as a new feature in the RDNA 4th generation GPU? It looks very interesting.
AMD SVP of Radeon Technologies Group, David Wang (Machine Translated via 4Gamer)
Moving to the AI side of things, David Wang talks about how the use of AI within Radeon-class GPUs for gamers. It is made clear that AI processing is mostly done on CPUs & they can do it fast enough. In fact, 95% of inference processing is said to be done on CPUs and AMD knows this since they are one of the leaders within the server CPU industry. Meanwhile, the AI Learning & training model is something that’s done primarily on the GPU side and Wang states that their recent commitment to the ROCm software suite is now on par with NVIDIA’s CUDA platform.
As for the performance of the inferencing accelerator on the Radeon RX 7900 “RDNA 3” GPUs, he states that installing an AI accelerator on the GPU is more of a business decision on what users want and what they don’t want. AMD doesn’t want to limit the AI accelerator to just upscaling features such as FSR, XeSS, and DLSS. Do note that AMD already has the 3rd generation of FSR known as FSR 3 in the works.
They see an opportunity in a host of other applications but don’t want users to be paying extra for features they don’t want. Wang states that FSR competes well with NVIDIA’s DLSS without AI-based acceleration. He gives full credit to NVIDIA’s AI strategy but doesn’t think that strategy applies to AMD GPUs (yet).
According to AMD, some other uses of AI accelerators within GPUs could be NPC Behavior and Movement with the learning model being applied to make those better and more fluid than the standard AI programming model from the developers. The AI accelerator can be harnessed by developers to make their gaming experience for users even better.
The AI can also be used for image processing which is something that’s picking up the pace within AI generational tools. Wang mentioned that Stable Diffusion for GPUs was implemented within the Vulkan ML API around the same time the Radeon RX 7900 “RDNA 3” series were announced. Looks like the green team (NVIDIA) has been taking full advantage of their AI prowess to power AI generation tools and AMD also wants a piece of that cake.
We think that what should be done with the inference accelerator installed in the GPU should not be limited to “utilization centered on image processing” represented by NVIDIA’s “DLSS”. Take a look at FidelityFX SuperResolution (FSR), one of the FidelityFX series. FSR’s anti-aliasing and super-resolution processing realized without using an inference accelerator provide performance and quality that can fully compete with NVIDIA’s DLSS.
The reason why NVIDIA is actively trying to use AI technology even for applications that can be done without using AI technology is that 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.
For example, the movement and behavior of enemy characters and NPCs are probably the most obvious examples.
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.
AMD SVP of Radeon Technologies Group, David Wang (Machine Translated via 4Gamer)
On a closing note, Rick Bergman, AMD’s EVP of Computing & Graphics Business Group) shared the GPU strategy and promised to evolve to the RDNA 4 generation of GPUs with higher performance than what we got with RDNA 3. What’s even more interesting is that the word “Near Future” was used which would suggest that the next-gen might be coming sooner but we still expect to see those in 2024 as the red team has various RDNA 3 launches planned throughout 2023.
Rick also states that they are currently making GPUs for a wide variety of platforms including mobiles, consoles, and automobiles, and that they will continue to develop GPUs that will not betray the expectations of gamers.
Currently, GPUs for PCs are steadily evolving from RDNA 1 to RDNA 2 and RDNA 3, and we promise to evolve to RDNA 4 with even higher performance in the near future.
In recent years, Samsung Electronics’ SoC “Exynos 2200” has entered the smartphone field by providing RDNA 2-based GPU IP cores. In addition, AMD’s GPUs have been provided to electric vehicle maker Tesla for its “Model S” and “Model X” in-vehicle systems. It is a foray into new fields. We will continue to develop GPUs that will not betray the expectations of gamers.
It’s definitely great to hear from company representatives in such interviews and we can’t wait to see what AMD’s RTG has in the stores for us in the coming years. The next-gen AMD Radeon GPU lineup based on the RDNA 4 architecture is expected to launch by 2024 and will be utilizing a new and advanced process node.
AMD RDNA Generational GPU Lineup
Radeon LineupRadeon RX 5000Radeon RX 6000Radeon RX 7000Radeon RX 8000
GPU ArchitectureRDNA 1RDNA 2RDNA 3 / RDNA 2RDNA 4
Process Node7nm7nm5nm/6nm?5nm/3nm?
GPU FamilyNavi 1XNavi 2XNavi 3XNavi 4X
Flagship GPUN/ANavi 21 (5120 SPs)Navi 31 (12288 SPs)Navi 41
High-End GPUNavi 10 (2560 SPs)Navi 22 (2560 SPs)Navi 32 (8192 SPs)Navi 42
Mid-Tier GPUNavi 12 (2560 SPs)Navi 23 (2048 SPs)Navi 33 (4096 SPs)Navi 43
Entry-Tier GPUNavi 14 (1536 SPs)Navi 24 (1024 SPs)Navi 34 (2560 SPs)Navi 44
The post AMD Talks RDNA 4, GPU-Based AI Accelerators, Next-Gen Graphics Pipeline: Promises To Evolve To RDNA 4 With Even Higher Performance In Near Future by Hassan Mujtaba appeared first on Wccftech.
