Senior Vice President David Wang and AMD Executive Vice President of Computing and Graphics Products Rick Bergman told ITMedia that the company could develop a GeForce RTX 4090-class graphics card within the RDNA 3 architecture for enthusiasts, but decided based on the potential height on the other hand, power consumption of such a solution and the costs.
“Technically, within the framework of the RDNA 3 architecture, it is possible to develop GPUs with characteristics that could compete with them (with NVIDIA and its GeForce RTX 4090 – ed.). The PC gamer community eventually accepted a $1,600 TDP graphics card. But after some deliberation, we decided not to follow this strategy. The Radeon RX 7900 XTX that we released costs $999, which is considered “high-end” among enthusiasts, most of whom are gamers. We also released the Radeon RX 7900 XT for $699. We’re repeating our pricing strategy that we used with RDNA 2 graphics cards when the flagship Radeon RX 6900 XT was $999 and the successor Radeon RX 6800 XT was $699.Rick Bergman commented.
He added that when developing the Radeon RX 7900 XTX and Radeon RX 7900 XT graphics cards, the company also tried to stick to common sense in terms of power supply and high-quality cooling. When developing these graphics cards, AMD decided that the new accelerators could work with older power supplies and should easily fit into not very spacious computer cases.
“We didn’t intend to release a $1,600 graphics card for gamers to compete with NVIDIA. Instead, we focused on developing a GPU that fits into the graphics card class. “for $1000“ and at the same time has a good cost-performance ratio. This $600 cost difference between graphics cards can be invested in other important parts of the PC such as: B. into the CPU, improving the overall performance of the game.added David Wang.
Wang also shared some details about future generations of Radeon GPUs. According to him, the future of the GPU lies in chiplet architectures, but the company has decided not to implement it “true chiplet structure” as part of RDNA 3 GPUs Formally, the current generation GPUs are actually the first GPUs with a chiplet layout. However, chiplets are only used here in the form of MCD blocks with memory controllers. The main GPU block with cores is still monolithic, but that will change in the future.
“The chiplet architecture will enable the development of high-performance graphics processors while leading to lower production costs. The cores of modern graphics processors contain more than 10,000 arithmetic cores (arithmetic units for floating point operations). This is more than a thousand times more than the composition of the cores of central processors. If you try to combine chiplets in a GPU under such conditions, you get an incredible amount of connections where reliable transmission of an electrical signal is not guaranteed. Because of this, releasing chiplet GPUs compared to chiplet CPUs is now a very expensive and difficult task. Not only in terms of the number of man hours required, but also in terms of the cost of creating all of these connections within the GPU. It’s now cheaper and more efficient to make monolithic GPU chips. Therefore, we have decided to defer the implementation of a true multi-chip GPU and not use this approach within the RDNA 3 architecture.said Wang.
As for the future RDNA 4 architecture, according to Wang, AMD will focus significantly more on its ray tracing technology while trying to maintain the same rate of increase in graphics performance as when switching from RDNA 2 to RDNA 3 – then the performance gain was around 60% %. Improved task performance and a more flexible computer system are also on the list of things to be implemented in RDNA 4.