Although the rumors that in the flagship processor AMD RDNA3 will offer more than 4 times greater computing power of single accuracy compared to RDNA2. NVIDIA presumably makes a similar update for its future flagship. It is expected that the AD102 graphic processor, based on the ADA LoveLace architecture, will provide a power of more than 100 TFLOPS, which is 2.5 times more than 40 TFLOPS offered by RTX 3090 Ti and 2.8 times more than that of RTX 3090. The power of the FP32 (single accuracy) is not anyway, automatically guarantees better game performance.
Both the Greymon55 insider and Kopite7kimi agree that the TFLOPS war will be only part of the battle for the fastest table graphics. The game has something else, for example, acceleration of rays tracing, supported super-high resolution technology and other functions that can persuade the scales in favor of any architecture.
To reach a capacity of 100 TFLOPS, the AD102 graphic processor with 18,432 CUDA cores should operate at a frequency of 2.7 GHz, but almost certainly RTX 4090 will be delivered with a partially disconnected graphic processor. Therefore, the clock frequency will be automatically higher. According to Greymon55, the next generation flagship cards can be delivered with very similar clock frequencies, which in the case of the AMD Navi 31 graphic processor means 3.0 GHz, provided that a complete graphic processor is used.
We already know a lot about the next generation graphic processors, some rumors have been going on for several months. But this does not mean that everyone knows everything. The number of CUDA/streaming processors FP32 can still change. What, apparently, all leaders agree with the fact that the next generation graphic processors will need a lot of energy.
#Graphic processors