New information on the AMD Navi 31, 32 and 33 GPUs have just been released. These contradict the technical bases of the next RDNA3 / RX 7000 GPUs with notably lower specifications. Obviously, for the moment we are on rumors that contradict rumors, so it is appropriate to take all this with hindsight even if the sources for the latter are fairly reliable about AMD.
The main curiosity of the next RX 7000 cards is partly based on Navi 31. It will be the first GPU from AMD based on the MCM design (Multi Chip Module, the technical feature that made Zen a success) and will probably change the game for the GPU industry.
AMD has to deal with the “performance” of its RX 7000 chips?
But let’s get back to the point. According to the latest information from Greymon55, the internal structure of the Navi 3X GPUs could be different from the information in circulation until now. The main data to take into account would be a reduction of Stream Processors from 15360 SP to 12288 for the Navi 31 GPU. This is still a 20% reduction in the number of cores compared to the 15360 cores previously announced. This reduction in the number of cores will logically have an impact on performance. Speculations placed the top of the range Navi 91 around 92 TFLOPS. After these new leaks, the new estimates would fall to 73 TFLOPS. 
The main explanation of this drop could be found in the yield (the performance of the GPU). Indeed, as you know, the chips are sorted according to their qualities. The more you reduce the final product specifications, the more dies you have and therefore the more products you have to deliver. Yield generally improves over time but the sharpest chips are rare. Let’s keep in mind that for the moment we are on speculations based on leaks. The power in particular is calculated with leaked specifications and a theoretical GPU frequency…This means that the uncertainties are still huge for the moment.