Since last July, The GPU of the AMD began to shift to nm RDNA 7 architecture, Now Radeon 5000 series is the first generation RDNA architecture, And big Navi/Navi 2X series this year, Will use nm 7 process RDNA2 architecture, Energy efficiency has increased by 50% again.

This year's flagship graphics card big Naiv is reported to have up to 80 sets of CU units, that is ,5120 stream processors ,122% more than RX 5700, Possibly RTX 3090 a war with N high - end models.

And back? The road map published before the AMD shows RDNA2 followed by a RDNA3 structure GPU, but with the exception of code names, AMD are reluctant to mention details, and the Zen4CPU after the Zen3 are clearly five processes, which can be used only

Why are officials so secretive? Perhaps the progress of the RDNA3 is a bit slow, now have not determined the specific process, but this statement in TSMC 5 nn clear production, AMD also determined to use 5 nm.

According to the latest disclosure, amd does not know the specific nodes,It is likely to be related to the new design of rdna3 architecture, that is, it will also adopt the way of small chip to separate the computing unit from the IO unit. The whole design is similar to that designed by AMD on the 7Nm zen2 processor.

The advantage of small chip design is that it can stack the required chips like building blocks. If rndna3 is designed like this, it will be easier to stack the Cu computing unit and expand the stream processor unit,For example, the computing core can be made in a 5nm process,The IO core is likely to be manufactured in the same 12 / 14 nm process as zen2 to reduce costs.

Of course, there is no final conclusion on the specific situation. At present, the architecture of rdna3 is only a guess, but from the perspective of technology trends, this path is most likely.

Considering that big Navi graphics card will not be available until November this year, I'm afraid rdna3 will not be available in a year, and it will not be available until 2022. Amd still has enough time to improve and optimize the rdna3 architecture.