Alright, alright, I already know what you are pondering—you do not care what the brand new GPUs appear to be so long as they carry out. The factor is, these renders truly may give us some perspective on what to anticipate from every of AMD’s next-gen Radeon GPUs.
The authentic renders truly embrace textual content over them labeling what every chip is in addition to its capabilities. Here now we have pictured Wild_C’s finest guess at what Navi 31 will appear to be. The GCD contains six shader engines (SE), every with two shader arrays (SA) comprising 4 workgroup processors (WGP). Keeping in thoughts that the WGP has been doubled in width in comparison with RDNA 2, that offers us a complete of 12,288 shaders.
Navi 31 is predicted to make use of a smaller GCD with 4 shader engines, just like the Navi 21 die used within the Radeon RX 6800 and bigger RX 6000-series GPUs. However, because of the aforementioned doubling of WGP density, it would have double the shader depend. It will get 4 MCDs, giving it a really respectable 256-bit reminiscence bus and 128 MB of Infinity Cache—no less than on fashions with out 3D V-Cache.
Finally, the Navi 33 processor is predicted to be a monolithic design. While it seems to be surprisingly giant within the comparability render above, Wild_C notes that he expects it to make use of TSMC’s N6 course of as an alternative of the newer N5. Despite the distinction of just one digit, N6 is definitely a refinement of the older N7 course of, and there’s a fairly huge distinction in transistor density between the 2.
Going by rumors, Navi 33 can have two RDNA 3 shader engines, giving it a complete of 16 WGPs for a most of 4096 shaders. It’s anticipated to have as much as 128 MB of Infinity Cache, like Navi 32. Wild_C predicts that Navi 33 can be restricted a PCIe 5.0 x8 interface, which is one thing we have seen elsewhere. It appears affordable sufficient, though it may restrict efficiency on machines nonetheless restricted to PCIe 3.0. Then once more, the CPU in such a system is prone to be bottleneck sufficient for a next-generation GPU.