RDNA 2 Compute Unit size

Hey, I have been wondering about the RDNA 2 CUs vs the Vega CUs, and what exactly has changed between them. Firstly we have Mark Cerny say that the PS5 CU has 62% more transistors than the PS4 CUs. I think we can safely assume this is also the case with the XSX and the Xbox One. Within the RDNA 2 CUs we have the addition of RT BVH capabilities. Looking at the following diagram from AMD, this is quite a small addition, and doesn’t really add much to the size of the CU.

And then when we compare a GCN CU to a RDNA 1 CU, other than changing around where the scheduler appears, there doesn’t seem to be a whole lot of additional tech in it. Nothing that really stands out anyway.

So where would this additional transistor count be coming from, and does it add to performance or is it more to do with power efficiencies?

Transistor count is mostly dependent on the process node, the Xb1 gen was built using 28nm process compared to the Series X which uses TMSC’s 7nm process.

Basically, the smaller the node the more transistors you can fit in the same area, this is given by the transistor density.

More transistors = more performance, improvement in architecture GCN<RDNA1<RDNA2 (i.e. how these transistors and other components are used) also lead to performance gains and lower power draw.