Originally posted by zexelon
View Post
die shring to 5nm TCSM for the GPU core
HBM3 modells with 32-64GB vram for high paying customers like apple
increase the 128mb infinity cache to 256MB and maybe go from 12nm to 12nm-LP+ or 7nm-
Also Global Foundries invest to double the 12nm capacity (FD-SOI)
right now 5nm and 7nm is complete overloaded without any capacity free so the 256bit infinity cache chip will be FD-SOI 12LP+ nm
I think AMD also workd on OoO Out of Order execution shaders.
Multi-GPU-Chiplet design for VR gaming... with 2 gpu cores for each eye one gpu and 2 infinity cache and HBM3 vram.
as you can see Nvidia is doomed...
Comment