It probably helps that the nvlink 2.0 isn't trying to bridge 2 separate gpus, and the dies are sitting next to each other physically. iirc the big thing that killed nvlink and all the crossfire stuff was interchip latency
It probably helps that the nvlink 2.0 isn't trying to bridge 2 separate gpus, and the dies are sitting next to each other physically. iirc the big thing that killed nvlink and all the crossfire stuff was interchip latency
In the video, Steve goes over the challenges AMD faced with their "infinity cache" on their RX7000 GPUs and how this latency is definitely still a thing from die to die, or chiplet to chiplet. Nvidia is talking a very big game with their NVLink making this thing act like a monolithic die. Early adopters of Ryzen know how this went with CPUs. I'll wait and see here.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.