Yup. They started this transition away from NVLINK in the same generation in which they introduced their tensor cores. It's part of their ML profit strategy.

Yup. They started this transition away from NVLINK in the same generation in which they introduced their tensor cores. It's part of their ML profit strategy. There's really nothing you can do outside of soldering on higher capacity RAM chips onto a 4090. There are some 20GB 3080 cards floating around out there.