For large-scale datasets, two A800 40GB Active GPUs can be connected with NVLink to provide 80GB of memory and 2 times the processing power. Funded a bunch of analysis On this space but many converging advances and infra means that decentralised AI networks will outperform centralised gigamodels in another couple https://johnathanfaqet.blogminds.com/a-secret-weapon-for-nvidia-a800-80gb-pcie-gpu-28842462