Along with the announcement of its new Ampere-based Tesla A100, Nvidia also officially announced its new artificial intelligence system DGX A100 based on these graphs, which are accompanied by two AMD EPYC processors to take advantage of not only its high processing power, but also its enormous number of PC-E 4.0 lines.
Nvidia decided to use not one but two AMD EPYC 7742, totaling 128 cores and 256 threads, and 256 PCI-E 4.0 lines for expansion. These PCI-E 4.0 lines are used to connect the 8 Tesla A100 cards, and they also provide connectivity to the 8-port Infiniband Mellanox ConnectX 200 Gbps NIC, NVLINK switches, two 1.92TB M.2 SSDs, and four U.2 SSDs totaling 15TB. Other features include 1TB of 16 channel RAM, and Ubuntu operating system.
Obviously such a system is not cheap. Nvidia is selling the new DGX A100 system for $ 199,999, and it's now available to pre-order.
What do you think about this new Nvidia DGX A100 system? Do you like the idea of using AMD EPYC instead of Intel Xeon?
Source: TechPowerUp