[ineedasername]

As and outsider, can someone explain the "synergy" (god I hate that word) for the companies? It looks like Mellanox is primarily a network equipment company. What is the "fit" for that within a graphics/AI chip producer?

[yvdriess]

Communication is a key bottleneck. Modern supercomputers budget roughly half to interconnect. Nvidia invested heavily in developing NVLINK to be able to make GPUs communicate at the bandwidth necessary to make multi-GPU boxes more practical (e.g. DGX-1). To make it scale out across multiple boxes, you need a high bandwidth, low latency interconnect, which is where IB comes in.

[chas]

If you want to run a computation on more than one of those AI (and scientific linear algebra) chips, you need some network to connect them. The higher the bandwidth and the lower the latency of that network, the less likely the network performance limits total system performance. See NVLink as an example of Nvidia’s related work. (https://en.m.wikipedia.org/wiki/NVLink)

[p1esk]

Nvidia sells supercomputers: https://www.nvidia.com/en-us/data-center/dgx-saturnv/