Newbie Here, Question about linking together AGX Xaviers for increased capability

Apologies if this is the incorrect forum for this:

I am attempting to build robots similar to the Atlas, but perhaps more complex.

Is it possible to link together multiple AGX Xaviers for increased capacity/performance?

Thank you.

Hi brotherklight, yes you can connect multiple Xavier’s in a mini-cluster configuration and use intercommunication between them to distribute the processing load. The easiest way to get started would be to connect them over gigabit ethernet, or attach additional networking cards for additional ethernet ports and bandwidth. Although more advanced, it’s also possible to interconnect them over PCIe, for more info please refer to the PCIe Endpoint Design App Note.

Note that each Xavier in the cluster with still function independently, and it’s up to your software to coordinate the transfer of data and schedule the processing functions. You could use a framework like MPI for automating this, which is similar software that is used on bigger HPC clusters.

Thank you very very much!

It’s so nice to have an affordable module/developer kit like the Xavier available for those coming into the industry. It sounds like it will be enough to build what I desire. Thank you NVIDIA. (And thank you Dusty!)