Our head of ML is interested in upgrading us from a Nano to a Xavier. That adds quite a bit of cost to our platform and we have power limitations. Most likely we’re going to have to nerf the Xavier to its lowest power mode of 7.5 Watts. This is in line with what the Nano consumers. Is the Xavier still much more powerful at that limited power level over a Nano?
I can’t answer that question, but Nano is old enough that it won’t receive any feature additions. If you want newer features, for example Ubuntu 20.04 instead of 18.04, or CUDA 11 instead of 10, then Xavier or Orin are your only choice (and Xavier can use both older software and newer software). Sorry, I can’t answer the computing power versus lowest power mode comparison.
Couldn’t agree more, about future functionality. Nvidia has the Nano Next scheduled for next year which will bring it inline. The problem is that the Xavier is like 4x the price of a Nano so there is that to deal with in the short term.
What is the motivation for wanting the upgrade? If it is for training, then you could probably skip the upgrade and use pretrained models from a desktop system. If you want more performance with something like vision detection, then the upgrade is probably justified. It seems that a lot of your question is not just about money, but whether the upgrade can succeed at a given power level, so your use case for power and training versus use of pretrained models might be part of giving a good answer.
I’d have to ping our ML lead but there are some specialty cores on the Xavier that are not present on the Nano. The overall performance is also attractive to him. You know the classic dilemma, engineers always want the faster machine. We’re sort of in a place where we’re trying to determine how much value can we deliver, ML wise with a Jetson Nano. We may end up having a hybrid system like you’ve described but we’ll see.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.