Jetson devices


Can we use any of Jetson devices to host Clara Train platform. This is required as Healthcare facility we are working with is not allowing to give their data outside their network.
If so, what variant of Jetson device do you suggest?

Clara train is requires pascal family or newer as Volta or Ampere. it could also work with RTX cards. However, I think the jetson is not powerful enough.

I don’t fully understand the reason behind your ask if centers don’t want to give out their data (which is common practice in health care) then you are looking for Federated learning where each site trains on their data and only the model weights are sent to a server for aggregation. If that is the case and you wanted to used clara train fl, please take a look at the recently release as open source NVFlare. NVFlare is the core which clara train is using to do FL.

Hope that helps.

Thanks @aharouni again.

As part of initial model building, we required to train model in healthcare facility. Once model start giving descent outcome, that’s when I thought it would be ideal to move to Federated learning approach.

We are using Monai Label/Train/Deploy for model annotation/training/ deployment. Could you please suggest any good hardware?


This really depends on your use case and budget. For large teams, multiple use cases we recommend DGX A100 with 8x A100 80Gb gpus. Smaller teams can get a DGX station with 4xA100. For prototyping you can any A cards or RTX keeping the gpu memory in mind for your use case and model size.