I have deployed liver segmentation module with clara deploy and was able to generate the segmented outputs.
I tried viewing the output with clara render server.However,the performance was very slow and it keeps getting stuck while viewing it with different angles.
The GPU available is T4 GPU and CUDA version 10.2.
If anyone could guide how to improve the performance and send a good documentation of viewing results with clara render server and also the pre-specification required would be of great help.
Thanks for the question and for your interest in Clara!
One thing that may cause some issues is if the TRITON inference server is still running from the pipeline on the same GPU that the render server is on. You can use ‘nvidia-smi’ command to view all processes running on GPU - you’ll likely see a process with ‘trtserver’ in the name. By default this stays up and running as there is some overhead to spinning it up for each pipeline, and in a 24/7 deployment where multiple pipelines may be run at any time, it helps to have it ready to go when a new inference job comes in.
One way to address this is to stop the TRITON kubernetes deployment. You can use ‘kubectl get pods’ to see what pod is running the TRITON server, and you can use ‘kubectl get deployments’ to see which deployment is running that pod. You’ll want to do ‘kubectl delete deployment < trtserver-deployment-name >’, and then view deployments/pods again to ensure that it was properly removed. This will not hurt anything, won’t remove results or models, and will automatically be restarted if a new pipeline is run.
Try to view on render server again once this deployment was deleted and let us know if it helps.