I agree, having tensorflow-gpu helps my models train and predict much faster

Agreed, particularly for ML inference serving. Either standalone GPU instance types or the ability to attach an elastic GPU would allow us to host ML inference servers on Render.

Any update on this? Would love GPU support!

Having a service using Tensorflow or another AI library is becoming the industry standard. Using them without GPU support is super slow and having to run them on another provider like gcloud is suboptimal as well. I would love to be able to have that option here as well.

how to add GPU Instance on Render ..? i need it

Would love this! Very helpful for ML serving as well.