[sashank_1509]

PaperSpace product recently has been really bad. Gradient Notebooks are a worse version of Google Collab. Useless for any serious DL product building.

Their DL virtual servers, Core I think you call them is horrible. Very slow internet, takes forever to copy datasets into them. Most of the time, it's an uphill battle to get ssh access to them, they create some pointless virtual console instead and then a GUI to copy datasets, run trainings etc which is confusing and a hassle over simple ssh access. Programmers just want a simple ssh access to a server with a GPU, we are not looking for a WYSWYG like editor! I really don't know who the customer for this is? Marketing execs who want to do deep learning? I tried powering through the documentation, but it was outdated, and was plain wrong at points. It took me a couple of hours just to figure out how to copy a large dataset into my server and find the path to that dataset. Once I discovered, Lambda and Vast ai, I never looked back and forgot Paperspace for good.

Really as a cloud provider, all you need to do is create send a ssh tunnel to a system with a certain amount of compute and memory. Maybe like AWS you can create storage buckets but it's not absolutely neccesary. Don't add GUI, interfaces etc, your customers are engineers and they prefer simple systems that give them control.

One feature I would like in Lambda/ Vast is the opportunity to copy the dataset into the server before the GPU hours start billing. When you have TB's of datasets like me, you end up wasting 8-9 hours just copying the dataset and it feels annoying that I pay for the cloud hours during that time. Amazon kind of solves this, but it slows down data access in return. I would like a cloud provider who just lets me copy everything before starting to bill me.