Spin up GPU on request

Repls allow for boosted K80 GPUs for ~$3 a day. For my specific use case, I have some website that is fairly low CPU and GPU usage. When the user clicks a button, it activates a script that using pytorch, tensorflow, etc that is GPU intensive and runs for several hours.

This script needs the dedicated K80 GPU to this task. Obviously this doesn’t scale at all, even to > 1 concurrent users. It would be sufficient to spin up a new K80 GPU and run the script on that for every time a new request is sent.

Does Replit support anything like this? If not, I will begrudgingly have to move over to some traditional cloud service, as it’s essential for the functioning of this app.

1 Like

I can’t say for sure, but it is unlikely the Repl would last hours without being rebooted. Repls reboot so that Replit can balance their servers, so it is likely your program would be disrupted. Unfortunately, this may mean Replit is not suitable for your needs…

1 Like

Thanks, that’s too bad, there’s so many things to like about Replit. :disappointed:

1 Like