LEAP-Pangeo More Hardware Access? #9
-
|
Another thing that came up is the issue with some researchers possibly hoping to run large models on the server. As far as my access goes, I see that if we want to run things on a GPU, we can use the Large + Nvidia K80 GPU, which has 24GB of VRAM. If the model we're hoping to train does not fit on 24GB of VRAM, is it possible to get access to multiple K80s connected via SLIs? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
No, right now that type of hardware is not available in LEAP-Pangeo. We can work with 2i2c to see if we can create a larger configuration. Are you sure this is something you need for your project? |
Beta Was this translation helpful? Give feedback.
-
|
Thank you Ryan for all your help! I'll know soon about the hardware requirements once I start running the models but this was one possible issue that I foresaw. I could post an update soon on my experience; however, for now, I am still running most things on my personal desktop. |
Beta Was this translation helpful? Give feedback.
No, right now that type of hardware is not available in LEAP-Pangeo. We can work with 2i2c to see if we can create a larger configuration. Are you sure this is something you need for your project?