Hi, I want to build (as few as possible) AWS Machine Images (AMIs) that include the tools we need and can be used to spin up compute nodes running on different GPU types (OS is assumed fixed). If I install the latest CUDA driver in a recent GPU, will the same AMI be usable to run any other GPU model? Probably not. Should I create a different AMI for any different GPU model? Hopefully not. Is there a matrix that tells me which classes of GPU models are compatible with each other at the driver level so that I can use the same AMI for that class?
Table 4 here should give you all you need.
So using the latest driver should cover you as far back as Maxwell.
1 Like
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.