Nvidia for a medical device product

Please tick the appropriate box to help us categorize your post
Bug or Error
Feature Request
Documentation Issue
Other

Hi! Our client is developing a medical device using an x86 based PC and a NVidia GPU for AI. I have some questions regarding the software stack NVidia provides.

FDA Long Term Support requirements

The FDA now requires off the shelf software (also called SOUP) to be supported for the lifetime of the product, ie, if a security issue is detected then the software provider will create patch releases for some Long Term Support versions. With Open Source this is normally a matter of backporting patches, but in Nvidia’s case:

FDA1: Does Nvidia provide Long Term Support releases for CUDA?

FDA2: Does NVidia provide Long Term Support for nvidia-cuddn?

Software Bill of Materials and OS support

On medical devices running Linux it is customary to use Yocto because it is much easier to create small and fully managed images with a reduced set of software, thus reducing the SBoM. In this case we are using the x86-64 architecture (aka amd64). Do you provide any kind of support in this case?

In case you don’t: which of the following options would you suggest?

SB1: install NVidia software using deb packages. Might work, but binary compatibility is not guaranteed?

SB2: install NVidia software using the shell installer. Might work, but binary compatibility is not guaranteed?

SB3: use a docker container.

SB4: other, please specify.

Hi @lpmeyer I’ve moved this post from under AI Workbench to just under the AI and Data Science category, so that we can make sure we get the right eyes on it and get you an answer! Thanks for your patience.

1 Like