Questions about this project.

[fighterhit]

Hi @mYmNeo , thank you for your work. I have a few questions about the usage.

1. How does the device index uniquely determine a certain GPU card? As far as I know, when a certain card breaks, the index will change.
2. What is the unit of device core/memory limitation? Could you give a specific example?
3. Can this project be smoothly integrated with vcuda-controller and gpu-manager?

[mYmNeo]

1. server_monitor.c is only a example to demonstrate a way to control gpu utilization, one who wants to use it in production should implement it by his own way.
2. core limitation is range from 1 to 100, memory is binary prefix unit.
3. No, being compatible with vcuda-controller and gpu-manager has lots of work to do. I've no time for that.

[fighterhit]

Thanks @mYmNeo , does this project support all current and future cuda versions (e.g. cuda >= 12.0)?

[mYmNeo]

Thanks @mYmNeo , does this project support all current and future cuda versions (e.g. cuda >= 12.0)?

Not guaranteed, but cuda 12 is supported recently.

[fighterhit]

All cuda 12.x supported? Could you give some guidance on how to support newer CUDA versions in the future? Thanks!

[mYmNeo]

All cuda 12.x supported? Could you give some guidance on how to support newer CUDA versions in the future? Thanks!

files with hook suffix, the mechanism is the same with vcuda-controller

[cwdsuzhou]

I also have a question: does this project support all nvidia driver versions ?

[mYmNeo]

I also have a question: does this project support all nvidia driver versions ?

Consider the drivers listed in https://github.com/NVIDIA/open-gpu-kernel-modules as supported.