NVIDIA Tesla K80 for ML RIG? ML Hardware Suggestions

Gaming and Visualization Technologies General Topics and Other SDKs Drivers - Linux, Windows, MacOS
1

I’m a grad student who want to build a ML rig on the cheap. NVIDIA tesla K80 seem to be a cheap/good option. I am aware that there are cooling issues, since the designed to be a data center. However, I’ve read that there are 3rd party fans that work well. Also is a possible to use multiple NVIDIA Tesla K80? This is the build I was thinking of: AMD Ryzen 7 5800x Noctua NH-U9S MSI PRO B550-VC G.SKILL Ripjaws V Series 4x16GB MLLSE AMD RX 550 4GB(Display only) NVIDIA Tesla K80(1 OR 2?) CORSAIR RM1000x SAMSUNG 980 PRO M.2 1TB SSD Seagate Firecuda 4TB HDD

Planning only so far. Looking for suggestions.

2

Hi there @azizusman868 and welcome to the NVIDIA developer forums.

I cannot give you any kind of suggestions in this context since NVIDIA does not officially support the kind of setup you have in mind. And as a Disclaimer please be aware that running Server GPUs in a Desktop system is not recommended.

That said, please keep in mind that mixing brands in the same system, especially a consumer GPU and Datacenter GPUs will be very challenging. It is highly unlikely to get Linux running on this “Out of the box”. Especially since the last Kepler series Linux drivers are the 470.xx series drivers. We are at 53x series now. So there might be issues with newer Linux kernels.

Good luck!

3

@MarkusHoHo

I think a lot of people, myself included, is on a similar position. The K80 has the reputation of being excellent value. You gotta start somewhere, and not everyone can effort to start at the high end :-)

With your disclaimers in mind, would you mind a clarification? You said:

Especially since the last Kepler series Linux drivers are the 470.xx series drivers. We are at 53x series now. So there might be issues with newer Linux kernels.

Could you clarify: exactly what do I have to look at in order to determine with which Linux kernels I can use the K80?

4

Hi joao-kepler,

I recommend sticking with the details that come with the CUDA documentation of the CUDA version you can, or are still able to use, see the other thread you are on regarding the K80..

The documentation on system requirements also mentions kernel versions.

Newer drivers make assumptions of the minimum kernel versions required, it can be found in the README of the respective drivers.

1 Like