Would it be at all possible to use an old Nvidia Kepler-class Tesla (but one with a crazy amount of RAM and shaders) via either Optimus (i.e., frame buffer is displayed by another GPU, usually integrated) or Looking Glass (VM video output redirected to another display)?
I’ve seen some old Teslas with very high amounts of shaders and RAM going for sale on eBay, and was wondering if these could effectively be repurposed as primary or passthrough GPUs.
Any thoughts, or is this a pipe dream?
EDIT: Or use for video capture/editing on a Linux workstation w/ a primary AMD GPU?
you can get to the very last step but the thing is the windows drivers nvidia gives you won’t work in a VM so looking glass is out
I’ve also tried optimus and lucid virtu, no dice, mine was a Tesla K40
you just might have better luck if you have a tesla with a display output but probably not
you can however use it for rendering AI since it has a large VRAM buffer for cheap
I actually found a video where @wendell got this working.
He used ESXi (which I have some experience with) but doesn’t go into great detail about the implementation. Perhaps there’s a corresponding post in the forum somewhere, but I haven’t found it yet. Since I don’t have $10K for a modern Tesla I was thinking of using something like the Tesla K80.
Other sites/Youtubers make mention of having to flash the firmware and make physical mods of K-series Teslas to get it to work with VMs. It doesn’t seem like Wendell had to go to those lengths, but maybe he glossed over it.
I’m pretty sure that it is using the GRID/Nvidia vGPU software, which is $$$, and the client are remoting in via VMware horizon, which is like RDP.
So the performance (mostly in terms of latency), is useable but not great for gaming, but is ok for CAD and stuff.
This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.