My boss asked me to look at a problem that one of our clients has and I have not found a solution yet… So I’ve created a support request on the general NVIDIA site last week and they told me to ask it again over here.
Bottom line is: they want to use remote accelerated graphics in a Windows Server 2010 R2 VM via a Tesla K20m card that is physically installed in a Dell PowerEdge R730 server running ESXi 6.0.
Has anyone ever succeeded in doing this?
I’ve tried it with multiple VIB versions (346.27, 346.42, 346.68 and 352.54).
The install Always works, but after a reboot the kernel module is not automatically loaded.
But you can do it yourself and then get some sane info from the relative command’s (vmkload_mod Nvidia, vmkload_mod -l | grep Nvidia, esxcfg-module -e Nvidia) but when you run /bin/nvidia-smi you get an error: Failed to initialize NVML: Unknown Error
What are we doing wrong or how do we get this to work?
K20m cannot provide graphics acceleration/support in a windows environment.
So under what operating system does it work then?
I’ve seen there are Windows drivers provided on multiple sites, do they only work natively?
The people that sold it to us guaranteed this was a good solution for our needs, or not?
I was responding principally to this statement:
“they want to use remote accelerated graphics in a Windows Server 2010 R2 VM via a Tesla K20m card”
The principal purpose of K20m is to provide support for compute operations, not graphics acceleration.
It can do this either in a windows environment or in a linux environment.
I believe it should be possible to get K20m working in your setup, however once it is “working” if you are trying to use it to provide windows graphics acceleration functions, I think you’re going to be disappointed.
Thanks for clearing this out so far! :-)
I thought it was possible for this kind of card to switch between compute and graphics mode if you boot with some kind of specific ISO but I have not found this ISO so I don’t know if what I’m saying is correct…
But I guess it’s a big no-no for my particular case, right? :-/
The specific case it’s needed for is also a wee bit silly. One of our clients has programmed something for their bookkeeping that runs on a VM and apparently it’s quite colourful, has some nifty different fonts, grids and scrolling things in it (their words) that behave a bit laggy, as a remote desktop connection is not about eye candy and this is more or less to be expected.
So I will have to dissapoint them? Is there any card you’d recommend otherwise?
And could you elaborate on getting this working in my setup?
Kind regards and sorry for all the questions,
I am afraid I have no idea what “ISO” means here.
This is completely outside my area of expertise (your use case seems to have nothing to do with CUDA?), but NVIDIA offers a line of GRID GPUs that might be appropriate for your use case:
http://on-demand.gputechconf.com/gtc/2014/presentations/S4725-hi-perf-graphics-nvidia-grid-virtual-gpus.pdf [NOTE: this overview is from 2014 and may not reflect latest advances in GRID technology]
Yes, you’re likely to have better results with a GRID setup or a Quadro GPU.
You don’t know what an ISO is? ;-)
An ISO-file is a (read-only) image from a storage device (like magnetic, optical or flash memory) but neatly packed in one file. This is easy for distribution and can be transferred back to one of the media types mentioned before.
It was mainly exclusively used for optical media but can be used as a back-up for your hard drive too for example since storage capacity keeps on growing…
Or for the distribution of operating systems. So you write them to an USB-drive, boot from it and install the OS.
And I posted my question here because I did not know where else to post it, this one seemed like the only one that has something about setup and installation.
But the gpumodesetting ISO is a bootable version that’s used for switching the card from compute mode to graphics mode.