when attempting to execute nvidia-smi from the command line returns "unable to determine the device handle for your GPU 0000:01:00.0: unknown error "
I am running 2 cards in this machine as one is just a Geforce 210 that is being used for the displays and then the Nvidia Tesla card M2090.
Both are functioning correctly, with no issue either with the display or running jobs on the compute card,
Is there a variable for nvidia-smi that will point it to the tesla card or should it pull back information for both cards
System info -
Windows server 2019
Tesla driver - 386.45 OpenCL:1.1
thanks for any input
The GeForce 210 is a GPU that is quite old. It is from a family of GPUs (compute capability 1.x) that is not supported by the 386.45 driver.
That is why you are having trouble. The solution is to remove that card from the system, or downgrade the installed driver to one that is compatible with the GeForce 210 (which would be a 34x.xx driver). However I don’t think there is any driver like that for windows server 2019. Your configuration is fundamentally unsupported. The solution is to remove the GeForce 210.
Note that the M2090 is also old, and is no longer supported by current drivers from NVIDIA. Support was dropped for that card after 390.xx drivers.
thanks for your reply, I was already aware of the driver issue with the tesla card as I have one running on an ubuntu machine with driver 390.xx, however I found that the stability for the windows machine works better with the 386.xx driver.
I was only using the 210 card as a display adaptor while setting up the server machine. Since I can access it With RDC and can remove that 210 card, Is the driver still to old to return information with nvidia-smi for the tesla card. The main reason I am trying to get that command to return information is for temp monitoring.
GPU-Z only show information related to GPU load, Memory controller load, Bus interface load and VDDC. it will not return a temp. Or does the M2090 not have a temp sensor on it? I have tried other monitoring software like HWMonitor- and that just displays the card info
My main goal here is to build a tesla compute node as part of my degree I am working on, and while I cannot afford the newest tesla model cards I was able to get my hands on multiple m2090s as well as a couple k40s, I am determined to build with the hardware I have.
Given that goal and the driver limitations for that model cards is it even practical for me to build with both models of cards or where there be issues with multiple drivers for the different models.
while Server 2019 was supplied by the school I do have pervious versions of server as well as the ability to use an open source OS as well
thanks
Your 386.xx driver should work fine with the Tesla M2090. Once you remove the GeForce 210, nvidia-smi should work correctly and show you the temperature of that M2090 GPU.
The M2090 requires server-assisted flow-through cooling. If you don’t provide that or some equivalent, it will likely overheat.