NVIDIA
Grid K1 & Cuda
Hi, I have a Nvidia Grid K1 card, in a Dell R730 Server, using with ESXi 6.5 vGPU to a Windows Server 2016 VM, using K180Q mode. I am using the latest driver pack: NVIDIA-vGPU-kepler-vSphere-6.5-367.123-370.17, and using 370.17_grid_win10_server2016_64bit_international.exe in the Windows Server 2016 VM. Everything looks correctly installed. But..... I am trying to use applications which utilise CUDA, so I am trying to install a CUDA driver, (I have tried CUDA version 7.5, 8 and 9) but when I try and install any of these CUDA drivers I just get the massage that no compatible hardware was found. Any ideas please?
Hi, I have a Nvidia Grid K1 card, in a Dell R730 Server, using with ESXi 6.5 vGPU to a Windows Server 2016 VM, using K180Q mode.
I am using the latest driver pack: NVIDIA-vGPU-kepler-vSphere-6.5-367.123-370.17, and using 370.17_grid_win10_server2016_64bit_international.exe in the Windows Server 2016 VM.
Everything looks correctly installed. But.....
I am trying to use applications which utilise CUDA, so I am trying to install a CUDA driver, (I have tried CUDA version 7.5, 8 and 9) but when I try and install any of these CUDA drivers I just get the massage that no compatible hardware was found.
Any ideas please?

#1
Posted 12/26/2017 08:39 PM   
CUDA will never work with vGPU K* (only with passthrough). CUDA with K180Q can be used but not allowed by NVidia (see [url]https://gridforums.nvidia.com/default/topic/735/[/url]). Technically due to NVidia stupid chips that are not able to use [b]preemtive[/b] multitasking up to Pascal generation and CUDA program can hung the chip (see [url]https://gridforums.nvidia.com/default/topic/743/[/url]). It is possible to enable CUDA but in Xen based hypervisor (see [url]https://gridforums.nvidia.com/default/topic/258/#5202[/url]).
CUDA will never work with vGPU K* (only with passthrough). CUDA with K180Q can be used but not allowed by NVidia (see https://gridforums.nvidia.com/default/topic/735/). Technically due to NVidia stupid chips that are not able to use preemtive multitasking up to Pascal generation and CUDA program can hung the chip (see https://gridforums.nvidia.com/default/topic/743/). It is possible to enable CUDA but in Xen based hypervisor (see https://gridforums.nvidia.com/default/topic/258/#5202).

#2
Posted 12/26/2017 10:01 PM   
Understood, thanks for the info. So I just changed it to do a straight passthrough (not vGPU) to the VM, installed the latest driver, but when trying to install the CUDA 8 driver, I still get the "no compatible hardware was found" message. What am I doing wrong?
Understood, thanks for the info. So I just changed it to do a straight passthrough (not vGPU) to the VM, installed the latest driver, but when trying to install the CUDA 8 driver, I still get the "no compatible hardware was found" message.
What am I doing wrong?

#3
Posted 12/27/2017 09:59 AM   
[quote=""]So I just changed it to do a straight passthrough (not vGPU) to the VM, installed the latest driver, but when trying to install the CUDA 8 driver, I still get the "no compatible hardware was found" message. What am I doing wrong?[/quote] Do not install [u]driver[/u] from "CUDA8 package" it is incompatible with Grid K1/K2 cards (even in passthrough mode) because package contains driver version [b]376.51[/b] (Grid K1/K2 cards are unsupported from version 375). Install only other components (like demos and header files if needed). The CUDA driver runtime (nvcuda32/64.dll ...) is already installed with driver 370.17.
said:So I just changed it to do a straight passthrough (not vGPU) to the VM, installed the latest driver, but when trying to install the CUDA 8 driver, I still get the "no compatible hardware was found" message. What am I doing wrong?

Do not install driver from "CUDA8 package" it is incompatible with Grid K1/K2 cards (even in passthrough mode) because package contains driver version 376.51 (Grid K1/K2 cards are unsupported from version 375). Install only other components (like demos and header files if needed). The CUDA driver runtime (nvcuda32/64.dll ...) is already installed with driver 370.17.

#4
Posted 12/27/2017 12:56 PM   
OK, thanks for your reply, I have a couple of questions: 1. Is CUDA 8 the last version that can be used with this K1 (I cant install or use CUDA 9, whihc is what I would prefer) ? 2. For CUDA 8, I can just install the driver package "370.17-quadro-winserv-2016-64bit-international-whql.exe" and CUDA 8 should work? 3. I would like to use this K1 card for some Mining (I know its not veru profitable, but it is spare currently, and I wanted to test), but whatever I use complains that it can't see any CUDA devices. Kind regards.
OK, thanks for your reply, I have a couple of questions:

1. Is CUDA 8 the last version that can be used with this K1 (I cant install or use CUDA 9, whihc is what I would prefer) ?
2. For CUDA 8, I can just install the driver package "370.17-quadro-winserv-2016-64bit-international-whql.exe" and CUDA 8 should work?
3. I would like to use this K1 card for some Mining (I know its not veru profitable, but it is spare currently, and I wanted to test), but whatever I use complains that it can't see any CUDA devices.

Kind regards.

#5
Posted 12/27/2017 04:48 PM   
Try to download GPU-Z and/or CUDA-Z to test CUDA in drivers ! I think that you are not limited by CUDA SDK Version (eg. compiler and header files) (CUDA SDK 9.0 has support for Compute Capability 3.0 – 7.x (Kepler, Maxwell, Pascal, Volta)) but you are limited by driver CUDA capabilities (seems to be version 8 in 370.17) and definitely by CUDA HW [b]Compute Capability[/b] (it is only 3.0 for GK107 (4 chips in GRID K1)). See [url]https://en.wikipedia.org/wiki/CUDA[/url] and continue in CUDA forums [url]https://devtalk.nvidia.com/[/url].
Try to download GPU-Z and/or CUDA-Z to test CUDA in drivers !
I think that you are not limited by CUDA SDK Version (eg. compiler and header files) (CUDA SDK 9.0 has support for Compute Capability 3.0 – 7.x (Kepler, Maxwell, Pascal, Volta)) but you are limited by driver CUDA capabilities (seems to be version 8 in 370.17) and definitely by CUDA HW Compute Capability (it is only 3.0 for GK107 (4 chips in GRID K1)). See https://en.wikipedia.org/wiki/CUDA and continue in CUDA forums https://devtalk.nvidia.com/.

#6
Posted 12/27/2017 09:07 PM   
Scroll To Top

Add Reply