Grid driver for Ubuntu and CentOS?

Are there vGpu drivers for the Ubuntu and CentOS ?

We currently do not support vGPU in a Linux guest. That said, tell me more about your use case? What hypervisor stack are you on?

Will there be support for Linux guests?

I use the KVM hypervisor,I find I can download the linux version driver for GRID K2 ,that means the driver does’t support the vGPU?

Correct.

Currently only passthrough (direct connection) is available for Linux Guests. Also, KVM does not support vGPU it is currently only available on XenServer and will appear on vSphere in the next major release.

I use the KVM hypervisor, and I am interested how to implement that KVM support VGPU.Can you give me some advice? I hava bought the nvidia grid k1 and I readlly want to research the detail of GPU virtualization.

Carter, South China University of Technology.
Email:541658708@qq.com

KVM currently only supports GPU passthrough.

Note that in an update to my comments above, GRID 2.0 now supports vGPU in Linux for Workstation profiles.

Jason, what do you mean by:

"GRID 2.0 now supports vGPU in Linux for Workstation profiles."

Are you simply referring to VFIO aka passthrough for the Grid GPUs ? I am pretty sure I saw the updated kernel from this year get vGPU support. Any thoughts ? Thanks.

Jason, what do you mean by:

"GRID 2.0 now supports vGPU in Linux for Workstation profiles."

Are you simply referring to VFIO aka passthrough for the Grid GPUs ? I am pretty sure I saw the updated kernel from this year get vGPU support. Any thoughts ? Thanks.

NVidia marketing decided to exclude support for vGPU attached to Linux guests (on XenServer/VMware hypervisors) with "old" grid hardware K1/K2 for now.

You can modify Linux guest nvidia paravirtualized drivers (not kernel) to enable this (https://gridforums.nvidia.com/default/topic/543/general-discussion/grid-2-0/post/1924/#1924, https://gridforums.nvidia.com/default/topic/543/general-discussion/grid-2-0/post/2351/#2351)

or "Jason Southern: You could submit a feature enhancement request via you Nvidia contact (either IBD, PBM or SA)"

or buy AMD solution (http://www.amd.com/en-us/solutions/professional/virtualization#) with standardized SR-IOV/FLR solution that should generally work with all SR-IOV/FLR aware virtualization hypervisors.

M.C>

I mean exactly what it say, that vGPU for Linux guests is supported in GRID 2.0. (Virtual Workstation and above). So you can share the single gpu with multiple Linux VM’s.

http://www.nvidia.com/object/grid-technology.html

Passthrough is supported in all versions of GRID.

mcerveny believes that all new features should be back ported, unfortunately our Engineering feels otherwise.

Thanks MCerveny and Jason.

We are looking to leverage NV Grid for VDI, we’d like to use Red Hat RHEV to do this. I was informed that RH is working with Nvidia to do a grid manager integration into the KVM hypervisor before we can take advantage of vGPU. What would be the recommended Nvidia cards for this, would it still be the K1 and K2 or a different one ? aka newer model.

PS: The hosts are on RHEV and VMs are Windows 7/10 Enterprise. We may even throw terradicci into the mix to output to a zero client.

We can’t comment on this until it’s released, though it’s safe to assume the latest release will be supported.

If you’re using Windows guests today, consider XenServer as an alternative with full support of vGPU. You can trail on this platfrom then migrate to RHEV once it’s released.

When I install Nvidia Grid driver for Redhat 7 or Centos, system show bellow error:

ERROR: Unable to load the kernel module ‘nvidia.ko’.

How can I solve this problem?
[My Nvidia card is Grid K2 and I want ]

Generally, you use bad driver for the card or card does not exist or card is used by other driver (nouveau). You can check official NVidia knowledge base http://nvidia.custhelp.com/app/answers/list/st/5/kw/grid/page/1 or http://nvidia.custhelp.com/app or https://devtalk.nvidia.com/default/board/98/linux/ or you can study install guides for NVidia drivers or you can study RachelBerry blog https://virtuallyvisual.wordpress.com/

You should share more information like

  • centos is installed directly on hardware (baremetal) …
  • if no, centos is installed as guest (virtual machine) …
  • if yes, kvm,xen,xenserver,esxi,hyperv … is used …
  • if *, you pass K2 as vGPU (shared) or as passthrough (dedicated, DDA, vDGA) …
  • if *, you see card in guest/baremetal centos with "lspci; lspci -nvv"
  • if yes, you can use driver from standard download location http://www.nvidia.com/Download/index.aspx?lang=en-us -> “GRID” -> “GRID Series” (for passthrough or baremetal) or “NVIDIA GRID vGPU” (for vGPU, but unsupported for linux guest until you modify it (see previous post))
  • you can see any relevant errors in log ("dmesg | grep -i nvidia" or "dmesg | grep -i nvrm" …)

Linux support was added in GRID 2.0 and up which is available on newer Maxwell cards e.g. M60/10 - in the driver release notes it stipulates which versions/genres of linux are supported for each version.

Best wishes,
Rachel

We would like to do a demo with Dell R730, VMWare, K2 Grid Card and Linux VM. We are willing to use K2 for their HPC trainings. Since all the programs run on Linux, this has to be a Linux VM. Below, please find the list of the products/software and their version. We are facing a problem while installing the driver on Linux VM. The error was; “ERROR: Unable to load the kernel module ‘nvidia.ko’.” when we run Nvidia-Grid driver.

Dell Poweredge R730 (Dual E5-2650 v3, 128GB, 4*100GB SAS HDD)

Single Nvidia K2 Grid Card

VMware ESXI 6.0 ver. 4192238 , Vcenter 6.0 ver. 4637290

Vmware Horizon 7.0.2-4350299 on Windows 2012R2

Linux VM as Centos 7.2

Drivers used K2 Grid Card Installation for host : NVIDIA-VMware_ESXi_6.0_Host_Driver_367.64-1OEM.600.0.0.2494585.vib

And for Linux VM : NVIDIA-Linux-x86_64-367.64-grid.run

If you could lead and/or help us, we would highly appreciate it.

Ok, some information finally posted.

There is only one supported option for linux K2 and ESXi - vDGA (eg. passthrough), you can attach one linux guest to one chip and K2 has two chips on board. CUDA will be available too.

The vGPU system (eg. GRID with described drivers) is unsupported for linux for K1/K2 (eg. K2xx and K1xx vGPU profiles) and you get your “ERROR:…” message in guest. You can modify NVIDIA-Linux-x86_64-367.64-grid.run guest driver (or any corresponding NVidia linux standard guest driver) to get vGPU working in linux guest (about 10 bytes to modify, I found it in 20 minutes - https://gridforums.nvidia.com/default/topic/543/general-discussion/grid-2-0/post/1924/#1924). BUT CUDA will NOT work ! CUDA is available only for “full chip” assignment with newer cards M60/M6/M10 (vGPU profiles M10-8Q, M6-8Q, M60-8Q) (http://nvidia.custhelp.com/app/answers/detail/a_id/4189/kw/cuda%20grid)

Better and cheaper solution for CUDA processing is to use Quadro cards. Quadro >=K2000, >=M2000 should work in vDGA mode but NVidia does not guarantee anything (or publish updated list of supported cards for passthrough). You must check compatibility list https://www.vmware.com/resources/compatibility/search.php?deviceCategory=vdga.