Login | Register For Free | Help
Search for: (Advanced)

Mailing List Archive: Xen: Users

VGA Passthrough (NVIDIA) - current status?

 

 

Xen users RSS feed   Index | Next | Previous | View Threaded


equilibrium87 at gmail

Dec 30, 2012, 2:47 PM

Post #1 of 3 (1259 views)
Permalink
VGA Passthrough (NVIDIA) - current status?

Hi everyone,

Sorry if this gets asked over and over again, but I just joined the list
and wanted to ask what the current status of the VGA-passthrough
possibilities is.
Today I set up my new workstation that has a Kepler card in it and compiled
XEN 4.2.1 on Ubuntu 12.04. The first problem I ran into was that the system
just hung up after the login screen, which (after quite some
google-research) seems to be a common problem caused by proprietary nvidia
drivers used
(I am using the experimental 310 binaries, which are required by CUDA 5.x).

What I want to achieve is the following: Being able to spawn several VMs
such that I am not required to mess around with my base OS if I am working
on
university assignments. Therefore, it can be assumed, that only one VM will
the GPU at a time. At the same time, close to native performance would be
nice.

I have read list entries from aprox. 2 years ago covering the topic
(applying patches, extracting the card's BIOS,...) - but is there something
more recent
that someone could point me to? Is there anyone who successfully runs a
Kepler card + CUDA in his/her VM?

Any help would be much appreciated.
Happy new year & Regards,
Paul Mayer


cdelorme at gmail

Dec 30, 2012, 3:03 PM

Post #2 of 3 (1245 views)
Permalink
Re: VGA Passthrough (NVIDIA) - current status? [In reply to]

I have had great success with an AMD consumer card. Some minor issues with
FLR that aren't well explained, but I put together a comprehensive tutorial
on Debian (similar to Ubuntu I'm sure) with a video walkthrough as well:
http://wiki.xen.org/wiki/Comprehensive_Xen_Debian_Wheezy_PCI_Passthrough_Tutorial

To my understanding nVidia takes a bit of extra effort, generally involving
some Xen source patching, but I do not know whether that is required for
their server grade cards (which I believe are built to support IOMMU). For
patching details check out David Techer's Blog:
http://www.davidgis.fr/blog/index.php?Xen

Maybe someone else can supply their experience with nVidia Server cards?


On Sun, Dec 30, 2012 at 5:47 PM, Paul Mayer <equilibrium87 [at] gmail> wrote:

> Hi everyone,
>
> Sorry if this gets asked over and over again, but I just joined the list
> and wanted to ask what the current status of the VGA-passthrough
> possibilities is.
> Today I set up my new workstation that has a Kepler card in it and
> compiled XEN 4.2.1 on Ubuntu 12.04. The first problem I ran into was that
> the system
> just hung up after the login screen, which (after quite some
> google-research) seems to be a common problem caused by proprietary nvidia
> drivers used
> (I am using the experimental 310 binaries, which are required by CUDA 5.x).
>
> What I want to achieve is the following: Being able to spawn several VMs
> such that I am not required to mess around with my base OS if I am working
> on
> university assignments. Therefore, it can be assumed, that only one VM
> will the GPU at a time. At the same time, close to native performance would
> be
> nice.
>
> I have read list entries from aprox. 2 years ago covering the topic
> (applying patches, extracting the card's BIOS,...) - but is there something
> more recent
> that someone could point me to? Is there anyone who successfully runs a
> Kepler card + CUDA in his/her VM?
>
> Any help would be much appreciated.
> Happy new year & Regards,
> Paul Mayer
>
> _______________________________________________
> Xen-users mailing list
> Xen-users [at] lists
> http://lists.xen.org/xen-users
>


gizmochicken at gmail

Dec 31, 2012, 12:48 AM

Post #3 of 3 (1227 views)
Permalink
Re: VGA Passthrough (NVIDIA) - current status? [In reply to]

Paul,

Apart from their high end workstation cards, such as the Quadro 2000,
Nvidia cards aren't known to be Xen friendly. On the other hand, even
mid-range AMD Radeon cards have been reported to support VGA-passthrough.
(See http://wiki.xen.org/wiki/XenVGAPassthroughTestedAdapters) So to save
yourself many headaches, I recommend that you invest in a mid-range (or
better) AMD Radeon card. In my case, I found that a GigaByte AMD Radeon HD
6670 (about $60 on sale) supported VGA-passthrough with an HVM domU.

For a well-written and very detailed tutorial describing something similar
to what are describing, have a look here:
http://forums.linuxmint.com/viewtopic.php?f=42&t=112013

Best regards,
GizmoChicken


On Sun, Dec 30, 2012 at 6:03 PM, Casey DeLorme <cdelorme [at] gmail> wrote:

> I have had great success with an AMD consumer card. Some minor issues
> with FLR that aren't well explained, but I put together a comprehensive
> tutorial on Debian (similar to Ubuntu I'm sure) with a video walkthrough as
> well:
>
> http://wiki.xen.org/wiki/Comprehensive_Xen_Debian_Wheezy_PCI_Passthrough_Tutorial
>
> To my understanding nVidia takes a bit of extra effort, generally
> involving some Xen source patching, but I do not know whether that is
> required for their server grade cards (which I believe are built to support
> IOMMU). For patching details check out David Techer's Blog:
> http://www.davidgis.fr/blog/index.php?Xen
>
> Maybe someone else can supply their experience with nVidia Server cards?
>
>
> On Sun, Dec 30, 2012 at 5:47 PM, Paul Mayer <equilibrium87 [at] gmail>wrote:
>
>> Hi everyone,
>>
>> Sorry if this gets asked over and over again, but I just joined the list
>> and wanted to ask what the current status of the VGA-passthrough
>> possibilities is.
>> Today I set up my new workstation that has a Kepler card in it and
>> compiled XEN 4.2.1 on Ubuntu 12.04. The first problem I ran into was that
>> the system
>> just hung up after the login screen, which (after quite some
>> google-research) seems to be a common problem caused by proprietary nvidia
>> drivers used
>> (I am using the experimental 310 binaries, which are required by CUDA
>> 5.x).
>>
>> What I want to achieve is the following: Being able to spawn several VMs
>> such that I am not required to mess around with my base OS if I am working
>> on
>> university assignments. Therefore, it can be assumed, that only one VM
>> will the GPU at a time. At the same time, close to native performance would
>> be
>> nice.
>>
>> I have read list entries from aprox. 2 years ago covering the topic
>> (applying patches, extracting the card's BIOS,...) - but is there something
>> more recent
>> that someone could point me to? Is there anyone who successfully runs a
>> Kepler card + CUDA in his/her VM?
>>
>> Any help would be much appreciated.
>> Happy new year & Regards,
>> Paul Mayer
>>
>> _______________________________________________
>> Xen-users mailing list
>> Xen-users [at] lists
>> http://lists.xen.org/xen-users
>>
>
>
> _______________________________________________
> Xen-users mailing list
> Xen-users [at] lists
> http://lists.xen.org/xen-users
>

Xen users RSS feed   Index | Next | Previous | View Threaded
 
 


Interested in having your list archived? Contact Gossamer Threads
 
  Web Applications & Managed Hosting Powered by Gossamer Threads Inc.