RE: GPU instance with debian HVM AMI?
The Debian AMIs are not yet blessed for G2. They would at this stage require the nvidia drivers at this stage, and as this would conflict with the Debian Free Software Gidelines, I have not (yet) produced a Debian + NVidia AMI. I am hoping that Nouveau will be blessed for use on this instance type in a few weeks, and at that stage I'll apply for the AMI to be approved for this instance type.
There are other issues that I have seen with the Debian AMI on the g1 instance - I am trying to find some time for me to look deeper into this (probably next week).
James Bromberger | Solution Architect | Amazon Web Services
E: email@example.com P: +61 422 166 708 T:@JamesBromberger
> -----Original Message-----
> From: Aryeh Leib Taurog [mailto:firstname.lastname@example.org]
> Sent: Thursday, 20 February 2014 6:39 AM
> To: email@example.com
> Subject: GPU instance with debian HVM AMI?
> I would like to launch GPU instances with the debian HVM AMI:
> debian-wheezy-amd64-hvm-2014-02-08-ebs - ami-9f675df6
> It doesn't allow me to do this. Apparently only certain AMIs are supported.
> I posted an inquiry about this on the EC2 forums.
> I received the following response:
> > I have investigated the issue in regards to the AMI - ami-9f675df6 and
> > unfortunately it would appear the instance type of g2.2xlarge is not
> > supported by this particular image. For the AMI itself it may be
> > lacking drivers and/or internal support for the g2.2xlarge instance type.
> > I have however been able to successfully launch the same AMI using a
> > variety of other instance types such as C3, M3 etc.
> > This is not something we can simply 'enable', it is a restriction of
> > the particular AMI itself. I would recommend either selecting a
> > difference Instance type during the provisioning or selecting a
> > different AMI that provides the features you are looking for.
> Can anyone shed any light on this? Could there be some missing
> configuration which is required for PCI passthrough? The Xen wiki indicates
> > HVM guests require no special configuration for the guest kernel, as
> > all accesses are emulated and virtualized by the IOMMU hardware.
> Aryeh Leib Taurog
> To UNSUBSCRIBE, email to firstname.lastname@example.org
> with a subject of "unsubscribe". Trouble? Contact email@example.com