When will ovirt add GPU/vGPU support in the admin portal ?
Since I recently saw cloudstack have add this( https://cwiki.apache.org/confluence/display/CLOUDSTACK/GPU+and+vGPU+support+... ). And iGVT( https://01.org/zh/xen/blogs/wangbo85/2014/intel-gvt-gxengt-pubic-release?lan...) is released. If ovirt add this, it'll be very competitive for ovirt in VDI platforms.
This is a multi-part message in MIME format. --------------030106030007070401060908 Content-Type: text/plain; charset="windows-1252" Content-Transfer-Encoding: quoted-printable On 01/19/2015 02=3A33 AM=2C lof yer wrote=3A =3E =3E Since I recently saw cloudstack have add this=28 =3E https=3A//cwiki=2Eapache=2Eorg/confluence/display/CLOUDSTACK/GPU+and+vG= PU+support+for+CloudStack+Guest+VMs =3E =29=2E =3E And iGVT=28 =3E https=3A//01=2Eorg/zh/xen/blogs/wangbo85/2014/intel-gvt-gxengt-pubic-re= lease=3Flangredirect=3D1=29 =3E is released=2E =3E If ovirt add this=2C it=27ll be very competitive for ovirt in VDI platf= orms=2E =3E See below the response I got about 1 year ago=2C I don=27t know the current= plans/status=2E=2E=2E =3E -------- Forwarded Message -------- =3E Subject=3A =09Re=3A =5BSpice-devel=5D =5BUsers=5D desktop virtualizatio= n and GPU =3E support =3E Date=3A =09Wed=2C 16 Oct 2013 14=3A08=3A47 +0200 =3E From=3A =09David Ja=9Aa =3Cdjasa=40redhat=2Ecom=3E =3E Reply-To=3A =09spice-devel=40lists=2Efreedesktop=2Eorg=2C users =3Cuser= s=40ovirt=2Eorg=3E =3E CC=3A =09spice-devel=40lists=2Efreedesktop=2Eorg=2C users =3Cusers=40ov= irt=2Eorg=3E =3E =3E =3E =3E Hi=2C =3E =3E I=27m afraid that none of these cards will work=2E IIRC it was explaine= d for =3E nVidia approach that it would be necessary to add their closed-source= =3E code to qemu which is not going to happen=2E =3E You might have better luck with AMD card =22passthrough=22 approach but= I=27m =3E not sure if GPUs can be passed through qemu/kvm and if so=2C how the =3E GPU-rendered image then gets displayed=2E So YMMV and don=27t expect mu= ch=2E =3E =3E There is a WIP effort for qemu though that is hardware vendor agnostic= - =3E it=27s based on creation of virtual GPU in qemu called Virgil that pass= es =3E 3D drawing commands to GPU on host=2E The project is in initial stage= =3E though so it won=27t be ready even for testing for quite some time=2E F= WIW=2C =3E there has been discussion about it on spice-devel in recent days and it= =3E will be presented on upcoming KVM forum so you can get pretty clear =3E picture about its status from these sources=2E =3E =3E HTH=2C =3E =3E David Met vriendelijke groet=2C With kind regards=2C Jorick Astrego Netbulae Virtualization Experts=20 ---------------- =09Tel=3A 053 20 30 270 =09info=40netbulae=2Eeu =09Staalsteden 4-3A =09KvK= 08198180 =09Fax=3A 053 20 30 271 =09www=2Enetbulae=2Eeu =097547 TA Enschede =09BTW= NL821234584B01 ---------------- --------------030106030007070401060908 Content-Type: text/html; charset="windows-1252" Content-Transfer-Encoding: quoted-printable =3Chtml=3E =3Chead=3E =3Cmeta content=3D=22text/html=3B charset=3Dwindows-1252=22 http-equiv=3D=22Content-Type=22=3E =3C/head=3E =3Cbody bgcolor=3D=22=23FFFFFF=22 text=3D=22=23000000=22=3E =3Cbr=3E =3Cdiv class=3D=22moz-cite-prefix=22=3EOn 01/19/2015 02=3A33 AM=2C lof= yer wrote=3A=3Cbr=3E =3C/div=3E =3Cblockquote cite=3D=22mid=3ACAKUHWYpedYihywiMGHfpi7dPpejZi6042x+FgLO0af7dV6-33w=40mail= =2Egmail=2Ecom=22 type=3D=22cite=22=3E =3Cp dir=3D=22ltr=22=3ESince I recently saw cloudstack have add this= =28 =3Ca moz-do-not-send=3D=22true=22 href=3D=22https=3A//cwiki=2Eapache=2Eorg/confluence/display/CLOUDSTACK/GPU+= and+vGPU+support+for+CloudStack+Guest+VMs=22=3Ehttps=3A//cwiki=2Eapache=2Eo= rg/confluence/display/CLOUDSTACK/GPU+and+vGPU+support+for+CloudStack+Guest+= VMs=3C/a=3E =29=2E=3Cbr=3E And iGVT=28 =3Ca moz-do-not-send=3D=22true=22 href=3D=22https=3A//01=2Eorg/zh/xen/blogs/wangbo85/2014/intel-gvt-gxengt-pu= bic-release=3Flangredirect=3D1=22=3Ehttps=3A//01=2Eorg/zh/xen/blogs/wangbo8= 5/2014/intel-gvt-gxengt-pubic-release=3Flangredirect=3D1=3C/a=3E=29 is released=2E=3Cbr=3E If ovirt add this=2C it=27ll be very competitive for ovirt in VDI= platforms=2E=3C/p=3E =3C/blockquote=3E See below the response I got about 1 year ago=2C I don=27t know the current plans/status=2E=2E=2E=3Cbr=3E =3Cbr=3E =3Cblockquote type=3D=22cite=22=3E-------- Forwarded Message --------= =3Ctable class=3D=22moz-email-headers-table=22 border=3D=220=22 cellp= adding=3D=220=22 cellspacing=3D=220=22=3E =3Ctbody=3E =3Ctr=3E =3Cth align=3D=22RIGHT=22 nowrap=3D=22nowrap=22 valign=3D=22BAS= ELINE=22=3ESubject=3A =3C/th=3E =3Ctd=3ERe=3A =5BSpice-devel=5D =5BUsers=5D desktop virtualizat= ion and GPU support=3C/td=3E =3C/tr=3E =3Ctr=3E =3Cth align=3D=22RIGHT=22 nowrap=3D=22nowrap=22 valign=3D=22BAS= ELINE=22=3EDate=3A =3C/th=3E =3Ctd=3EWed=2C 16 Oct 2013 14=3A08=3A47 +0200=3C/td=3E =3C/tr=3E =3Ctr=3E =3Cth align=3D=22RIGHT=22 nowrap=3D=22nowrap=22 valign=3D=22BAS= ELINE=22=3EFrom=3A =3C/th=3E =3Ctd=3EDavid Ja=9Aa =3Ca class=3D=22moz-txt-link-rfc2396E=22 h= ref=3D=22mailto=3Adjasa=40redhat=2Ecom=22=3E=26lt=3Bdjasa=40redhat=2Ecom=26= gt=3B=3C/a=3E=3C/td=3E =3C/tr=3E =3Ctr=3E =3Cth align=3D=22RIGHT=22 nowrap=3D=22nowrap=22 valign=3D=22BAS= ELINE=22=3EReply-To=3A =3C/th=3E =3Ctd=3E=3Ca class=3D=22moz-txt-link-abbreviated=22 href=3D=22m= ailto=3Aspice-devel=40lists=2Efreedesktop=2Eorg=22=3Espice-devel=40lists=2E= freedesktop=2Eorg=3C/a=3E=2C users =3Ca class=3D=22moz-txt-link-rfc2396E=22 href=3D=22mailto=3Au= sers=40ovirt=2Eorg=22=3E=26lt=3Busers=40ovirt=2Eorg=26gt=3B=3C/a=3E=3C/td= =3E =3C/tr=3E =3Ctr=3E =3Cth align=3D=22RIGHT=22 nowrap=3D=22nowrap=22 valign=3D=22BAS= ELINE=22=3ECC=3A =3C/th=3E =3Ctd=3E=3Ca class=3D=22moz-txt-link-abbreviated=22 href=3D=22m= ailto=3Aspice-devel=40lists=2Efreedesktop=2Eorg=22=3Espice-devel=40lists=2E= freedesktop=2Eorg=3C/a=3E=2C users =3Ca class=3D=22moz-txt-link-rfc2396E=22 href=3D=22mailto=3Au= sers=40ovirt=2Eorg=22=3E=26lt=3Busers=40ovirt=2Eorg=26gt=3B=3C/a=3E=3C/td= =3E =3C/tr=3E =3C/tbody=3E =3C/table=3E =3Cbr=3E =3Cbr=3E =3Cpre=3EHi=2C I=27m afraid that none of these cards will work=2E IIRC it was explained fo= r nVidia approach that it would be necessary to add their closed-source code to qemu which is not going to happen=2E You might have better luck with AMD card =22passthrough=22 approach but I= =27m not sure if GPUs can be passed through qemu/kvm and if so=2C how the GPU-rendered image then gets displayed=2E So YMMV and don=27t expect much= =2E There is a WIP effort for qemu though that is hardware vendor agnostic - it=27s based on creation of virtual GPU in qemu called Virgil that passes= 3D drawing commands to GPU on host=2E The project is in initial stage though so it won=27t be ready even for testing for quite some time=2E FWIW= =2C there has been discussion about it on spice-devel in recent days and it will be presented on upcoming KVM forum so you can get pretty clear picture about its status from these sources=2E HTH=2C David =3C/pre=3E =3C/blockquote=3E =3Cbr=3E =20= =3CBR /=3E =3CBR /=3E =3Cb style=3D=22color=3A=23604c78=22=3E=3C/b=3E=3Cbr=3E=3Cspan style=3D=22c= olor=3A=23604c78=3B=22=3E=3Cfont color=3D=22000000=22=3E=3Cspan style=3D=22= mso-fareast-language=3Aen-gb=3B=22 lang=3D=22NL=22=3EMet vriendelijke groet= =2C With kind regards=2C=3Cbr=3E=3Cbr=3E=3C/span=3EJorick Astrego=3C/font= =3E=3C/span=3E=3Cb style=3D=22color=3A=23604c78=22=3E=3Cbr=3E=3Cbr=3ENetbul= ae Virtualization Experts =3C/b=3E=3Cbr=3E=3Chr style=3D=22border=3Anone=3B= border-top=3A1px solid =23ccc=3B=22=3E=3Ctable style=3D=22width=3A 522px=22= =3E=3Ctbody=3E=3Ctr=3E=3Ctd style=3D=22width=3A 130px=3Bfont-size=3A 10px= =22=3ETel=3A 053 20 30 270=3C/td=3E =3Ctd style=3D=22width=3A 130px=3Bf= ont-size=3A 10px=22=3Einfo=40netbulae=2Eeu=3C/td=3E =3Ctd style=3D=22wid= th=3A 130px=3Bfont-size=3A 10px=22=3EStaalsteden 4-3A=3C/td=3E =3Ctd sty= le=3D=22width=3A 130px=3Bfont-size=3A 10px=22=3EKvK 08198180=3C/td=3E=3C/tr= =3E=3Ctr=3E =3Ctd style=3D=22width=3A 130px=3Bfont-size=3A 10px=22=3EFax= =3A 053 20 30 271=3C/td=3E =3Ctd style=3D=22width=3A 130px=3Bfont-size= =3A 10px=22=3Ewww=2Enetbulae=2Eeu=3C/td=3E =3Ctd style=3D=22width=3A 130= px=3Bfont-size=3A 10px=22=3E7547 TA Enschede=3C/td=3E =3Ctd style=3D=22w= idth=3A 130px=3Bfont-size=3A 10px=22=3EBTW NL821234584B01=3C/td=3E=3C/tr=3E= =3C/tbody=3E=3C/table=3E=3Cbr=3E=3Chr style=3D=22border=3Anone=3Bborder-top= =3A1px solid =23ccc=3B=22=3E=3CBR /=3E =3C/body=3E =3C/html=3E --------------030106030007070401060908--
----- Original Message -----
From: "Jorick Astrego" <j.astrego@netbulae.eu> To: users@ovirt.org Sent: Monday, January 19, 2015 10:14:35 AM Subject: Re: [ovirt-users] When will ovirt add GPU/vGPU support in the admin portal ?
On 01/19/2015 02:33 AM, lof yer wrote:
Since I recently saw cloudstack have add this( https://cwiki.apache.org/confluence/display/CLOUDSTACK/GPU+and+vGPU+support+... ). And iGVT( https://01.org/zh/xen/blogs/wangbo85/2014/intel-gvt-gxengt-pubic-release?lan...) is released. If ovirt add this, it'll be very competitive for ovirt in VDI platforms.
See below the response I got about 1 year ago, I don't know the current plans/status...
-------- Forwarded Message -------- Subject: Re: [Spice-devel] [Users] desktop virtualization and GPU support Date: Wed, 16 Oct 2013 14:08:47 +0200 From: David Jaša <djasa@redhat.com> Reply-To: spice-devel@lists.freedesktop.org, users <users@ovirt.org> CC: spice-devel@lists.freedesktop.org, users <users@ovirt.org>
Hi,
I'm afraid that none of these cards will work. IIRC it was explained for nVidia approach that it would be necessary to add their closed-source code to qemu which is not going to happen. You might have better luck with AMD card "passthrough" approach but I'm not sure if GPUs can be passed through qemu/kvm and if so, how the GPU-rendered image then gets displayed. So YMMV and don't expect much.
There is a WIP effort for qemu though that is hardware vendor agnostic - it's based on creation of virtual GPU in qemu called Virgil that passes 3D drawing commands to GPU on host. The project is in initial stage though so it won't be ready even for testing for quite some time. FWIW, there has been discussion about it on spice-devel in recent days and it will be presented on upcoming KVM forum so you can get pretty clear picture about its status from these sources.
HTH,
David
Hello, There is quite active effort to implement device passthrough (including GPU, nics, usb etc.)[1]. Physical GPU passthrough can be considered generic PCI passthrough which should be ready for oVirt 3.6. As for the vGPU, that will depend on technology used - SR-IOV would be supported along PCI passthrough, anything else might need additional work on our side. [1] http://www.ovirt.org/Features/hostdev_passthrough mpolednik
Met vriendelijke groet, With kind regards,
Jorick Astrego
Netbulae Virtualization Experts
----------------
Tel: 053 20 30 270 info@netbulae.eu Staalsteden 4-3A KvK 08198180 Fax: 053 20 30 271 www.netbulae.eu 7547 TA Enschede BTW NL821234584B01
----------------
_______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
On 19.01.2015 12:24, Martin Polednik wrote:
----- Original Message -----
From: "Jorick Astrego" <j.astrego@netbulae.eu> To: users@ovirt.org Sent: Monday, January 19, 2015 10:14:35 AM Subject: Re: [ovirt-users] When will ovirt add GPU/vGPU support in the admin portal ?
On 01/19/2015 02:33 AM, lof yer wrote:
Since I recently saw cloudstack have add this( https://cwiki.apache.org/confluence/display/CLOUDSTACK/GPU+and+vGPU+support+... ). And iGVT( https://01.org/zh/xen/blogs/wangbo85/2014/intel-gvt-gxengt-pubic-release?lan...) is released. If ovirt add this, it'll be very competitive for ovirt in VDI platforms.
See below the response I got about 1 year ago, I don't know the current plans/status...
-------- Forwarded Message -------- Subject: Re: [Spice-devel] [Users] desktop virtualization and GPU support Date: Wed, 16 Oct 2013 14:08:47 +0200 From: David Jaša <djasa@redhat.com> Reply-To: spice-devel@lists.freedesktop.org, users <users@ovirt.org> CC: spice-devel@lists.freedesktop.org, users <users@ovirt.org>
Hi,
I'm afraid that none of these cards will work. IIRC it was explained for nVidia approach that it would be necessary to add their closed-source code to qemu which is not going to happen. You might have better luck with AMD card "passthrough" approach but I'm not sure if GPUs can be passed through qemu/kvm and if so, how the GPU-rendered image then gets displayed. So YMMV and don't expect much.
There is a WIP effort for qemu though that is hardware vendor agnostic - it's based on creation of virtual GPU in qemu called Virgil that passes 3D drawing commands to GPU on host. The project is in initial stage though so it won't be ready even for testing for quite some time. FWIW, there has been discussion about it on spice-devel in recent days and it will be presented on upcoming KVM forum so you can get pretty clear picture about its status from these sources.
HTH,
David
Hello,
There is quite active effort to implement device passthrough (including GPU, nics, usb etc.)[1].Physical GPU passthrough can be considered generic PCI passthrough
IMHO not quite in case of GPUs. I tested that recently with a combination of EL7's new default VFIO support [1]. Two things could be concluded working with Windows 7 guests and ATI Fire Pro GPUs: - 'Old' PCI passthough unstable - VFIO working and performing better (tested with Cinebench OpenGL [2], about 98% of the bare metal performance) As for the stability, the results with VFIO where better in this case. The tricky part seems to be the driver installation in the guest, there the mandatory device probe in the installer. While PCI passthough resulted in Blue Screens, VFIO went though. [1] https://access.redhat.com/sites/default/files/pages/attachments/rhel_whatsne... [2] http://www.maxon.net/products/cinebench/overview.html which should be ready for oVirt 3.6. As for the vGPU, that will
depend on technology used - SR-IOV would be supported along PCI passthrough, anything else might need additional work on our side.
[1] http://www.ovirt.org/Features/hostdev_passthrough
mpolednik
Met vriendelijke groet, With kind regards,
Jorick Astrego
Netbulae Virtualization Experts
----------------
Tel: 053 20 30 270 info@netbulae.eu Staalsteden 4-3A KvK 08198180 Fax: 053 20 30 271 www.netbulae.eu 7547 TA Enschede BTW NL821234584B01
----------------
_______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
_______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
-- Daniel Helgenberger m box bewegtbild GmbH P: +49/30/2408781-22 F: +49/30/2408781-10 ACKERSTR. 19 D-10115 BERLIN www.m-box.de www.monkeymen.tv Geschäftsführer: Martin Retschitzegger / Michaela Göllner Handeslregister: Amtsgericht Charlottenburg / HRB 112767
----- Original Message -----
From: "Daniel Helgenberger" <daniel.helgenberger@m-box.de> To: "Martin Polednik" <mpolednik@redhat.com>, "Jorick Astrego" <j.astrego@netbulae.eu> Cc: users@ovirt.org Sent: Monday, January 19, 2015 3:45:01 PM Subject: Re: [ovirt-users] When will ovirt add GPU/vGPU support in the admin portal ?
On 19.01.2015 12:24, Martin Polednik wrote:
----- Original Message -----
From: "Jorick Astrego" <j.astrego@netbulae.eu> To: users@ovirt.org Sent: Monday, January 19, 2015 10:14:35 AM Subject: Re: [ovirt-users] When will ovirt add GPU/vGPU support in the admin portal ?
On 01/19/2015 02:33 AM, lof yer wrote:
Since I recently saw cloudstack have add this( https://cwiki.apache.org/confluence/display/CLOUDSTACK/GPU+and+vGPU+support+... ). And iGVT( https://01.org/zh/xen/blogs/wangbo85/2014/intel-gvt-gxengt-pubic-release?lan...) is released. If ovirt add this, it'll be very competitive for ovirt in VDI platforms.
See below the response I got about 1 year ago, I don't know the current plans/status...
-------- Forwarded Message -------- Subject: Re: [Spice-devel] [Users] desktop virtualization and GPU support Date: Wed, 16 Oct 2013 14:08:47 +0200 From: David Jaša <djasa@redhat.com> Reply-To: spice-devel@lists.freedesktop.org, users <users@ovirt.org> CC: spice-devel@lists.freedesktop.org, users <users@ovirt.org>
Hi,
I'm afraid that none of these cards will work. IIRC it was explained for nVidia approach that it would be necessary to add their closed-source code to qemu which is not going to happen. You might have better luck with AMD card "passthrough" approach but I'm not sure if GPUs can be passed through qemu/kvm and if so, how the GPU-rendered image then gets displayed. So YMMV and don't expect much.
There is a WIP effort for qemu though that is hardware vendor agnostic - it's based on creation of virtual GPU in qemu called Virgil that passes 3D drawing commands to GPU on host. The project is in initial stage though so it won't be ready even for testing for quite some time. FWIW, there has been discussion about it on spice-devel in recent days and it will be presented on upcoming KVM forum so you can get pretty clear picture about its status from these sources.
HTH,
David
Hello,
There is quite active effort to implement device passthrough (including GPU, nics, usb etc.)[1].Physical GPU passthrough can be considered generic PCI passthrough
IMHO not quite in case of GPUs. I tested that recently with a combination of EL7's new default VFIO support [1]. Two things could be concluded working with Windows 7 guests and ATI Fire Pro GPUs: - 'Old' PCI passthough unstable - VFIO working and performing better (tested with Cinebench OpenGL [2], about 98% of the bare metal performance)
I'm sorry, let me correct myself a bit: the legacy PCI-stub is not really supported at all, it's VFIO only (so "generic PCI" was meant as "generic VFIO"). What is not currently supported and will not be in a reasonable time frame is gVirt or overall intel GVT.
As for the stability, the results with VFIO where better in this case. The tricky part seems to be the driver installation in the guest, there the mandatory device probe in the installer. While PCI passthough resulted in Blue Screens, VFIO went though.
[1] https://access.redhat.com/sites/default/files/pages/attachments/rhel_whatsne... [2] http://www.maxon.net/products/cinebench/overview.html
which should be ready for oVirt 3.6. As for the vGPU, that will
depend on technology used - SR-IOV would be supported along PCI passthrough, anything else might need additional work on our side.
[1] http://www.ovirt.org/Features/hostdev_passthrough
mpolednik
Met vriendelijke groet, With kind regards,
Jorick Astrego
Netbulae Virtualization Experts
----------------
Tel: 053 20 30 270 info@netbulae.eu Staalsteden 4-3A KvK 08198180 Fax: 053 20 30 271 www.netbulae.eu 7547 TA Enschede BTW NL821234584B01
----------------
_______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
_______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
-- Daniel Helgenberger m box bewegtbild GmbH
P: +49/30/2408781-22 F: +49/30/2408781-10
ACKERSTR. 19 D-10115 BERLIN
www.m-box.de www.monkeymen.tv
Geschäftsführer: Martin Retschitzegger / Michaela Göllner Handeslregister: Amtsgericht Charlottenburg / HRB 112767
participants (4)
-
Daniel Helgenberger -
Jorick Astrego -
lof yer -
Martin Polednik