On 12/09/14 10:52, Elliot Hallmark wrote: > > Thanks. Those keys exist, but how would I get one? Nvidia website > says I need to have a personal contact at nvidia already to be considered. > I'm sorry I can't comment on that any more than I already have in my previous rant: http://lists.devloop.org.uk/pipermail/shifter-users/2014-June/000888.html
Only to say that the NVENC SDK version 4 is now out and that I will take a look at it soon to see what broke: http://xpra.org/trac/ticket/653 > > Yes, I forgot about virtualgl. I had got a two head system working > and forgot that not dealing with video connections on the card would > free me up. > > I am fine with serving applications, though a virtual machine could be > an application if that ends up being useful. I intend to make a > python based menu on the clients for choosing applications, rather > than give clients the whole desktop. > That's what winswitch does, and more. If you only care about xpra and access over ssh, you can easily write something more simple. Cheers Antoine > On Sep 11, 2014 10:14 PM, "Antoine Martin" <[email protected] > <mailto:[email protected]>> wrote: > > On 12/09/14 09:29, Elliot Hallmark wrote: > > I just want to check in and make sure I am thinking correctly. > My aim is > > to serve hardware opengl accelerated applications to thin clients. > > > > I wanted to use nvenc from a gtx 660, but this requires a > developer key. > Those keys exist. I haven't tested them myself with GTX6XXs, only with > GTX7XXs, but I believe others have. > > So, for a prototype before investing in a $2000 grid card ( k1 > and k2 > > support nvenc through xpra without a dev key), I will use the > x264 encoder. > If you want to use a pro card to avoid the license key issue, > there are > cheaper options than the grid cards. > The NVENC chips on those cards are not faster than in the consumer > cards > (if anything slower because of the lower clock and older > architecture), > the only benefit of the grid cards is that they have multiple > chips per > card. (4 in a K1, 2 in a K2) > > Clients will run a stripped down linux that only runs xpra > client in full > > screen, ultimately with a python based application or session > selector > > interface. Server will serve applications (potentially virtualized > > desktops) at the full screen resolution of the client. The > server will > > render applications on the gtx 660 before encoding. > If I understand this correctly, you want to use xpra to serve a full > desktop instead of individual windows? > That would work but it would be less efficient than letting the > clients > manage the windows themselves. > > I believe this > > requires some hacking to make each application in a common X > session (ie > > :100.1, 100.2, etc) inorder to "share" the gpu. > I'm not sure I understand this part. > If you want multiple server sessions getting accelerated OpenGL > rendering, you should look at VirtualGL. > If you want to share the NVENC encoding chip for multiple server > sessions, then it is just a permission issue on the video device. > > I can do the last bit with X, ie I had two monitors displaying > seperate > > hardware accelerated programs (like minecraft) from a single gpu > through > > creative use of xorg.conf, so I'm assuming it can happen with > xpra too. > This sounds like a client-side setup. > > Has anyone done this before? Any heads up would be appreciated. > Hope this helps! > > Cheers > Antoine > > > > Thanks, > > Elliot > > _______________________________________________ > > shifter-users mailing list > > [email protected] > <mailto:[email protected]> > > http://lists.devloop.org.uk/mailman/listinfo/shifter-users > > _______________________________________________ > shifter-users mailing list > [email protected] > <mailto:[email protected]> > http://lists.devloop.org.uk/mailman/listinfo/shifter-users > _______________________________________________ shifter-users mailing list [email protected] http://lists.devloop.org.uk/mailman/listinfo/shifter-users
