NVIDIA's Server-Side GLVND Implementation

[phoronix]

Phoronix: NVIDIA's Server-Side GLVND Implementation

Earlier this week I wrote about Adam Jackson's work at Red Hat on a prototype of an X.Org server-side GLVND implementation. NVIDIA has also now announced their coincidentally similar work...

http://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-Server-GLVND

[bitman]

I hope they can come up with per-window opengl lib selection, otherwise it will be useless for optimus laptops once again.

[bug77]

I hope they can come up with per-window opengl lib selection, otherwise it will be useless for optimus laptops once again.

That would be nice, but how do you do per window rendering, when the monitor is physically connected to one video card or another? You'd need the window manager to dispatch each window to a video card, gather the output and compose a frame. Probably not impossible, but highly improbable if lesser things took this long to implement.

[cj.wijtmans]

That would be nice, but how do you do per window rendering, when the monitor is physically connected to one video card or another? You'd need the window manager to dispatch each window to a video card, gather the output and compose a frame. Probably not impossible, but highly improbable if lesser things took this long to implement.

You draw to a texture/buffer/plane whatever terminology and transfer it around. Not a difficult concept is it? Thats how i want my computer to boot my desktop onto the iGPU(even if my monitor is on my gfx card) then for games i want it to use my gfx card.

[bug77]

You draw to a texture/buffer/plane whatever terminology and transfer it around. Not a difficult concept is it? Thats how i want my computer to boot my desktop onto the iGPU(even if my monitor is on my gfx card) then for games i want it to use my gfx card.

Yeah, dead simple. And then you'd want V-sync, G-sync, FreeSync and what not to still work properly.

[cj.wijtmans]

Yeah, dead simple. And then you'd want V-sync, G-sync, FreeSync and what not to still work properly.

i dont see any reason why it couldnt work properly in such a scenario.

[polarathene]

That would be nice, but how do you do per window rendering, when the monitor is physically connected to one video card or another? You'd need the window manager to dispatch each window to a video card, gather the output and compose a frame. Probably not impossible, but highly improbable if lesser things took this long to implement.

You have bumblebee doing this atm with a 2nd xorg session that sends the rendered frame to the iGPU framebuffer which it then composes on the iGPU connected display. I thought this sort of work is to avoid the need for an additional server like bumblebee approach.

[ChristianSchaller]

Adam Jackson is on vacation now, but the plan is for him to work with NVidia once he is back on both server side glvnd and HDR support across X and Wayland. So if anyone was worried we would have two competing efforts here that is not in the cards.

[bug77]

You have bumblebee doing this atm with a 2nd xorg session that sends the rendered frame to the iGPU framebuffer which it then composes on the iGPU connected display. I thought this sort of work is to avoid the need for an additional server like bumblebee approach.

Afaik, bumblebee works to redirect the full screen, not individual windows. But I may be wrong.

[boxie]

You draw to a texture/buffer/plane whatever terminology and transfer it around. Not a difficult concept is it? Thats how i want my computer to boot my desktop onto the iGPU(even if my monitor is on my gfx card) then for games i want it to use my gfx card.

that sounds like a lot of expensive copying, which would hurt graphics performance, battery life (both in terms of bus usage and having 2 gpus somewhat active), and implementation of fancy features.