Announcement

Collapse
No announcement yet.

Proof Of Concept: Open-Source Multi-GPU Rendering!

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • susikala
    replied
    This is the second time I see you write "has went" this month, Michael. Is this a common trend now in American English?

    When I was in school, we learnt 'I go, I went, I have gone', and I don't remember going over the dictionary lists for irregular verbs and it being any different there -- go / went / gone. So is this an American English / Internet trend?

    Leave a comment:


  • rohcQaH
    replied
    Originally posted by mtippett View Post
    I would imagine that part of Dave's evil plan would be to allow runtime switching between devices, but then allow existing clients continue to run until the former device finishes running clients - at which time it can be turned off.
    Wouldn't a single, long-running client cause both GPUs to be permanently turned on in that scenario?
    You know, long-running clients like your desktop or a window manager.

    The way I see it, client migration is a must for true GPU switching. It's already possible with an additional proxy between the client and X server (forgot the name of the program though), but that'd disable direct rendering.

    Leave a comment:


  • mtippett
    replied
    Originally posted by Zhick View Post
    AFAIK there's no way to seamlessly switch between gfx-cards, even on windows (let alone Linux/X). There'll always be a moment of black-screen.
    I like this approach. And what I find awesome is that nVidia advertises this feature as something totally new and exclusive to their new chips, and then airlied just goes ahead and implements it for completely different hardware. Goes to show how hardware-vendors try to use software-features to sell their hardware. And the name's awesome too.
    That's correct.

    The drivers and the hardware share a lot of state. The solution to get this going means pushing state further up so that it can be replicated to a secondary device. That or you pre-emptively doublepump state to allow transparent switching. This is what I believe NV does for their Xinerama accelerated 3D.

    I would imagine that part of Dave's evil plan would be to allow runtime switching between devices, but then allow existing clients continue to run until the former device finishes running clients - at which time it can be turned off.

    Regards,

    Matthew

    Leave a comment:


  • Zhick
    replied
    Originally posted by uwgandalf View Post
    Soo why don't you just switch seamlessly to the discrete gfx card to display your rendered stuff instead of having to go to these hacks to get just a portion of your discrete cards results composited in your onboards output?
    AFAIK there's no way to seamlessly switch between gfx-cards, even on windows (let alone Linux/X). There'll always be a moment of black-screen.
    I like this approach. And what I find awesome is that nVidia advertises this feature as something totally new and exclusive to their new chips, and then airlied just goes ahead and implements it for completely different hardware. Goes to show how hardware-vendors try to use software-features to sell their hardware. And the name's awesome too.

    Leave a comment:


  • uwgandalf
    replied
    soo?

    Soo why don't you just switch seamlessly to the discrete gfx card to display your rendered stuff instead of having to go to these hacks to get just a portion of your discrete cards results composited in your onboards output?

    Leave a comment:


  • Dragonlord
    replied
    I would really not call the windows way of doing things "good". I prefer an off-screen screen any time over having to create a visible window to create proper contexts.

    Leave a comment:


  • mar04
    replied
    Hey, lets clone airlied several times and we will have decent graphic stack in a few months

    Leave a comment:

Working...
X