Announcement

Collapse
No announcement yet.

Radeon Gets Multi-Screen Reverse Optimus Support

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

  • Radeon Gets Multi-Screen Reverse Optimus Support

    Phoronix: Radeon Gets Multi-Screen Reverse Optimus Support

    Hours after David Airlie posted a new Nouveau X.Org driver with "multi-screen reverse Optimus" support, AMD's Alex Deucher ported the code to the open-source Radeon driver...

    http://www.phoronix.com/vr.php?view=MTQyNDA

  • #2
    Alex is on fire lately

    Comment


    • #3
      Does this mean we might finally be able to have two gpus driving 4 displays, but with a single gpu drawing everything (so that there no xinerama and etc issues)?
      That would be an awesome way to have 4 displays for working with old crappy graphics cards.

      Comment


      • #4
        I think what this means is that you can have your monitors plugged into your discrete GPU, but have your integrated Intel GPU doing the actual rendering.

        Comment


        • #5
          Originally posted by duby229 View Post
          I think what this means is that you can have your monitors plugged into your discrete GPU, but have your integrated Intel GPU doing the actual rendering.
          Yes, but my use case is more: 2 radeon gpus, 2 monitors attached each, and having radeon gpu 1 doing the rendering for them all (don't really care about gaming, just pretty opengl desktop).

          Is the solution generic enough so that I can replace "intel gpu" with "radeon gpu", is basically what I'm asking.

          Comment


          • #6
            What is the real world use of this?

            Have there been any AMD-based laptops which have an Nvidia discrete GPU built into them?

            Have there been any AMD-based laptops which have an Intel discrete GPU built into them?

            Have there been any Intel-based laptops which have an AMD discrete GPU built into them?

            Have there been any AMD-based laptops which have a more powerful AMD discrete GPU?

            Or is this part of some "foundational groundwork" which will help lead to crossfire/SLI support in the OSS driver? (Ditto for nouveau)

            Or...... ?

            I haven't looked at laptops in quite a long time, so I genuinely do not know. My old laptop still serves me well for now.

            Comment


            • #7
              Background on reverse optimus:
              http://airlied.livejournal.com/76723.html

              Comment


              • #8
                Originally posted by halfmanhalfamazing View Post
                1) Have there been any AMD-based laptops which have an Nvidia discrete GPU built into them?
                2) Have there been any AMD-based laptops which have an Intel discrete GPU built into them?
                3) Have there been any Intel-based laptops which have an AMD discrete GPU built into them?
                4) Have there been any AMD-based laptops which have a more powerful AMD discrete GPU?
                1) Not that I recall, but it wouldn't surprise me.
                2) No (Intel don't make discrete GPUs)
                3) Yes
                4) Yes (there have also been laptops with a weak discrete GPU and a more powerful integrated GPU (e.g. discrete HD 7450, integrated HD 6650))

                :-)

                Comment


                • #9
                  Well what about an OpenCL render slave? Will the OSS drivers allow you to use an AMD APU as strictly an OCL churner while the dedicated GPU does all the graphics?

                  What about mixed GPU generations and vendors? Can you have an Nvidia iGPU running Nouveau with an AMD GPU all extending the desktop?

                  Comment


                  • #10
                    Hello!

                    Hello everyone I'm new to the forum and good Phoronix, the issue I'm interested Amd eto know if my graphics can help Amd 7870 and I have a 4250 built this can help you? or not?
                    _________
                    Espaņol
                    Hola a todos soy nuevo en el foro de Phoronix y bueno, en el tema de Amd me interesaria sabe si eto puede ayudar a mi grafica Amd 7870 y tengo una integrada 4250 esto puede ayudarla en algo? o no?

                    Comment


                    • #11
                      That sound fantastic. So with that patch I can use my radeon and my intel card together ( sony vaio VPC-Z21 with a fgx card in the dockingstation)?
                      Because now it is not possible to use the external displays that are connected to the dockingstation and the internal display together.

                      Does anyone has a reference how to use that feature? maybe a xorg.conf example ?

                      Comment


                      • #12
                        Originally posted by halfmanhalfamazing View Post
                        Or is this part of some "foundational groundwork" which will help lead to crossfire/SLI support in the OSS driver? (Ditto for nouveau)
                        No.
                        The purpose of crossfire and SLI is having both GPU working together on the same output is if it was a huge GFX card with more GPUs on it.
                        The whole array of cards renders 1 ouput (this is done by either having 2 cards drawing alterning frames, or spliting the output frame in two and having each card render 1 half, or much more complicated schemes).

                        The purpose of DMA-BUF (the technology behind OSS implementation of Optimus and Reverse-Optimus) is to send the output from one card to another. They are not collaborating on 1 output.
                        The first card does output the frame alone on its own. In addition to its normal output port (HDMI connector 1, DisplayPort connector 2 and 3, etc.), thanks to DMA-BUF it also see a virtual output port. Trying to display things on this port will actually transfert the output frame to another Graphic card which didn't had anything to do with it up to now.

                        Optimus example:
                        - You have a laptop
                        - The laptop has an onboard Intel, this intel is connected to the DVI output port and to the built-in screen.
                        - The laptop has a GeForce card, this card isn't connected to absolutely anything and can't output a damn thing on its own.
                        - You setup the Geforce as your main desktop.
                        - thanks to DMA-BUF you create a "virtual connector" on the GeForce as if it had a DVI output.
                        - this virtual connector actually sends (or maps) the output to the Intel embed.
                        - the Intel embed actually displays the output using it's own output connectors.

                        Reverse example:
                        - You have a workstation, on which you also develop OpenCL code.
                        - You have a small punny Radeon, embed in the APU of your main processor.
                        - You have the latest possible overkill monster as a descreet card.
                        - You want to test OpenCL code and you want it to run on the big card, and you don't want anything else disturbing it (you want 100% of your ressources to produce your bitcoins :-P )
                        - The embed Radeon is more than enough to display the desktop.
                        - BUT you want to still by able to use your 3 monitors setup (2 or 3 of them are connected to the big card).
                        - You setup the embed as your main desktop.
                        - Thanks to DMA-BUF, in addition to the crappy built-in VGA-out of the motherboard, you give access to a few virtual connectors to your embed card.
                        - These connectors send (or share) the output frame with the big GFX card.
                        - The big GFX card displays the output to your 2-3 monitors.
                        - All the drawing is done by the embed.
                        - The big card doesn't do much beside displaying the frame to the monitors. Most of the RAM is free, and the GPU doesn't do a shit.
                        - So you have all the ressource free for your precious OpenCL bitcoins.


                        USB example:
                        - you bought a nice all-in-one Picture Frame / portable TV / external display combo.
                        - when in external display, the widget works over USB
                        (That's simply a framebuffer that can be displayed. No 3D-acceleration-over-USB or whatever).
                        - (There is no actual connector like HDMI, DVI, DP+, VGA, ...) beside an useless composite in.
                        - But you want full 3D.
                        - So you use your main machine (laptop, desktop, whatever) as normal (with your normal big monitor connected to your output).
                        - Thank to DMA-BUF, you creat an additional virtual connector for the USB display.
                        - This additionnal connector writes the output frames into the framebuffer of the USB display.
                        - Bam! You get 3D acceleration on a accelerator-less USB monitor.


                        The same strategies can also work for an in-kernel always working VNC that directly works on top of KMS/DMA-buff (you still have the display even if the X or Wayland crashes and kicks you to the console. VNC works as if it where a normal monitor connected on an extra DVI port).
                        It could also work for gameplay streaming (Play PC games actually running on your workstation while playing them on some under-powered portable console or tablet)

                        Comment


                        • #13
                          If anyone cares, yesterday I was able to get a multiscreen opengl-composited desktop across two radeon X1250 on the same machine, using the latest kubunty nightly (it even set them up automatically for me).

                          It worked fine for 2 minutes and then was hit by massive slowdowns, probably some kind of software fallback, but I tried multiple times and it always happened, even if the desktop just sat there doing nothing.

                          So... works but not perfectly, for now. I'll retry again when 13.10 is nearing release, and maybe file some bug reports, although mysterious X server slowdown will be interesting to decide against what I should file it.

                          (I also tried with a radeon and an nvidia 6200le but couldn't even get both of them to light up at the same time).

                          Comment


                          • #14
                            I've been running this for a month with a Radeon HD 6750 as a primary and a Radeon X300 as an output slave. Biggest annoyance was cursor flickering on the outputs connected to the primary graphics card, which I solved by disabling "framebuffer object" and "always use buffer swapping" options in the OpenGL plugin of Compiz in CCSM. I can even run Unigine Heaven across 3 monitors in this configuration.

                            Comment


                            • #15
                              There are AMD athlon X2 laptops with nvidia graphics

                              Originally posted by halfmanhalfamazing View Post
                              Have there been any AMD-based laptops which have an Nvidia discrete GPU built into them?
                              .
                              Yes, my sister has one, so I have to make sure all kernel updates are done at my place in case
                              the NVIDIA blob doesn't work right after the update. It stays on Ubuntu 10.04 to preserve the
                              GNOME 2 environment, works great. If ever a newer OS were needed it would have to be Mint/MATE.

                              In 10.04, Nouveau was still very new, the "experimental" 3D driver will work but that machine has heat
                              issues unless constantly blown clear of dust, and I don't experiment on machines used by people
                              who are not hackers and need months of reliability between visits.

                              Comment

                              Working...
                              X