Announcement

Collapse
No announcement yet.

A 14-Way Comparison Of NVIDIA vs. Nouveau Drivers

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

  • A 14-Way Comparison Of NVIDIA vs. Nouveau Drivers

    Phoronix: A 14-Way Comparison Of NVIDIA vs. Nouveau Drivers

    Back in September I provided the most comprehensive AMD Radeon Linux graphics comparison that took 28 graphics cards from all supported ATI/AMD Radeon product families and tested them under Linux using the latest Catalyst driver as well as the open-source Mesa/Gallium3D driver. In this article is a similar comparison on the NVIDIA side as I take most of the GeForce graphics cards at my disposal and try them under the NVIDIA binary Linux driver and the community-developed open-source "Nouveau" driver. Not only is the OpenGL performance looked at for multiple generations of NVIDIA hardware, but the thermal and power consumption is compared too. In certain OpenGL workloads, the open-source Linux driver is now faster than NVIDIA's own driver for select graphics cards in a fair comparison, but overall the NVIDIA blob still reigns supreme.

    http://www.phoronix.com/vr.php?view=16676

  • #2
    Shame there isn't any current hardware in there (I'd have liked to have seen my GTX580) for nothing else but for a comparison against the 460, which seems to trounce everything else.

    Having had my 580 for a while, I'm not sure I'm getting my money's worth in Linux.

    Comment


    • #3
      Isn't benchmarking three games based on ioquake a bit pointless? The result will all be similar.

      And to me there's a clear link between CPU usage and driver performance.

      Wish I could help in Nouveau development, but I just don't know were to start hacking

      Comment


      • #4
        Well, I for one find Nouveau to be amazing work. Granted I'm not much of a gamer, but Nouveau works great with my 9600GT and I can't say I've experienced any performance loss in Blender sculpturing/modeling since I switched from the binary blob to Nouveau. Obviously depending on your demands and hardware configuration, Nouveau will not be a solution for everyone and for them there's the binary blob. From a purely technical standpoint it's incredibly impressive that the Nouveau guys have been able to reach this level of performance through reverse-engineering something as advanced as these top-of-the-line discrete gpu's. When the project first started I thought it would never amount to much and now it's providing me with accelerated graphics on my Linux installations, massive respect!

        Comment


        • #5
          Great to see that you have listened to the criticism on the forum Michael, providing a summary and some thoughts at the end! Might I ask for a header there though? Something like

          Summary and comments:

          would be nice, as well as an index with links. The latter is less important than the former.

          Comment


          • #6
            I am quite surprised that the GeForce 6600 GT works so bad at Michael. My GeForce 6600 GT (AGP) from XFX works very well, with a Nouveau kernel module from Git.

            Comment


            • #7
              Originally posted by XorEaxEax View Post
              ... Nouveau works great with my 9600GT ...
              Maybe for 2D, but burning the same amount of power as official blobs while providing a 10th the framerate, isn't exactly "great" in my book.
              I'll always buy nvidia because of their Linux support, but for the same reason I'll always install their driver.

              Comment


              • #8
                Pages 11 and 12 show only bars for one of the drivers. Error?

                Comment


                • #9
                  On the "Power" and "CPU" utilization graphs, there is a comment at the top that "Less is better". I do not believe that this is what you intended, as it does not correlate with the performance data. If a part draws 30% more power, but accomplishes the task in >30% less time or performs >30% additional operations per second, then the "less is better" statement is untrue. The obvious exceptions being the idle-utilization graphs and fixed-duration tests (playing back a DvD for example).

                  Comment


                  • #10
                    Originally posted by bug77 View Post
                    Maybe for 2D, but burning the same amount of power as official blobs while providing a 10th the framerate, isn't exactly "great" in my book.
                    Well, like I said I do use it with Blender for 3d scultping/modeling and I haven't noticed (as in felt) any performance decreases. That said, I haven't actually measured the performance (given that it's been running fine I haven't seen any reason to). I was expecting Blender to become noticeably slower when switching to Nouveau and was happy when I didn't notice any decrease in performance.

                    Comment


                    • #11
                      Power consumption

                      GTX 460 @ nouveau: 2 watts
                      GTX 460 @ blob: 13 watts

                      Your explanation: "this may be due to the Nouveau driver not initializing all areas of the GPU"

                      Really ? *Maybe* it is because all areas of the GPU are clocked at 1/10 of the frequency the blob drives them at ?
                      Don't you think this is kind of *important* information when doing a *performance* comparison ?

                      Comment


                      • #12
                        Originally posted by oliw View Post
                        Shame there isn't any current hardware in there (I'd have liked to have seen my GTX580) for nothing else but for a comparison against the 460, which seems to trounce everything else.

                        Having had my 580 for a while, I'm not sure I'm getting my money's worth in Linux.
                        First off there are 2 500 series GPUs in the review, second, if you bought a current top of the line flagship class GPU for use with the OSS drivers you are an idiot since even the officially supported AMD OSS drivers have allot of issues wit the HD6900 flagship series GPUs as unlike Nvidia and their previous series of GPUs the HD6900 series actually did change the GPU core allot more then just adding more shader processors.

                        Comment


                        • #13
                          But if there weren't 500 series cards in the article, people would complain that the newest cards were missing.
                          This way it's actually proved that they work bad with Nouveau.

                          Comment


                          • #14
                            The overall power consumption of the system with the different graphics cards was always lower when using the Nouveau driver, but this may be due to the Nouveau driver not initializing all areas of the GPU and taking full advantage of the hardware[...]
                            It could also be attributed to the CPU not being used as much, if that meant it were able to operate at a lower power state.

                            Obviously, the drivers will need to take more advantage of the GPU in order to close the gap further. I was going to say it would need to take more advantage of the CPU too, but then I realized the reason that the CPU might not be working as hard could be that it's waiting for the GPU.

                            Anyway, I'm quite happy and grateful for what the nouveau developers have been able to accomplish so far, and I can't wait to see what they do next. ^_^

                            Comment


                            • #15
                              The difference between 8800GT and 9800GT GPUs is striking because it's basically the same GPU.

                              Michael, I smell something very fishy with your results, e.g. on this page the fps difference is fourfold which is kinda wrong.

                              Comment

                              Working...
                              X