Announcement

Collapse
No announcement yet.

NVIDIA Linux 2008 Year in Review

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

  • NVIDIA Linux 2008 Year in Review

    Phoronix: NVIDIA Linux 2008 Year in Review

    Yesterday we had published our AMD Linux 2008 Year in Review where we had provided a recap of all major features introduced in their Linux driver over the past year as well as taking the time to re-benchmark every Catalyst driver released this year. We have been doing this process annually going back to 2005, and now it's time to look at the NVIDIA Linux driver releases from this year.

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    looking at the benchmark, there isn't really much to say if not: +10 for NVIDIA. also considering their stable driver, XV working as it should with composite, coolbits, SLI, VDPUA, WINE really working good and other things I don't even know, we can consider NVIDIA the clear winner of GPU on Linux side.
    Last edited by bulletxt; 16 December 2008, 07:21 PM.

    Comment


    • #3
      The gtkDrawingArea test doesn't show a figure for the final test.

      I'm surprised if nvidia managed to optimise the gtkDrawingArea to be 20 times faster! Especially considering that the ATI drivers do the test in about 17 seconds (note: different hardware).

      Comment


      • #4
        Hum... I hope we will have a GPL driver soon.
        Indeed, if we have a BSD driver (currently nouveau has a BSD license which nullifies the GPL), we will probably end with a sub-optimal open source driver, with only proprietary forks which will be optimal, for instance with hardware secrets shared between apple/nvidia. Really, there, the "GPL only thing" is really important.
        Moreover, I use the proprietary driver with a compositing window manager... the direct rendering is a real mess, namely it's full of bugs (indirect rendering is fine but very slow without vblank sync). The real answer is to let the compositing window manager have a finer grained control of the VRAM. Some will tell me that's the role of the opengl driver... well, currently VRAM management seems to be very badly done by the opengl driver, at least for composited desktops.

        Comment


        • #5
          @bulletxt

          The benchmarks are pretty useless to compare fglrx+nvidia. The used gfx cards are pretty old now - of course they have to exist 1 y ago too, but currently you would have to compare newer ones. You can only see a tendency between the drivers, not much more.

          What is not shown are of course the video abilities, rendering issues, which are not visable in numbers. Of course you can have got a fast driver for many apps, but when the image which is rendered is incorrect for the app you want to use then the card/driver is useless. Also stability is not shown in numbers, but when you see how easy you can crash a system with fglrx, then nvidia is the clear winner. vdpau hardcrashed some times too, but it improves from one driver to the next.

          Comment


          • #6
            I'll quickly summarize what I think in one short sentence. Nvidia, please die. Thanks.

            Comment


            • #7
              I'm wondering if there is still a significant performance delta between the Windows drivers and these Linux drivers for the Geforce 8/9 series nowadays? Is SLI performance in Linux progressing as well, presuming that this wasn't optimal either?

              I'm aware that the results from the NVIDIA Workstation performance article (http://www.phoronix.com/scan.php?pag...ion_perf&num=1) from March were very promising as if these issues seem fixed, but the benchmarks results are from SPECViewPerf and not the well known gaming titles. Are they comparable in that regard? Only Windows Vista is used, perhaps XP performs way better than Vista?

              Perhaps this is worth a future benchmark/article? That would be nice I think..

              Comment


              • #8
                Originally posted by chromis View Post
                I'm wondering if there is still a significant performance delta between the Windows drivers and these Linux drivers for the Geforce 8/9 series nowadays? Is SLI performance in Linux progressing as well, presuming that this wasn't optimal either?

                I'm aware that the results from the NVIDIA Workstation performance article (http://www.phoronix.com/scan.php?pag...ion_perf&num=1) from March were very promising as if these issues seem fixed, but the benchmarks results are from SPECViewPerf and not the well known gaming titles. Are they comparable in that regard? Only Windows Vista is used, perhaps XP performs way better than Vista?

                Perhaps this is worth a future benchmark/article? That would be nice I think..
                The performance difference on the 8/9 compared to windows is next to nil. Some stuff is slightly faster in windows others are slightly faster in linux.

                As far as the SLi stuff goes. There is development going on with that. I've been working with one of the NV devs since August getting some stuff corrected (speedups) on SLi. Sad reality is though that most of the games that are on linux simply don't push the hardware hard enough to see noticable increase of speeds, single cards barely break a sweat.

                Comment


                • #9
                  Originally posted by jamei View Post
                  The gtkDrawingArea test doesn't show a figure for the final test.
                  There's a very short bar, too small for numbers to fit in.

                  I'm surprised if nvidia managed to optimise the gtkDrawingArea to be 20 times faster! Especially considering that the ATI drivers do the test in about 17 seconds (note: different hardware).
                  Something seems screwy. My GeForce 7200SE using the 177.82 driver and X Server 1.4.2 runs gtkDrawingArea Pixbufs thru 100 iterations in only 0.23 seconds.

                  (My CPU is a lot less beefy, too, a 64X2 4000+, running a 64-bit kernel and 32-bit userland.)

                  Comment


                  • #10
                    Is it possible then that Nvidia cards cheat, and don't draw every frame? That kind of speed difference could be the result of that.

                    Remember the Geforce FX times, where not the full screen was antialiased to get some more fps..

                    Comment

                    Working...
                    X