Announcement

Collapse
No announcement yet.

51 GPUs Tested, From The Radeon HD 2900XT To RX 580 & R9 Fury: Testing The 2017 Linux Driver Stack

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

  • 51 GPUs Tested, From The Radeon HD 2900XT To RX 580 & R9 Fury: Testing The 2017 Linux Driver Stack

    Phoronix: 51 GPUs Tested, From The Radeon HD 2900XT To RX 580 & R9 Fury: Testing The 2017 Linux Driver Stack

    It's that time of the year where we see how the open-source AMD Linux graphics driver stack is working on past and present hardware in a large GPU comparison with various OpenGL games and workloads. This year we go from the new Radeon RX 580 all the way back to the Radeon HD 2900XT, looking at how the mature Radeon DRM kernel driver and R600 Gallium3D driver is working for aging ATI/AMD graphics hardware. In total there were 51 graphics cards tested for this comparison of Radeon cards as well as NVIDIA GeForce hardware for reference.

    http://www.phoronix.com/vr.php?view=24724

  • #2
    The progress these benchmarks show are pretty phenomenal. The 8600GTS is only ~10 years old, and yet in some benchmarks modern cards are pushing out over a 100x more FPS, or two full orders of magnitude. Comparing a 10-year old CPU with modern ones is unlikely to show anywhere near one order of magnitude difference.

    I wonder what the next 10 years holds and if we can hope for even a fraction of these kinds of gains.

    Comment


    • #3
      Originally posted by existensil View Post
      The progress these benchmarks show are pretty phenomenal. The 8600GTS is only ~10 years old, and yet in some benchmarks modern cards are pushing out over a 100x more FPS, or two full orders of magnitude. Comparing a 10-year old CPU with modern ones is unlikely to show anywhere near one order of magnitude difference.

      I wonder what the next 10 years holds and if we can hope for even a fraction of these kinds of gains.
      You can see some ~10 year old CPU comparisons from last year's Phoronix birthday benchmarking - http://www.phoronix.com/vr.php?view=21751 http://www.phoronix.com/vr.php?view=21900 or even Socket 478 compared to Raspberry Pi http://www.phoronix.com/vr.php?view=22529
      Michael Larabel
      http://www.michaellarabel.com/

      Comment


      • #4
        Did you plug in a few graphics cards at a time or one by one? How many hours did it take to run the tests? You do testing like I used catch and train Pokémon!

        Comment


        • #5
          Originally posted by Jabberwocky View Post
          Did you plug in a few graphics cards at a time or one by one? How many hours did it take to run the tests? You do testing like I used catch and train Pokémon!
          One at a time... A number of days.
          Michael Larabel
          http://www.michaellarabel.com/

          Comment


          • #6
            Originally posted by Michael View Post

            One at a time... A number of days.
            serious dedication to the cause!

            Comment


            • #7
              Aaaand still no Triangle. Being a massive-way comparison, I am a little frustrated. I wanted to see Triangle since in the past the Fury (and then the RX 480) were beating the NVIDIA cards, and see if this is still the case (which I doubt because NVIDIA has done dark magic in the last benchmark, since their numbers went up by 50%).

              ​​​​​​Or are you waiting for Vega to run Triangle again, due to HBM2?

              Or does NVIDIA pay you to not show results where they are not in crest?

              Or is it because it is not a real world test at all? (yet it tests memory bandwidth, which no other test does, and real world is not just all games and Unigine stuff, by the way).

              I could simply merge results in OpenBenchmarking.org, but then I'd have mixed outdated results (older Mesa) and therefore an unfair comparison.
              And I wish I could finally use the performance showdown feature, but still it tells me "we were not able to analyze enough results". Really? 18k results and that is not enough?

              This is the final time I mention Triangle, by the way.

              Comment


              • #8
                Originally posted by tildearrow View Post
                Aaaand still no Triangle. Being a massive-way comparison, I am a little frustrated. I wanted to see Triangle since in the past the Fury (and then the RX 480) were beating the NVIDIA cards, and see if this is still the case (which I doubt because NVIDIA has done dark magic in the last benchmark, since their numbers went up by 50%).

                ​​​​​​Or are you waiting for Vega to run Triangle again, due to HBM2?

                Or does NVIDIA pay you to not show results where they are not in crest?

                Or is it because it is not a real world test at all? (yet it tests memory bandwidth, which no other test does, and real world is not just all games and Unigine stuff, by the way).

                I could simply merge results in OpenBenchmarking.org, but then I'd have mixed outdated results (older Mesa) and therefore an unfair comparison.
                And I wish I could finally use the performance showdown feature, but still it tells me "we were not able to analyze enough results". Really? 18k results and that is not enough?

                This is the final time I mention Triangle, by the way.
                Triangle isn't in here since these results are just an extension of the previous tests I've been doing over the past week, where you also mentioned triangle, rather than a brand new comparison started from scratch. Given all the tests being done and the spectrum of cards, I was limiting the number of tests used to try to save some time. I'll be sure to add Triangle for my next GPU comparison though for you
                Michael Larabel
                http://www.michaellarabel.com/

                Comment


                • #9
                  Originally posted by Michael View Post

                  Triangle isn't in here since these results are just an extension of the previous tests I've been doing over the past week, where you also mentioned triangle, rather than a brand new comparison started from scratch. Given all the tests being done and the spectrum of cards, I was limiting the number of tests used to try to save some time. I'll be sure to add Triangle for my next GPU comparison though for you
                  It's fine, you do not have to add it for me. I'd be happier to see the "Performance Showdown" feature fixed, though.

                  Comment


                  • #10
                    Originally posted by Michael View Post

                    You can see some ~10 year old CPU comparisons from last year's Phoronix birthday benchmarking - http://www.phoronix.com/vr.php?view=21751 http://www.phoronix.com/vr.php?view=21900 or even Socket 478 compared to Raspberry Pi http://www.phoronix.com/vr.php?view=22529
                    Thanks for the links. It's kind of surprising to see that in a way I was wrong... in certain workloads 10-year-old CPUs are indeed ~10x slower. Also, the ARM comparisons and wattage comparisons in those charts indicate there may have been something in the neighborhood of an order of magnitude improvement in efficiency over the same period. Still not quite the same performance and efficiency gains as GPUs but the little regular improvements really add up over the years.

                    Thanks for all the beautiful benchmarks.
                    Last edited by existensil; 06-02-2017, 12:54 PM. Reason: typo

                    Comment

                    Working...
                    X