Announcement

Collapse
No announcement yet.

AMD APU vs. Radeon GPU Open-Source Comparison

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

  • AMD APU vs. Radeon GPU Open-Source Comparison

    Phoronix: AMD APU vs. Radeon GPU Open-Source Comparison

    Earlier this month I ran some benchmarks showing that with the very latest open-source AMD Linux graphics driver code, the AMD APU Gallium3D performance can be ~80%+ the speed of Catalyst, the notorious Linux binary graphics driver. For end-users curious what the AMD A10-6800K "Richland" APU performance is comparable to when it comes to discrete Radeon graphics cards with the R600 Gallium3D driver, here's some weekend comparison benchmarks.

    http://www.phoronix.com/vr.php?view=19438

  • #2
    RAM speed?

    What was the speed of RAM used in the test? It is well known that APUs are bottlenecked by memory bandwidth, and especially AMD's apus scale well with increased RAM bandwidth.

    Comment


    • #3
      Originally posted by TemplarGR View Post
      What was the speed of RAM used in the test? It is well known that APUs are bottlenecked by memory bandwidth, and especially AMD's apus scale well with increased RAM bandwidth.
      I was going to ask the same question. There is a huge difference between 1333 and even 1866.

      Comment


      • #4
        Originally posted by benmoran View Post
        I was going to ask the same question. There is a huge difference between 1333 and even 1866.
        The performance scales with DDR3 speed, seeing that all AMD mobos claim to be able to hit at least DDR2 2133 and that DDR3 2133 stock speed low voltage and timing ram is within $5 of 1866 for the same capacity it only makes sense to use that as a baseline.

        Ram OC tests with the AMD APUs show that the GPU is heavily limited by the ram and will continue to scale pretty much linearly with that speed if you don't let the timings get out of hand and kill the performance gains.

        For the APUs upping your latency timings to 13-13-13-31 just to hit a higher MHZ is going to be no better then 11-11-11-27 at a lower Mhz. I forget the formula right now to calculate it.

        If you intend to game with them go for at least DDR3 2.4Ghz and disable the power saving modes.

        Comment


        • #5
          Originally posted by templargr View Post
          what was the speed of ram used in the test? It is well known that apus are bottlenecked by memory bandwidth, and especially amd's apus scale well with increased ram bandwidth.
          ddr3-2133.
          Michael Larabel
          http://www.michaellarabel.com/

          Comment


          • #6
            Offtopic

            Unigine Valley 1.0 and Heaven 4.0 works pretty nice on RadeonSI (and R600g) now.

            Comment


            • #7
              Originally posted by Kivada View Post
              If you intend to game with them go for at least DDR3 2.4Ghz and disable the power saving modes.
              Disabling power saving modes will not increase your performance, it will only waste power.

              Comment


              • #8
                Thank you Michael for linux OSS APU performance.

                Comment


                • #9
                  Originally posted by Calinou View Post
                  Disabling power saving modes will not increase your performance, it will only waste power.
                  The tests don't show that, disabling the power saving features raises your minimum framerate. While this does nothing for your max framerate, the max framerate doesn't matter, minimum and frame latency do. So what if you can hit 500FPS nose up to a blank wall when your framerate consistently falls below 30FPS during heavy action?

                  I take it you don't understand this concept because minimum framerate is yet another thing Larabel doesn't do that all of the decent Windows hardware reviewers have been doing forever.

                  Comment


                  • #10
                    Originally posted by Kivada View Post
                    The tests don't show that, disabling the power saving features raises your minimum framerate. While this does nothing for your max framerate, the max framerate doesn't matter, minimum and frame latency do. So what if you can hit 500FPS nose up to a blank wall when your framerate consistently falls below 30FPS during heavy action?

                    I take it you don't understand this concept because minimum framerate is yet another thing Larabel doesn't do that all of the decent Windows hardware reviewers have been doing forever.
                    Does that even apply now that the CPU governor was fixed in kernel 3.12?

                    Comment


                    • #11
                      Originally posted by mmstick View Post
                      Does that even apply now that the CPU governor was fixed in kernel 3.12?
                      It does, since there is still a delay between when the game is requesting more performance from the CPU/GPU and the CPU/GPU coming back out of it's lower power state, that latency is what causes the extra hit to the minimum framerate.

                      If you are marginal in a game and don't want to OC, just cap the speeds at their stock max and disable the power saving features and it will squeeze a few extra FPS out at the low end at the cost of a bit more heat and power consumption without having to OC or upgrade the cooling.

                      Comment


                      • #12
                        Originally posted by mmstick View Post
                        Does that even apply now that the CPU governor was fixed in kernel 3.12?
                        Michael, could you please run tests with the Linux kernel 3.12 & 3.13 with ondemand & performance governors to check if there are still differences?
                        Afaik this was never tested after the fix landed.
                        I would like to see some gaming benchmarks with min/average/max fps graphs included

                        Comment


                        • #13
                          Thanks for the OSS APU benchmarks.

                          I do have a few suggestions though,
                          1. Compare Intel and AMD APUs. I'd really like to see them benchmarked side by side with open drivers.
                          2. OpenCL and AMD's HSA architecture will become more and more important in the near future so we need to see OpenCL testing baked into new benchmarks. I bring this up in this thread due to OpenCL being very GPU related. Oh I do realize that open source wise support is lagging so testing with closed source drivers should be a goal,too.
                          3. While testing against discreet cards is nice, it would be nice to have some APU context here. That is result from a previous generation APU should be in the graphs.
                          4. non gaming OpenGL performance should be addressed to a greater extent! Games are nice but not everybody these days is considering APUs for gaming.

                          The idea here is that integrated GPUs are becoming good enough for many of us to consider these days. For years I dismissed integrated GPUs out of hand but we are certainly at the tipping point right now. Effectively they are processors that are now good enough. Of course good enough depends upon what you want to do with the machine, in this case "engineering" apps are a part of my interests.

                          Comment


                          • #14
                            Originally posted by Kivada View Post
                            minimum framerate is yet another thing Larabel doesn't do that all of the decent Windows hardware reviewers have been doing forever.
                            Clearly I must be seeing things. Is anyone else also hallucinating the low and peak numbers?



                            (Yes, he doesn't do them when he can't, ie when there is no frame time info. I too would like him to get frame time info on all apps, which would be a day's coding away, but apparently not a big priority.)

                            Comment


                            • #15
                              Originally posted by curaga View Post
                              Clearly I must be seeing things. Is anyone else also hallucinating the low and peak numbers?

                              (Yes, he doesn't do them when he can't, ie when there is no frame time info. I too would like him to get frame time info on all apps, which would be a day's coding away, but apparently not a big priority.)
                              More like this http://hardocp.com/article/2013/12/0..._card_review/5

                              Comment

                              Working...
                              X