Announcement

Collapse
No announcement yet.

27-Way NVIDIA & AMD Graphics Card Benchmarks On Linux

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

  • #11
    Originally posted by extraymond View Post
    As long as we can have luxmark result, I really don't mind the form to display it <3
    I'm thinking luxmark cos that is one remarkable benchmark people can have to imagine performance like bitcoin or opencl rendering and so on.
    The operations in each application differ and play to the strengths or weaknesses of each architecture. Just because Solution A beats Solution B in one openCL benchmark, it does not mean that you can expect the same results across all openCL application implementation.

    (as far as bitcoin goes, you would be nuts to mine with any of the GPU's now that they have dedicated cheap ASICs that slaughter GPGPU mining.)

    Something like Rodinia and/or SHOC will give you a better overall computing performance.

    https://www.cs.virginia.edu/~skadron...h_Accelerators

    http://keeneland.gatech.edu/software/keeneland/shoc

    Comment


    • #12
      Originally posted by deanjo View Post
      The operations in each application differ and play to the strengths or weaknesses of each architecture. Just because Solution A beats Solution B in one openCL benchmark, it does not mean that you can expect the same results across all openCL application implementation.

      (as far as bitcoin goes, you would be nuts to mine with any of the GPU's now that they have dedicated cheap ASICs that slaughter GPGPU mining.)

      Something like Rodinia and/or SHOC will give you a better overall computing performance.

      https://www.cs.virginia.edu/~skadron...h_Accelerators

      http://keeneland.gatech.edu/software/keeneland/shoc


      Sorry for draging bitcoin in... That's a bad explanation.
      People have been saying gpgpu for a long time. So it would be nice to have it in the benchmark.
      Maybe in a more real world use cases I think.

      Like
      1.darktable with opencl on(and off)
      2.luxmark(I love luxrender!!!)
      3.There will be more softwares introducing opencl I think, not sure if amd's hsa is talking about the same thing.

      Just pointing these out, because seeing game benchmark is nice, but it would be nice to know the strengh of its opencl power.
      Anyone can think of any other applicaions that might be benifit by opencl?

      Comment


      • #13
        Originally posted by extraymond View Post
        Just pointing these out, because seeing game benchmark is nice, but it would be nice to know the strengh of its opencl power.
        Anyone can think of any other applicaions that might be benifit by opencl?
        It would be nice to know the openCL computing power, however the limited scope applications of what you are benching will not give a good overall picture (or what openCL is primarily used for).

        Comment


        • #14
          Originally posted by deanjo View Post
          It would be nice to know the openCL computing power, however the limited scope applications of what you are benching will not give a good overall picture (or what openCL is primarily used for).
          True!
          Hope opencl can go faster than opengl in mesa (which is difficult I think...)
          Opencl is really benefiting small budget pc's.

          Comment


          • #15
            I don't think most of these results are relevant, because all they say is that at ludicrous framerates well beyond even 120hz monitors, the nvidia driver has more headroom.

            I'd like to see Metro Last Light or something at like 1440p or 4k that could push these gpus below the refresh rates, so we get real world impact data.

            I get that the trend is "catalyst is a slow POS" but I'm of the opinion that the combined fps of the radeon + fglrx drivers beats the singular nvidia one, so I'd still buy AMD parts just because they at least have a few people working on the FOSS drivers. Also, they don't brutalize the gpgpu capabilities of their cards.

            These stats at least show if you want to game on Linux, pretty much any mid range gpu is already overkill.

            Comment


            • #16
              Originally posted by remm View Post
              Ok, so the 770 looks like the best buy right now for a higher end card on Linux.
              Actually, the old (?!?!) 780 is a very interesting deal.
              I just bought a (Gigabyte) 780OC'ed for 100$ more than price of a 770OC'ed. (and >200$ less than the price of a 780Ti).

              FWIW, I'm connecting it to a Dell U2711 1440p display.

              - Gilboa
              Last edited by gilboa; 12-03-2013, 02:30 AM.
              DEV: Intel S2600C0, 2xE52658V2, 32GB, 4x2TB + 2x3TB, GTX780, F21/x86_64, Dell U2711.
              SRV: Intel S5520SC, 2xX5680, 36GB, 4x2TB, GTX550, F21/x86_64, Dell U2412..
              BACK: Tyan Tempest i5400XT, 2xE5335, 8GB, 3x1.5TB, 9800GTX, F21/x86-64.
              LAP: ASUS N56VJ, i7-3630QM, 16GB, 1TB, 635M, F21/x86_64.

              Comment


              • #17
                Originally posted by deanjo View Post
                Does it seem to make a difference in performance?
                no - only in a distant future with big AAA titles. I think AMD should also release cheaper 2 GB cards.
                Either way - I just wanted to know on which basis you did the price comparison. So again: are those all 4 GB cards?

                Comment


                • #18
                  OpenCL performance AMD vs Nvidia:

                  Low complexity -> Amd have almost double performance over Nvidia
                  Normal Complexity -> ~Same performance
                  above normal -> Nvidia wins about double or more performance than AMD
                  High Complexity -> Amd stop working or does slower than a 2007 CPU.

                  So if you want to do complex calculations with Opencl Go Nvidia if you want to do fairly simple things amd would give you the best results.

                  Comment


                  • #19
                    Originally posted by Sdar View Post
                    OpenCL performance AMD vs Nvidia:

                    Low complexity -> Amd have almost double performance over Nvidia
                    Normal Complexity -> ~Same performance
                    above normal -> Nvidia wins about double or more performance than AMD
                    High Complexity -> Amd stop working or does slower than a 2007 CPU.

                    So if you want to do complex calculations with Opencl Go Nvidia if you want to do fairly simple things amd would give you the best results.

                    What's the reason behind this?
                    It seems weird...

                    Comment


                    • #20
                      Originally posted by extraymond View Post
                      What's the reason behind this?
                      It seems weird...
                      The most widespread theory is that cause Amd don't have support for function calls or spill everything must be inlined causing an Out of resources problem...

                      Anyway AMD admits that there's a problem and they already said that they're working on it (i think they said this in late 2011 or 2012) in October of this year they said that probably the only GPUs that are going to be fixed are the GCN ones...

                      Comment

                      Working...
                      X