Announcement

Collapse
No announcement yet.

29-Way GPU Comparison On Linux From Kepler & Cypress To Today's Pascal & Vega

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

  • #11
    My R9 290 now rocks in openGL. Totally satisfied.

    I hope the RadV devs give it some love though. It's a bit too far behind the RX 580 in RadV (considering the openGL numbers, or windows numbers).

    Comment


    • #12
      Why is the r9 285 missing in so many tests? The Fury is present and both use the same GCN revision, so support should be roundabout equal.

      Comment


      • #13
        Originally posted by HoermI View Post
        Why is the r9 285 missing in so many tests? The Fury is present and both use the same GCN revision, so support should be roundabout equal.
        Forgot to mention it in the article, but stability/display issues.
        Michael Larabel
        https://www.michaellarabel.com/

        Comment


        • #14
          Why is the GTX 1070 Ti so much faster than the GTX 1080 in many tests?

          Comment


          • #15
            Originally posted by GruenSein View Post

            Without the intention to start an argument, it still isn't even "practically" CPU-bound. That would imply that the CPU is limiting the performance of high end cards because it cannot supply them with enough tasks or the time it takes to render a frame is negligible in comparison with other tasks like game logic, physics etc. However, this is not the case. More powerfull GPUs significantly cut down the frame time.

            I do agree that at way beyond 200fps nobody cares about the actual number, though.
            Well, if we want to be pedantic, unless CPU hits 100% usage, there's no CPU limitation, but "software limitation", Source engine defaults "fps_max 300", but yes, on 60Hz displays it's usless to play at 200 FPS, on 100Hz+ it doesn't really matter, I prefer using fps_max 60 over v-sync because lag.

            Comment


            • #16
              Originally posted by linuxjacques View Post
              Why is the GTX 1070 Ti so much faster than the GTX 1080 in many tests?
              This ...

              Comment


              • #17
                Originally posted by tildearrow View Post
                Why is the 1080 Ti usually always a lot faster than the next fastest card?
                The 1080 Ti is a *lot* bigger than the 1080 - 1.7x the number of transistors, 1.5x the die size, 2x the memory bandwidth - but you can see how scaling limits start to kick in because it is often only ~15-20% faster than 1080.

                Originally posted by linuxjacques View Post
                Why is the GTX 1070 Ti so much faster than the GTX 1080 in many tests?
                For the "a bit faster than 1080" cases I think that's just a function of the SKU being introduced more recently, so the boards/VBIOSes were configured with faster clocks. When performance tracks with clocks more than with number of shader cores that usually means the chip is geometry-bound on that workload, ie limited by the fixed-function logic rather than by the shader core.

                For the "much faster than 1080" cases no idea. My impression was that 1070, 1070 Ti and 1080 all used the same die so first thought would be some kind of quirk in the driver behaviour.
                Last edited by bridgman; 02 June 2018, 07:22 AM.
                Test signature

                Comment


                • #18
                  Originally posted by GruenSein View Post
                  Without the intention to start an argument, it still isn't even "practically" CPU-bound.
                  (snip)
                  I do agree that at way beyond 200fps nobody cares about the actual number, though.
                  Yes that is true. Let us call it "refresh rate bound" then

                  Comment


                  • #19
                    For me this article is the most useful in 2018 so far.

                    Now I see that Vega 64 is not even twice faster than my old R9 290, actual speedup being around 70-80%. Not worth it to upgrade, especially considering the crazy prices because of the miners.

                    Well, even my old HD6950 that is still gathering dust somewhere would give a decent fps in most games in 1080p. Good news, very good news. I will swap the 6950 in if the 290 dies, no sense to waste any money on a new card.

                    Comment


                    • #20
                      Originally posted by Veto View Post
                      Yes that is true. Let us call it "refresh rate bound" then
                      That's true of pretty much every result in his tests - they were almost all over 60fps.

                      Portal was very clearly GPU bound, but obviously what Michael was trying to get at was the fact that it's so lightweight that even the slowest GPUs are more than powerful enough to run it. That's very different from being CPU bound.

                      Comment

                      Working...
                      X