Announcement

Collapse
No announcement yet.

Initial Radeon vs. GeForce Vulkan Ray-Tracing Performance On Linux

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

  • #21
    Originally posted by bridgman View Post
    I suspect that cb88's point was that the tests Michael ran all seemed to fall into one of two buckets, where all of the tests in a bucket had similar performance characteristics.

    My understanding is that the two buckets are (in DXR-speak) triangle mesh geometry vs procedural primitive geometry. At a 10,000 foot level, NVidia seems to be faster with triangle mesh geometry while we seem to be faster with procedural primitive geometry, which makes sense when you consider the implementations.

    https://microsoft.github.io/DirectX-...-mesh-geometry
    That is vaguely what I meant yes, since we are in fact looking at RT microbenchmarks effectively, though you added more knowledge to it... real games aren't going to usually be entirely one or the other.

    Comment


    • #22
      Originally posted by Jumbotron View Post
      AMD DID get crushed. Look at the Geometric Mean results again.
      There is not enough data to make geometric mean relevant. Also we don't know how much those benchmarks are significative, and if their significativeness is equal. If there was 5 more GeeXLab-like benchmarks tested the geometric mean would be very different and Nvidia would looks like a joke, though in reality there would be nothing tested more than what we see.

      Comment


      • #23
        Originally posted by tildearrow View Post
        <sarcasm>Oh wow, AMD outperforms NVIDIA even in ray-tracing in some cases?
        And this is only their first try on ray-tracing. Amazing!</sarcasm>
        FTFY.

        Comment


        • #24
          Man, I did not expect AMD to be winning on any RT application at this point. Good work; I hope they sponsor some of that Blender Vulkán work.

          Comment


          • #25
            Originally posted by birdie View Post

            FTFY.
            Self evidently you didn't actually fix anything at all. What we know today is that we don't know enough but we can expect interesting developments in the near future. Michael's testing and many comments on this thread are enlightening but in no way conclusive vis-à-vis the potential performance of AMD's ray tracing.

            Comment


            • #26
              Since nvidia already opened this can of worms long time ago, I see no alternative for AMD but to compete in that.

              The "proper way" for ray/path-tracing in games would be something like "probability hit boxes", materials dependency and randomization, to get "more realistic graphics". Either way, the reality of the situation is that you can acomplish similar graphic output with 10 times less compute power without r/p-t. It probably have it's limited use for simulations, for games, I just can't see it as "worth doing".

              Comment


              • #27
                What about ProRender? Do the vulkan-rt powered full-spectrum modes work with this driver?

                Comment


                • #28
                  Well it was looking fantastic but Cornell box and Lucy benches make me scratch my head allot. Do those do more RT workloads (rays) thus the RDNA2 basically fail at being able to keep up? Hopefully AMD can figure it out and optimize the drivers further or something in these areas.....

                  Comment


                  • #29
                    Very interesting outcome, although like other commenters have said, it seems to mostly fall into two buckets for now.

                    I'd be interested in more benchmarks: trough wine/proton, and Quake II RTX.

                    Besides this, can this vulkan-based RT API be leveraged by blender? Would that be more efficient than OpenCL?

                    Comment


                    • #30
                      Blender does not support Vulkan, although it is planned. I guess that it could accelerate evee realtime renderer but I doubt that cycles could be ported to Vulkan. Vulkan-RT is supported by AMD's ProRender (which integrates well into Blender) but according the tests I did today, it works only on Windows, not on Linux with these released drivers. However, as it is now covered on driver side, hopefully ProRender for Linux will introduce support in the next release...

                      Comment

                      Working...
                      X