Announcement

Collapse
No announcement yet.

Early OpenGL vs. Vulkan Linux Benchmarks With Talos Principle

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

  • #31
    Originally posted by gouchi View Post

    Just press Esc to exit.

    Performance monitoring, you can display fps (fps_show = "true") and also add performance counters (perfcnt_enable = "true") in the configuration file.

    For automated test case, there is "demo" core using opengl 4.3.
    If it doesn't dump performance data in a way that can be parsed it isn't really usable.

    Comment


    • #32
      Originally posted by duby229 View Post

      If it doesn't dump performance data in a way that can be parsed it isn't really usable.
      Yep, and having to hit ESC tends to be less than scientific results :/
      Michael Larabel
      https://www.michaellarabel.com/

      Comment


      • #33
        Also, there are a few key performance features that aren't yet enabled in the Intel driver - we were racing to get it done - so I wouldn't put a ton of stock in benchmark results yet. That should all be coming soon enough.
        Free Software Developer .:. Mesa and Xorg
        Opinions expressed in these forum posts are my own.

        Comment


        • #34
          Hey not directly related but maybe an indicator on what to get soon in Vulkan from AMD? The chopper Nvidia demo seems very rough to me...
          First DX12 gaming benchmark shows R9 290X going toe-to-toe with a GTX 980 Ti.

          Comment


          • #35
            Originally posted by Kayden View Post
            Also, there are a few key performance features that aren't yet enabled in the Intel driver - we were racing to get it done - so I wouldn't put a ton of stock in benchmark results yet. That should all be coming soon enough.
            Which are ? Could you give examples ?

            Comment


            • #36
              Originally posted by pal666 View Post
              so, we have slow, unreleased and immature drivers and only one of them is open source atm
              In this case of Talos Principle the performance issue is mostly not due to the drivers. the dev made the below post. They have only just achieved Vulkan API calls using a wrapper and managed to get it out as early optional beta. But this is only the first step for them in porting the engine over to Vulkan. Steps 2 to 3 below to have their game engine properly utilizing Vulkan are not done yet and beta users will have to await more patches over a period of time.

              engine design for Vulkan is basically consited of three major parts:\

              1) Port. Make it work as fast as possible just by wrapping current engine design around Vulkan. Avoid all pitfalls and bottlenecks. This is what we did by now and released as patch for Talos.

              2) Use Vulkan for multi-threaded rendering. Our engine is designed really well for multi-threaded rendering, but we have only our wrapper for it - calls to graphics API (like Vulkan) are not multi-threaded. Yet.
              That being said, this is the next step what we'll do. And probably release that also as patch for Talos. I tried to do that with Direct3D 11 long time ago (support for its deffered contexts), but it was too much pain and too little or even no gain. That's just one of reasons why we decided to stick with our own approach for MT renderer for that long.

              3) Redesign engine for Vulkan. This is the biggest step and can be split in two:

              3a) Precache all rendering states (which mostly mean materials in game) up front. This will make rendering calls much simplier and faster. So, instead of deciding at rendering time what is needed for a material to be rendered via Vulkan, do this at loading time and then when material needs to be rendered just give it to Vulkan, via one or two simple function calls.

              3b) Precache all geometry, material, textures, everything that is needed for rendering an object up front. This basically creates so called command buffer ready for Vulkan, and nothing extra needs to be set or created at render time.

              3rd part of port is, obviously, the most complex one, and it'll take time to change engine design for it, step-by-step.
              https://steamcommunity.com/app/25751...1651559970/#p2

              This mirrors what Chris Roberts said about taking time to properly refactor your engine to get the best out of DX12 and Vulkan
              Last edited by humbug; 20 February 2016, 06:23 AM.

              Comment


              • #37
                Originally posted by Kayden View Post
                Also, there are a few key performance features that aren't yet enabled in the Intel driver - we were racing to get it done - so I wouldn't put a ton of stock in benchmark results yet. That should all be coming soon enough.
                But still Intel is listed as conformant on the khronos site. So the conformance tests are just crap?

                Comment


                • #38
                  Originally posted by V10lator View Post
                  But still Intel is listed as conformant on the khronos site. So the conformance tests are just crap?
                  He spoke about "few key performance features". Be conforms to the specification is one thing while being fast is another.

                  Comment


                  • #39
                    So when the readme https://cgit.freedesktop.org/mesa/me...n.txt?h=vulkan says

                    Unsupported Features: - Tesselation shaders - Push constants in GS and VS on HSW and prior - Sparse resources - Input attachments
                    Is Vulkan still feature complete in the driver?

                    Comment


                    • #40
                      Originally posted by pal666 View Post
                      so, we have slow, unreleased and immature drivers and only one of them is open source atm
                      That said, this is probably the most broadly supported launch for an open graphics standard in history, isn't it ?
                      Test signature

                      Comment

                      Working...
                      X