Search results for query: *

  1. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    exactly as predicted, thanks for confirming my predictions of the X3D-Cache
  2. T

    Benchmark Result Discussion

    8000MT/s ram, sheesh, that's not possible on AMD
  3. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    I really dont think its the matter of GPU. Baseline1 had a frametime of 3.27ms avg, which means out of 1000ms of a second it took just 3.27ms to finish a full vam frame, corresponding to 305fps. The actual fps in baseline1 however was 70fps, here is a chart: The benchmark measures the time...
  4. T

    Benchmark Result Discussion

    It's not just about ST score, for VAM it's about memory latency. The unity engine does a lot of physics which is wasn't designed for. The physics simulation does a lot of memory reading/writing in a non-optimized memory layout. Crazy memory speeds or huge caches lower this latency and that's why...
  5. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    Huge improvement in frame times, the worst is 7.93ms avg in baseline3, which translates into 126fps. You have configured something wrong in your oculus drivers which lock the framerate to 45fps. VAM creates atleast 126fps in all your benchmarks, but only 45fps arrive in your headset for some...
  6. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    I created a scene with 2 girls, naturalis, some dance animation, 6 lights, softbody and dreamstreetbedroom and I get around 100 fps. The dreamstreetbedroom has no impact on fps for me. If you give me a var I can do some profiling, maybe there is some weird interaction that spins the CPU for some...
  7. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    you could try previous versions that dont replace the mono runtime.
  8. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    as far as I understand it's just a unity asset you import? whats the poly count on that bedroom? I mean if it has as many polygons as the "cubes" in starfield, the GPU needs to render extra hard and that tanks the FPS.
  9. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    the benchmark sometimes doesnt start because there is a bug in the scene starting code. The benchmark creator put async code right at the start which NULL pointers out if the scene loads too fast and softlocks the whole scene. I think your 3070ti is already saturated, 65% usage is already a very...
  10. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    it used to impact loading time because all clothing was recursively deactivated on scene load. I fixed it to only reload clothing once. It does not impact runtime, only when the clothing menu is OPEN it drops fps more if you have more clothing installed.
  11. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    no idea, havent seen vam2 code yet
  12. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    the problem is that the jsons are huge and they are loaded bit by bit from a zipstream instead of unzipping them once and then loading them in one go. Unzipping the the .var first and then loading atoms would increase loading speeds 100x on a ssd, but would make every scene require a few gigs of...
  13. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    it depends on the scene, skin meshing can benefit from hyperthreads if there are many characters to skin. For example a scene with 5 characters can run 3 threads each on 15 threads from which 7 are hyperthreads. This is faster than running 5 characters on 8 non-hyperthreads. I think I will try...
  14. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    its not otherwise looks fine, no idea what single-instance means though
  15. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    well I manged to improve performance in a 5 character scene from 80fps to 200fps, but it only helps ryzen cpus with 32 threads and 2 CCDs like the 5950x. The trick was to park all unity threads to CCD1 while putting all character related threads to CCD2 to use more threads without the inter-CCD...
  16. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    Porting mono to the newest runtime is kind of a shitshow and many plugins break, I think I'm just gonna wait for vam2 beta and then ask @meshedvr if I should commit improvements I find to the source code. Unity 2022.3 should give a lot more room for optimization, especialy openxr and the burst...
  17. T

    Benchmark Result Discussion

    in vam and unity in general, RAM latency and single core performance is king. The X3D versions of ryzen CPUs have bigger caches which make RAM latency less meaningful and therefor have better benchmarks in unity.
  18. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    oh, you are video streaming to your oculus via wifi? Yeah that's probably why the benchmark is so bad: every frame has to be video encoded and that's eating up the CPU. Can you connect an oculus via displayport to your computer?
  19. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    Probably not since that unity version doesnt support openxr, but there is also this https://gitlab.com/znixian/OpenOVR This forwards openvr calls to openxr, potentially bypassing steamvr or oculus middleware
  20. T

    Other CPU Performance Patch (Up to 30% faster physics, up to 60% more FPS)

    I have almost the same CPU and GPU (5950x and 4090) as you and I get constant 90fps min/max everywhere in the vr benchmark. Image reprojection is not ASW, it is used if your HMD is not natively supported and the picture gets first rendered to a screen and the screen is rendered again for your...
Back
Top Bottom