[Bug 109216] 4-27% performance drop in Vulkan benchmarks

bugzilla-daemon at freedesktop.org bugzilla-daemon at freedesktop.org
Thu Jan 3 17:09:53 UTC 2019


https://bugs.freedesktop.org/show_bug.cgi?id=109216

            Bug ID: 109216
           Summary: 4-27% performance drop in Vulkan benchmarks
           Product: Mesa
           Version: unspecified
          Hardware: Other
                OS: All
            Status: NEW
          Severity: normal
          Priority: medium
         Component: Drivers/Vulkan/intel
          Assignee: intel-3d-bugs at lists.freedesktop.org
          Reporter: eero.t.tamminen at intel.com
        QA Contact: intel-3d-bugs at lists.freedesktop.org
                CC: jason at jlekstrand.net

Created attachment 142959
  --> https://bugs.freedesktop.org/attachment.cgi?id=142959&action=edit
Avoid Vsync / Mailbox mode in fullscreen

Setup:
* Ubuntu 18.04 / Unity desktop
* Mesa git version
* X server & drm-tip kernel built from git within last couple of months
* Modifiers enabled in X server configuration ("dmabuf_capable")


Vulkan performance noticeably dropped on all Vulkan supporting platforms (at
least from HSW to CFL) between following Mesa commits:
2018-12-17 17:52:23 00e2cbc049 v3d: Fix the argument type for vir_BRANCH()
2018-12-18 18:47:54 29e4b949b4 Revert "nir/lower_indirect: Bail early if modes
== 0"

E.g. on KBL 7500U (GT2), performance drops were following:
- 27% Deferred Multisampling (*)
- 9% Compute NBody (*)
- 6% Compute Raytracing (*)
- 4% GfxBench Vulkan Aztec Ruins normal

(*) Sacha Willems' Vulkan demo, run at FullHD fullscreen, with Vsync disabled
i.e. NOT using mailbox mode as that seems to get Vsynched although Vsync is
supposed to be disabled (see attachment). For example:
 ./deferredmultisampling --fullscreen --benchmark --benchwarmup 3
--benchruntime 20

E.g. in Deferred Multisampling case; GPU, CPU and memory power usage drops
clearly with the performance, so I assume it's become more synchronous.

Performance didn't drop in following of the Vulkan demos:
* Triangle (so it's unlikely to be related to modifiers)
* Multithreading
* Indirect Draw

(Don't have data for the rest of the Vulkan demos.)

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20190103/72925eec/attachment.html>


More information about the intel-3d-bugs mailing list