[Bug 111731] [GEN9+] large perf drop (up to 1/3) in most 3D benchmarks from force-enabling IOMMU

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=111731

Eero Tamminen <eero.t.tamminen at intel.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[GEN9(+)] large perf drop   |[GEN9+] large perf drop (up
                   |(up to 1/3) in most 3D      |to 1/3) in most 3D
                   |benchmarks from             |benchmarks from
                   |force-enabling IOMMU        |force-enabling IOMMU
      i915 platform|                            |ICL

--- Comment #7 from Eero Tamminen <eero.t.tamminen at intel.com> ---
ICL-U with IOMMU
----------------

Performance didn't improve in any test from enabling IOMMU. Largest performance
drops are following:

*    30% SynMark VSTangent & ZBuffer
*    24% MemBW GPU write (with i965, 6-7% with Iris & higher FPS (*))
*    18% GLB 2.7 Egypt (windowed), SynMark VSDiffuse*
*    17% GpuTest v0.7 Triangle (windowed)
*    15% MemBW GPU blit (with i965, 20% with Iris & lower FPS (*))
*    14% GLB 2.7 T-Rex (windowed), SynMark HdrBloom
*  5-15% HEVC decode & download
* 11-12% GfxBench Manhattan 3.0 & 31, GpuTest Triangle, SynMark DrvRes
*  9-11% GfxBench T-Rex & CarChase, SynMark TerrainFly*
*   7-9% GfxBench AztecRuins, SynMark TexMem128, DeferredAA, ShMapVsm
*   5-7% Unigine Heaven & Valley, GfxBench ALU2, SynMark TexMem512, Deferred,
TerrainPan*, CSCloth
*   2-6% 8-bit <FullHD Media transcode

(*) Iris doesn't support E2E RBC, so in onscreen tests it uses more bandwidth /
frame, and therefore has lower perf.

PS. 30% perf drop from enabling IOMMU == >40% perf increase from disabling
IOMMU.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
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-gfx-bugs/attachments/20190924/bcdc9e8a/attachment.html>