[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
----------------------------------------------------------------------------
      i915 platform|                            |KBL, BXT
            Summary|[SKL GT4e] large perf drop  |[GEN9(+)] large perf drop
                   |(up to 27%) in most 3D      |(up to 1/3) in most 3D
                   |benchmarks from             |benchmarks from
                   |force-enabling IOMMU        |force-enabling IOMMU

--- Comment #6 from Eero Tamminen <eero.t.tamminen at intel.com> ---
Largest performance drops from IOMMU with recent user-space 3D / Media stack
are following (all machines have dual channel memory)...

SKL GT4e (i7-6770HQ)
--------------------

* 25-30% SynMark CSDof (fullscreen)
* 20-25% GpuTest windowed 1/2 screen Triangle, SynMark VSTangent
* 20-25% SynMark Fill* [1]
*    20% Unigine Heaven
*    20% GpuTest fullscreen Triangle, SynMark DeferredAA [2]
* 10-20% GLB 2.7 windowed 1/2 screen Egypt & T-Rex
* 10-20% SynMark ZBuffer & Deferred [2]
* 10-15% SynMark TexMem512, GPU write [1]
*  5-10% GfxBench T-Rex, Manhattan 3.0 & 3.1, CarChase, SynMark TerrainFly*
*     5% Unigine Valley, GfxBench AztecRuins
*   2-3%  8-bit, max FullHD, FFmpeg/MediaSDK GPU transcode/downscale

[1] With June user-space. With latest Mesa, Fill* & write tests drop is only
3%, and TexMem512 perf somehow improves by 2%.  Latest Mesa is several percent
faster than older one in these tests, due to Mesa slice/subslice balance
optimization, no idea how that can reduce impact of IOMMU.

[2] With June user-space. With latest user-space, drop in these specific tests
is half of that, or less.  For fullscreen Triangle case, potentially relevant
user-space change could be latest X server disabling atomic commits.  See:
https://gitlab.freedesktop.org/xorg/xserver/issues/888


KBL GT3e (i7-7567U)
-------------------

* 30-35% SynMark Fill* [1]
* 20-25% GpuTest windowed 1/2 screen Triangle, SynMark VSTangent
* 10-15% GLB 2.7 windowed 1/2 screen Egypt & T-Rex, SynMark ZBuffer
* 10-15% GpuTest fullscreen Triangle, GPU write [1]
* 10-15% 4K HEVC GPU decode + hwdownload
*  5-10% Unigine Heaven, GfxBench T-Rex, Manhattan 3.0 & 3.1 & CarChase,
SynMark CSDof
*  5-10% SynMark Deferred* [1]
*  5-10% 10-bit, 4K HEVC GPU transcode
*   2-3% 8-bit, max FullHD, FFmpeg/MediaSDK GPU transcode/downscale

[1] With June user-space. With latest user-space, drop in these specific tests
is about half.


SKL GT2 (i5-6600K)
------------------

*   >25% GPU write, SynMark VSTangent
*    20% SynMark Fill*, GpuTest windowed Triangle
* 15-20% GLB Egypt
*    15% SynMark ZBuffer
*    10% GLB T-Rex, GpuTest fullscreen Triangle
*  5-10% GfxBench Manhattan 3.0 & 3.1, T-Rex, CarChase
*     5% GfxBench AztecRuins, Unigine Heaven
*   2-6% 8-bit, max FullHD, FFmpeg/MediaSDK GPU transcode/downscale

With the June user-space, there are some differences in how much performance
drops, but nothing major like with GT3e & GT4e (where slice/subslice issue
balance had noticeable impact).


BXT J4205
---------

Results similar to other devices (not reported here as this has higher
variances than them).  Similarly to SKL GT2, user-space version doesn't have
significant impact on how much IOMMU regresses performance.


BDW GT2
-------

As expected, no impact (kernel skips IOMMU for BDW).


Summary
-------

* IOMMU can lose up to third of performance in worst synthetic case, and 5-15%
in real GPU (3d/Media) use-cases.

* Seems that badly balanced sub/slice utilization could have noticeable impact
in IOMMU performance impact for some use-cases.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are on the CC list 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/20190919/29f59a34/attachment-0001.html>