[Intel-gfx] Benchmarking Intel, Phoronix's PTS etc. [was: xf86-video-intel 2.8.0]

[Timo Jyrinki]

I also dislike some of the PTS benchmarks, and sometimes question the
results, but it is just one software project, not the whole site.

2009/8/1 Ben Gamari <bgamari.foss at gmail.com>:
> In sum, it is not Intel's responsibility to teach Mr. Larabel about the
> limitations of microbenchmarks (even if it might do their brand some
> good).

Well, I wouldn't want to have gfx development going all politics and
marketing, but simply put Phoronix cannot be ignored. It is AFAIK
_the_ site everyone watches for FOSS gfx news and benchmarks, and it
is a very good site. For a general public media it is, it's very much
up-to-date with various technical details and ongoings in the whole
community. Also Larabel personally is a valued member of X.org
community (eg. all FOSDEM video recording + reporting stuff if I
recall correctly), so I'd say he deserves good feedback instead of
being ignored.

When there is such a meaningful media like Phoronix, it cannot be
ignored simply by (perceived) technical merits. If the Intel driver
guys would go to discuss with the marketing department, I'd bet they
would be strongly ordered to do good dialog with such a media :) I
believe there is already maybe more dialog between AMD and Phoronix
than Intel and Phoronix, because people knowing each other already
beforehand? Article comments are not good (it seems they are not often
read, I have noticed), I believe direct communication is better, or in
case of mailing lists using the general xorg mailing list with
benchmark/Phoronix topic to catch attention.

I wouldn't hope to see such headings like "Intel/AMD trying to
manipulate PTS results" and such like has been on the proprietary side
when certain manufacturers have been doing "optimizations" :), but
patches, extra tests and simple suggestions to PTS would be very much
welcome. The Phoronix will anyway continue doing benchmarks using PTS,
and publishing results on the latest developments in graphics and
other areas.

Of course competitive website with loads of results is welcome as
well, but simply saying "yeah it's faster" does not change the
perception of Intel graphics as long as there are clear slowdowns and
waiting-for-gfx in basic usage (window switching, browser usage,
compiz effects) or lower-than-before fps in specific games. The
slowness of using netbooks (GMA 950 mostly) with Intel graphics is a
well published issue in various places.

Build bots doing performance regression testing (including comparing
to pre-GEM/UXA era drivers for example!) would be valuable, whatever
way it would be done. At least i915 and i965 should be always
included.

-Timo