Intel SNA vs. Modesetting GLAMOR - DDX Benchmarks

[phoronix]

Phoronix: Intel SNA vs. Modesetting GLAMOR - DDX Benchmarks

Following all of the Intel 3D graphics tests this week for DRM-Next code for Linux 4.7, Mesa 11.3-devel, and more, there's been a few readers requesting a fresh xf86-video-intel vs. xf86-video-modesetting comparison...

http://www.phoronix.com/scan.php?page=news_item&px=Intel-DDX-May-Tests

[FLHerne]

I was using xf86-video-modesetting on this Sandy Bridge laptop for a while, there wasn't any noticeable difference in use or power consumption.

[bug77]

I was using xf86-video-modesetting on this Sandy Bridge laptop for a while, there wasn't any noticeable difference in use or power consumption.

Most Sandy Bridge mobile parts are 35W or less. Considering the IGP took much less space on the die back in SB days, it's not unexpected IGP power consumption differences aren't noticeable.

[davidbepo]

xf86-video-intel 4ever

[CrystalGamma]

So no differences outside of the synthetic benchmarks? Why keep using xf86-video-intel if we have code that can be used for many platforms, and thus probably is much less buggy than the alternative? (I've used both Intel and AMD GPUs in the last years, and neither is particularly stable …)

[SaucyJack]

Oh man I can't wait to install the modesetting driver and get that sweet gtkperf performance boost.

[Ericg]

Oh man I can't wait to install the modesetting driver and get that sweet gtkperf performance boost.

Michael didn't inverse the graphs. SNA won.

[Ericg]

So no differences outside of the synthetic benchmarks? Why keep using xf86-video-intel if we have code that can be used for many platforms, and thus probably is much less buggy than the alternative? (I've used both Intel and AMD GPUs in the last years, and neither is particularly stable …)

Because the SNA architecture has been finely tuned and is known to be basically stable? Modesetting is still under development. I'll be happy when everyone's switched over too, but the developers have decided we're not there quite yet.

[franglais125]

I'll be happy when everyone's switched over too, but the developers have decided we're not there quite yet.

My experience has been that modesetting provides a smoother experience. This does not necessarily mean better performance though. I might be getting lower FPS for this or that, but I see no flickering/tearing, or glitches (SandyBridge mobile here).

Of course this is just my anecdotal experience, and this might be very different for other people! Do you know what the recommendation from, say, Fedora is? I find it a bit difficult to find reliable information on what is currently recommended by distributions. There is a lot of misinformation on various forums/reddit. Chris WIlson seems to think it's better to stay with intel [1].

FWIW, here are the notes from Debian testing (stating the obvious: not stable) on xserver-xorg-video-intel: The use of this driver is discouraged if your hw is new enough (ca. 2007 and newer). You can try uninstalling this driver and let the server use it's builtin modesetting driver instead. [2]

[1] https://bugs.freedesktop.org/show_bug.cgi?id=94843
[2] https://packages.debian.org/testing/...rg-video-intel

[Sloth]

As far as [1] goes, I was the one reporting the comparative numbers in there. Almost immediately after that bug was filed there was an absolute flurry of xf86-video-intel commits, and the performance got quite a bit better. It now beats modesetting in my tests, which was not previously the case. More importantly, it was constantly GPU hanging on me, whereas modesetting was rock solid, and the changes made xf86-video-intel stable. Finally.