Intel Linux Graphics Shine With Fedora 12

[phoronix]

Phoronix: Intel Linux Graphics Shine With Fedora 12

Intel's Linux graphics driver stack is often at the forefront of X.Org / Mesa innovations, from Intel being the first driver having in-kernel video memory management to being the first driver with mainline kernel mode-setting support to even being the driver that often first receives support for new OpenGL extensions in Mesa. The Intel Linux driver stack can be attributed with many firsts, but continually pushing this driver while putting out quarterly timed releases has led to some pains. Earlier this year in fact the driver stack was rather buggy -- especially in Ubuntu 9.04 -- that impaired many users with stability issues, performance problems, and other headaches. Most of the regressions from overhauling the Linux driver stack have been resolved, but where is the driver stack at now? The Intel stack in Ubuntu 9.10 is performing rather well, but where it's more important is its status within Fedora as more of the bleeding-edge graphics packages are pulled into this release that often don't make it into other distributions until months later when they roll out their next releases. To see where the Intel Linux graphics are at in Fedora 12, we ran the same set of benchmarks in the Fedora 10, 11, and 12 releases with an Intel G43 IGP.

http://www.phoronix.com/vr.php?view=14388

[Kano]

It is really unclear how h264 should be faster with a newer intel drivers. You can only use xv for it (no vaapi support yet) and that's more or less the same for every driver.

If you encode mpeg2 full hd or fetch a mpeg2 (interlaced) broadcast from somewhere then you should be able to see a difference with a new mplayer-vaapi. Sadly deinterlace is not available with it yet - but OSD works at least. I have got no script for fedora, but with debian or ubuntu with libdrm >= 2.4 this one should work.

http://kanotix.com/files/fix/mplayer-vaapi-latest.txt

pal/ntsc res is no real test for this as mpeg2 is not really demanding for current cpus.

[bridgman]

I suspect the difference comes from improvements in the memory management code.

[m4rgin4l]

I have a notebook with a GM4500HD chipset and KMS is totally broken. The system freezes just starting X or two or three seconds later. Supplying nomodeset fixes the problem, but then I can't access the virtual consoles. Compiz is broken too.

Nothing new here, we know the first days of a Fedora release are a pain in the ass.

[Dragoran]

I have a notebook with a GM4500HD chipset and KMS is totally broken. The system freezes just starting X or two or three seconds later. Supplying nomodeset fixes the problem, but then I can't access the virtual consoles. Compiz is broken too.

Nothing new here, we know the first days of a Fedora release are a pain in the ass.

I have a laptop with a GM45 (4500MHD) that works just fine with F12 without any manuall tuning.

KMS, OpenGL, Compiz (no tearing!),OpenGL ontop of compiz, Video playback, Flash all are working fine.

I yet have to find something that is "broken".

[m4rgin4l]

I have a laptop with a GM45 (4500MHD) that works just fine with F12 without any manuall tuning.

KMS, OpenGL, Compiz (no tearing!),OpenGL ontop of compiz, Video playback, Flash all are working fine.

I yet have to find something that is "broken".

Must be something specific of the configuration of my laptop (That's good news).

Thanks

[DoDoENT]

The Intel performance may shine on F12, but ATi performance is terribly bad (https://bugzilla.redhat.com/show_bug.cgi?id=539525).

[cruiseoveride]

I'd like to see a n00b comparison of fc12 and ubuntu9.10 . Its been a while since we had that sort of comparison. Not a benchmark per se (but that wont hurt), but more of general use and usability.

[whizse]

I have a notebook with a GM4500HD chipset and KMS is totally broken. The system freezes just starting X or two or three seconds later. Supplying nomodeset fixes the problem, but then I can't access the virtual consoles. Compiz is broken too.

Nothing new here, we know the first days of a Fedora release are a pain in the ass.

You have filed a bug report about this, right? If not, now would be a good time, as upstream have removed support for non kernel modesetting.

[Melcar]

The Intel performance may shine on F12, but ATi performance is terribly bad (https://bugzilla.redhat.com/show_bug.cgi?id=539525).


That's odd. My 200M works beautifully. Must be with specific card families.

[uaaquarius]

I have a notebook with a GM4500HD chipset and KMS is totally broken. The system freezes just starting X or two or three seconds later. Supplying nomodeset fixes the problem, but then I can't access the virtual consoles. Compiz is broken too.

There is a very similar problem with GMA950 (https://bugzilla.redhat.com/show_bug.cgi?id=531755). Your problem might be related somehow.

[AdamW]

kano, I maintain a Fedora repository with libva and mplayer-accelerated (vaapi) packages: http://www.happyassassin.net/video-experimental/

m4rgin4l, as others have suggested, that's not typical behaviour: please file a bug, including all the info requested at https://fedoraproject.org/wiki/How_to_debug_Xorg_problems .

dodoent: as the closing of your bug as a dupe implies, poor performance with KMS enabled on some ATI chipsets is a known issue. I should really call it out explicitly in common_bugs, actually. It is something that's known and will be worked on, we appreciate it's a bit of a pain in the mean time.

[DoDoENT]

dodoent: as the closing of your bug as a dupe implies, poor performance with KMS enabled on some ATI chipsets is a known issue. I should really call it out explicitly in common_bugs, actually. It is something that's known and will be worked on, we appreciate it's a bit of a pain in the mean time.

I've seen now that my bug is reported as duplicated and that poor performance is a known issue. Thank you very much for pointing that out.

I just hope that you'll quickly find a way to increase performance, at least to enable normal animations performance in OpenOffice Impress.

[Kano]

@AdamW

Did you enable the 965 driver in the libva package?

[AdamW]

we'll try. For now, the workaround of disabling KMS should work for most cases. If your graphics work okay with KMS disabled, it's alright to run that way.

[AdamW]

@AdamW

Did you enable the 965 driver in the libva package?

In the package in my personal repository, yeah. I have a review request in for Fedora proper; if that's ever accepted, it'll have to be without the 965 driver, as it implements patented stuff.

[DoDoENT]

we'll try. For now, the workaround of disabling KMS should work for most cases. If your graphics work okay with KMS disabled, it's alright to run that way.

The problem is that disabling KMS breaks suspend on my graphics card on 2.6.31. It does suspend, but it doesn't resume. I even tried --quirk-vbestate-restore and that works once, but after initiating second suspend, the screen is corrupted and hard reboot is required.

So for now, I'm using KMS for my everyday work and non-KMS when I need to perform an OOo Impress presentation.

[Dragoran]

To first recap some important pieces of information for Fedora 10, Fedora 11, and Fedora 12 are the package versions found in each release. Fedora 10 shipped with the Linux 2.6.27 kernel, GNOME 2.24.1, X Server 1.5.3, xf86-video-intel 2.5.0, and Mesa 7.3-devel. Fedora 11 provided the Linux 2.6.29 kernel, GNOME 2.26.1, X Server 1.6.2 RC1, xf86-video-intel 2.7.0, and Mesa 7.5-devel. The brand new Fedora 12 release is using the Linux 2.6.31 kernel, GNOME 2.28.1, X Server 1.7.1, xf86-video-intel 2.9.1, and Mesa 7.7-devel. All three Fedora releases were left in their stock configurations after installation. The x86_64 version of all three Fedora releases was used.

When doing such benchmarks you should issue an update first.

F11 ships 2.6.30 now.

[AdamW]

dodoent: ah, I see. That does suck, then :/. Sorry about that. All I can say is to watch the bug report, Dave and Jerome will get to it.

[Kano]

@AdamW

How to use your repo with rawhide?

[AdamW]

@AdamW

How to use your repo with rawhide?

just add it like any other repo, with a file in /etc/yum.repos.d/ . follow the format of the other files in there, change the name and use the appropriate path for my repo. something like:

[video-experimental]
name=Experimental video for Fedora Rawhide
baseurl=http://www.happyassassin.net/video-experimental/rawhide/x86_64
enabled=1
metadata_expire=7d
gpgcheck=0

ought to do it. remember to change the arch if appropriate for your system.

[Craig73]

I second the call to see the performance of Ubuntu 9.10 included here... is there a benefit (video wise) to running a distribution using newer packages?

[FunkyRider]

Now we are waiting on the new stack to run on real 3D accelerator. Not this intel 3D joke. If Radeons can run as good as this, it will be one of it's own kind. sigh...

[squirrl]

Now we are waiting on the new stack to run on real 3D accelerator. Not this intel 3D joke. If Radeons can run as good as this, it will be one of it's own kind. sigh...

Intel is real 3D acceleration. The tests should have included Windows XP as a control group. This would add a measure to compare to.

Consider adding UT2004 as a litmus test.

[whizse]

I guess you could argue about what "real" means here, Intel really doesn't target gamers.

Anyway, comparisons with Win would be welcome.

[FunkyRider]

Oh please, not a single 'real' Intel's 3D accelerator can beat the bottom of the line GeForce 8500 or even an 785G integrated chipset, and if you've ever seen PC games in recent like 10 years, there is no place for intel's graphics. We are living in the full HD gaming era where people treat 1920x1080, 4AA resolution as "normal". functional wise, intel's graphics can't even generate an acceptable render frame in modern games. I don't know if there is any better word than 'joke' to describe this. Intel never took gaming seriously. Or maybe they tried, it's either just too easy or too hard.

“They’re the world’s leading designers and manufacturers of CPUs – how hard could it be to build a GPU? I mean, come on, how hard could it be? That crummy little company down the road builds them – we could build them in our sleep. Come on, how hard could it be?” ——NVIDIA David Kirk

[Craig73]

Oh please, not a single 'real' Intel's 3D accelerator can beat the bottom of the line GeForce 8500 or even an 785G integrated chipset, ...I don't know if there is any better word than 'joke' to describe this. Intel never took gaming seriously. Or maybe they tried, it's either just too easy or too hard.


I don't think that anyone was disputing that Intel accelerators have not served the needs of gamers. I would guess that 80% of the broader market has not cared about games nor 3D to any significant degree. So while Intel chips have not served the needs of games, that does not mean Intel is out of touch of the broader market.

It's also not hard to argue that 3D is serving a useful need in the general desktop, and there are innovative UI designs that can be accelerated by a GPU, and that the GPU's role in general purpose computing is becoming very significant (offloading video decoding from CPI, compositing desktops, accelerating photo editing, etc.). And Intel can be seen bringing stronger 3D offerings to their product lineup.

Either way, Intel is also very engaged in the Open Source community with open source drivers, the Moblin platform, etc., so finding other ways to describe this other than "joke", in order to soften your tone, would offer more respect to a significant contributor to the community.

[BlackStar]

Either way, Intel is also very engaged in the Open Source community with open source drivers, the Moblin platform, etc., so finding other ways to describe this other than "joke", in order to soften your tone, would offer more respect to a significant contributor to the community.

One word: Poulsbo.

Not to mention shitty, non-conformant OpenGL drivers across the board. Check out the OpenGL forums if you wish to how developers feel about the nightmare of supporting Intel hardware.

You may not like it, but Intel is single-handedly holding back the adoption of OpenGL. It's one of the leading reasons why Direct3D is the only real option for consumer graphics.

So you think *that* is good for the community? Heh, good one - tell us more!

[caramerdo]

Oh please, not a single 'real' Intel's 3D accelerator can beat the bottom of the line GeForce 8500 or even an 785G integrated chipset, and if you've ever seen PC games in recent like 10 years, there is no place for intel's graphics. We are living in the full HD gaming era where people treat 1920x1080, 4AA resolution as "normal". functional wise, intel's graphics can't even generate an acceptable render frame in modern games. I don't know if there is any better word than 'joke' to describe this. Intel never took gaming seriously. Or maybe they tried, it's either just too easy or too hard.

“They’re the world’s leading designers and manufacturers of CPUs – how hard could it be to build a GPU? I mean, come on, how hard could it be? That crummy little company down the road builds them – we could build them in our sleep. Come on, how hard could it be?” ——NVIDIA David Kirk

Please think before you post.

Intel graphics cards are not directed to "gamers" - they don't give a shit about those people. Integrated graphics are extremely useful for business people and mom/dad types which only want decent 2D and Google Earth. And guess what, they are the most lucrative market since they represent 80% of the graphics card sales.

Myself, I just bought a new laptop with the Intel X4500 because of 3 things:
a) I don't play games
b) Integrated graphics gives 25%+ more battery life
c) Intel has excellent working open source drivers, and having been burned by Ati and their crappy fglrx I really appreciate that

Sorry if I'm aggressive, but I'm tired of seeing people complaining about integrated graphics performance in games - come on, do you really expect them to play games? Just don't buy them for that.


BlackStar, can you provide a link for that claim? According to Intel, the latest drivers are OpenGL 2.1 compliant...

[BlackStar]

BlackStar, can you provide a link for that claim? According to Intel, the latest drivers are OpenGL 2.1 compliant...

According to the 2.1 specs (pdf) (http://www.opengl.org/registry/doc/glspec21.20061201.pdf), "OpenGL 2.1 implementations must support at least revision 1.20 of the OpenGL shading language." (page 351, J.1). The latest drivers I could find would only support GLSL 1.10 - and badly at that.

Personally, I spent the better part of last week rewriting an application to work on a X4500 chip: downgraded to GLSL 1.10, disabled FBO blits, floating point attachments, MRT (which translates to no HDR, bloom, shadows or antialiasing) reduced texture resolution and finally... the driver produced an utterly broken picture. Imagine memory corruption, similar to TV snow, overlayed on top of the actual rendering.

That was on Vista, by the way. On Ubuntu, the driver simply refused to render any geometry touched by a vertex shader. In the end I simply gave up: Ati and Nvidia cards consume the application just fine, but Intel ones simply do not offer any meaningful OpenGL support. Maybe if you limited yourself to 15-year old GL1.1-level features the drivers might work correctly - but that's simply not an option.

It's sad to see the largest IHV (50%+ marketshare) produce such garbage. Their competitors are shipping OpenGL 3.2, when Intel is still struggling with 2.1 - and yes this *does* drive developers away from OpenGL and into Direct3D (and Microsoft).

Edit: Search for Intel on opengl.org (http://www.google.gr/search?client=opera&rls=el&q=site:http://www.opengl.org/discussion_boards/+intel&sourceid=opera&ie=utf-8&oe=utf-8) to see how depressing the situation really is.

[m4rgin4l]

You have filed a bug report about this, right? If not, now would be a good time, as upstream have removed support for non kernel modesetting.

I tried using abrt (a kernel oops is generated) but it didn't work, I'll try using bugzilla but that thing is a pain in the ass.

Edit: Filed the bug: https://bugzilla.redhat.com/show_bug.cgi?id=541593

[AdamW]

m4rgin4l: just checking - you saw I closed your bug as a dupe of https://bugzilla.redhat.com/show_bug.cgi?id=540218 and a kernel build with a candidate fix is now available (well, will be there in a few minutes), right? http://koji.fedoraproject.org/koji/buildinfo?buildID=143335

[m4rgin4l]

I can log in now, but Compiz crashes. It's not "shiny" but it is enough for me. Thanks!

[AdamW]

there's probably a bug report for Compiz crashing already. Probably https://bugzilla.redhat.com/show_bug.cgi?id=539789 ?