Mailing List Archive

MythTV experience with Intel Sandy Bridge graphics
There has been frequent discussion about the state of Intel graphics
and their Sandy Bridge chip capabilities, so I wanted to share our
recent experiences. Here's the hardware and software being used for
the MythTV front-end system.

CPU: Intel i5 2405S (2.5 GHz)
MB: MSI Z68MA-ED55 (HDMI, DVI-D, VGA)
Memory: 4 GB
Drive: Intel SSD (caught a good! Black Friday sale)
OS: Gentoo Linux x86_64 3.1.6
Video drivers
3D driver: mesa 7.11.2 (plus one patch)
2D driver: xf86-video-intel 2.17.0-r3
libdrm: libdrm-2.4.29
X: xserver-1.10.4-r1
Compiler: gcc-4.5.3-r2
Linker: binutils-2.21.1-r1
C library: glibc-2.12.2

MythTV Version : v0.24.1-27-g30993d6
MythTV Branch : fixes/0.24
Network Protocol : 63
Library API : 0.24.20110505-1
QT Version : 4.7.2

Display: Samsung C8000 series plasma

Video and audio are over HDMI. We're using the on-board Z68 graphics
(no add-in video card) and the classic mesa driver.

Myth experience: We started with the CPU++ standard playback profile
(ffmpeg & Xvideo). There was a noticeable horizontal tear about half
way down the screen, most apparent during vertical panning of a
scene. This wasn't surprising given prior reports of similar problems.

We tried switching to ffmpeg & opengl, but Myth would crash on starting
any playback. A bit of debugging identified a null-pointer dereference
in the mesa video driver, which we reported to freedesktop.org. They
fairly quickly posted a proposed (simple) patch which we have now
incorporated. With the patch, playback works *very* well with ffmpeg
& opengl. While I'm not a video engineer (nor do I play one on TV),
there are no apparent video defects and playback is completely smooth.
CPU load is up to ~25-30% on one core playing a 1080i HD recording from
the backend. Audio quality is good as well.

We have not experimented with libva yet.

Power profile: At idle on the X desktop, the system pulls 31.2 watts, as
measured by a Kill-a-Watt meter. With all 4 cores spinning at 100%, the
load is 66.1 watts. Playing a video via mythfrontend draws 34.4.

Keith
_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://www.mythtv.org/mailman/listinfo/mythtv-users
Re: MythTV experience with Intel Sandy Bridge graphics [ In reply to ]
Keith Pyle <kpyle@austin.rr.com> wrote:

> There has been frequent discussion about the state of Intel graphics
> and their Sandy Bridge chip capabilities, so I wanted to share our
> recent experiences.
>
> We started with the CPU++ standard playback profile
> (ffmpeg & Xvideo).  There was a noticeable horizontal tear about half
> way down the screen, most apparent during vertical panning of a
> scene.  This wasn't surprising given prior reports of similar problems.

I have a lower class of CPU than you (Core2 with G35 chipset) and get
silky smooth playback 1920x1080p50 with de-interlacing, the fix for
the tearing is quite simple, once you know ...

add

Option "XvPreferOverlay" "True"

to the Display section

of your xorg.conf, if you have "configless" xorg, create a config file
first using

Xorg :1 -configure
_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://www.mythtv.org/mailman/listinfo/mythtv-users
Re: MythTV experience with Intel Sandy Bridge graphics [ In reply to ]
On Sun, Jan 15, 2012 at 04:49:12PM -0600, Keith Pyle wrote:
> There has been frequent discussion about the state of Intel graphics
> and their Sandy Bridge chip capabilities, so I wanted to share our
> recent experiences. Here's the hardware and software being used for
> the MythTV front-end system.

I am working on a similar system. Here are my specs:

CPU: Intel i3-2100 (dual core 3.1 GHz)
MB: Asrock H67M-ITX
Memory: 4 GB
Drive: Kingston SSD
OS: CentOS 6.1, kernel-ml 2.6.39-4.1 (elrepo-kernel)
Video drivers
3D driver: mesa 7.11-12.1 (elrepo-testing)
2D driver: xf86-video-intel 2.17.0-1 (elrepo-testing)
libdrm: libdrm-2.4.27-1 (elrepo-testing)

MythTV Version : v0.24.1-106-g31e9528
MythTV Branch : fixes/0.24
Network Protocol : 63
Library API : 0.24.20110505-1
QT Version : 4.7.2

Display: Panasonic tc-p65v10 (plasma)

> Myth experience: We started with the CPU++ standard playback
> profile (ffmpeg & Xvideo). There was a noticeable horizontal tear
> about half way down the screen, most apparent during vertical
> panning of a scene. This wasn't surprising given prior reports of
> similar problems.

I have very slight horizontal tearing across the top 1/10th of my
screen. It's not constant; exists in some scenes, and not in
others. I can *almost* ignore it.

I tried enabling the XvPreferOverlay setting in xorg.conf that was
suggested here; that had no effect. Same goes for the
i915.powersave=0 module param that I read about somewhere else.

> We tried switching to ffmpeg & opengl, but Myth would crash on
> starting any playback. A bit of debugging identified a
> null-pointer dereference in the mesa video driver, which we
> reported to freedesktop.org. They fairly quickly posted a
> proposed (simple) patch which we have now incorporated. With the
> patch, playback works *very* well with ffmpeg & opengl. While I'm

Do you happen to have a link to that patch? Or if it's small enough
can you post it? I too have the "crash when using opengl" issue.

> Power profile: At idle on the X desktop, the system pulls 31.2
> watts, as measured by a Kill-a-Watt meter. With all 4 cores
> spinning at 100%, the load is 66.1 watts. Playing a video via
> mythfrontend draws 34.4.

This is why I'm really interested in getting a Sandy Bridge-based
FE going. I underclocked my memory from 1333 to 1066 and
undervolted it to 1.325 Volts. With a PicoPSU my idle system pulls
about 18 Watts (also measured with a Kill-a-Watt). Watching videos
puts me in the 25--30 Watt range (depending on the video).


_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://www.mythtv.org/mailman/listinfo/mythtv-users
Re: MythTV experience with Intel Sandy Bridge graphics [ In reply to ]
On Mon, Jan 16, 2012 at 2:35 PM, Matt Garman <matthew.garman@gmail.com> wrote:
> On Sun, Jan 15, 2012 at 04:49:12PM -0600, Keith Pyle wrote:
>> There has been frequent discussion about the state of Intel graphics
>> and their Sandy Bridge chip capabilities, so I wanted to share our
>> recent experiences.  Here's the hardware and software being used for
>> the MythTV front-end system.
>
> I am working on a similar system.  Here are my specs:
>
>    CPU: Intel i3-2100 (dual core 3.1 GHz)
>    MB: Asrock H67M-ITX
>    Memory: 4 GB
>    Drive: Kingston SSD
>    OS: CentOS 6.1, kernel-ml 2.6.39-4.1 (elrepo-kernel)
>    Video drivers
>        3D driver: mesa 7.11-12.1 (elrepo-testing)
>        2D driver: xf86-video-intel 2.17.0-1 (elrepo-testing)
>        libdrm:    libdrm-2.4.27-1 (elrepo-testing)
>
>    MythTV Version   : v0.24.1-106-g31e9528
>    MythTV Branch    : fixes/0.24
>    Network Protocol : 63
>    Library API      : 0.24.20110505-1
>    QT Version       : 4.7.2
>
>    Display: Panasonic tc-p65v10 (plasma)
>
>> Myth experience: We started with the CPU++ standard playback
>> profile (ffmpeg & Xvideo).  There was a noticeable horizontal tear
>> about half way down the screen, most apparent during vertical
>> panning of a scene.  This wasn't surprising given prior reports of
>> similar problems.
>
> I have very slight horizontal tearing across the top 1/10th of my
> screen.  It's not constant; exists in some scenes, and not in
> others.  I can *almost* ignore it.
>
> I tried enabling the XvPreferOverlay setting in xorg.conf that was
> suggested here; that had no effect.  Same goes for the
> i915.powersave=0 module param that I read about somewhere else.
>
>> We tried switching to ffmpeg & opengl, but Myth would crash on
>> starting any playback.  A bit of debugging identified a
>> null-pointer dereference in the mesa video driver, which we
>> reported to freedesktop.org.  They fairly quickly posted a
>> proposed (simple) patch which we have now incorporated.  With the
>> patch, playback works *very* well with ffmpeg & opengl.  While I'm
>
> Do you happen to have a link to that patch?  Or if it's small enough
> can you post it?  I too have the "crash when using opengl" issue.
>
>> Power profile: At idle on the X desktop, the system pulls 31.2
>> watts, as measured by a Kill-a-Watt meter.  With all 4 cores
>> spinning at 100%, the load is 66.1 watts.  Playing a video via
>> mythfrontend draws 34.4.
>
> This is why I'm really interested in getting a Sandy Bridge-based
> FE going.  I underclocked my memory from 1333 to 1066

As a casual observer, this may have something to do with under performance?

Have any of you used the combination of master and vaapi?

> and
> undervolted it to 1.325 Volts.  With a PicoPSU my idle system pulls
> about 18 Watts (also measured with a Kill-a-Watt).  Watching videos
> puts me in the 25--30 Watt range (depending on the video).
>
>
> _______________________________________________
> mythtv-users mailing list
> mythtv-users@mythtv.org
> http://www.mythtv.org/mailman/listinfo/mythtv-users
_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://www.mythtv.org/mailman/listinfo/mythtv-users
Re: MythTV experience with Intel Sandy Bridge graphics [ In reply to ]
On Sun, Jan 15, 2012 at 7:38 PM, Nick Rout <nick.rout@gmail.com> wrote:
> On Mon, Jan 16, 2012 at 2:35 PM, Matt Garman <matthew.garman@gmail.com> wrote:
>> This is why I'm really interested in getting a Sandy Bridge-based
>> FE going.  I underclocked my memory from 1333 to 1066
>
> As a casual observer, this may have something to do with under performance?

Are you suggesting because of my memory underclock? Or the CPU/GPU in
general? As for the memory clock, there's no perceivable difference
in playback with the memory clocked at 1333 or 1066. I.e., tearing
still exists with the memory at the higher speed and voltage.

As for the CPU/GPU performance: this hardware combination works just
fine for all kinds of playback (including high bit rate Bluray) on
Windows. So I really don't think it's a hardware limitation.

> Have any of you used the combination of master and vaapi?

By "master" are you referring to the svn 0.25 dev branch of mythtv?
AFAIK that's the only way to get vaapi support in myth.

Is it possible to run the stable 0.24 fixes on the backend and
0.25-svn on the frontend? If so, I might be willing to give it a try.
But otherwise, I don't want to risk the stability of an otherwise
perfectly working system (too much WAF risk :)).

Anyway, I'll add: in the initial post of this thread, a
freedesktop.org mesa patch was mentioned that fixed the crashing issue
when using mythtv with opengl rendering. I too have that crash (and
can't find the patch that was mentioned). Interestingly though, if I
play videos using mplayer -vo gl (or gl2), the video plays fine
(without crashing).

-Matt
_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://www.mythtv.org/mailman/listinfo/mythtv-users
Re: MythTV experience with Intel Sandy Bridge graphics [ In reply to ]
On Tue, Jan 17, 2012 at 6:35 AM, Matt Garman <matthew.garman@gmail.com> wrote:
> On Sun, Jan 15, 2012 at 7:38 PM, Nick Rout <nick.rout@gmail.com> wrote:
>> On Mon, Jan 16, 2012 at 2:35 PM, Matt Garman <matthew.garman@gmail.com> wrote:
>>> This is why I'm really interested in getting a Sandy Bridge-based
>>> FE going.  I underclocked my memory from 1333 to 1066
>>
>> As a casual observer, this may have something to do with under performance?
>
> Are you suggesting because of my memory underclock?  Or the CPU/GPU in
> general?  As for the memory clock, there's no perceivable difference
> in playback with the memory clocked at 1333 or 1066.  I.e., tearing
> still exists with the memory at the higher speed and voltage.

I was suggesting that underclocking might have affected performance,
but if you find the same problems at the higher speed then that rules
that out :)

>
> As for the CPU/GPU performance: this hardware combination works just
> fine for all kinds of playback (including high bit rate Bluray) on
> Windows.  So I really don't think it's a hardware limitation.
>
>> Have any of you used the combination of master and vaapi?
>
> By "master" are you referring to the svn 0.25 dev branch of mythtv?
> AFAIK that's the only way to get vaapi support in myth.


Yes, now that the development is in git, i understand master is the
term for the current development version - it's not 0.25 until
released.
>
> Is it possible to run the stable 0.24 fixes on the backend and
> 0.25-svn on the frontend?  If so, I might be willing to give it a try.
>  But otherwise, I don't want to risk the stability of an otherwise
> perfectly working system (too much WAF risk :))

I don't believe it is possible, so maybe better to wait until 025 is released :)

> Anyway, I'll add: in the initial post of this thread, a
> freedesktop.org mesa patch was mentioned that fixed the crashing issue
> when using mythtv with opengl rendering.  I too have that crash (and
> can't find the patch that was mentioned).  Interestingly though, if I
> play videos using mplayer -vo gl (or gl2), the video plays fine
> (without crashing).
>
> -Matt
> _______________________________________________
> mythtv-users mailing list
> mythtv-users@mythtv.org
> http://www.mythtv.org/mailman/listinfo/mythtv-users
_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://www.mythtv.org/mailman/listinfo/mythtv-users
Re: MythTV experience with Intel Sandy Bridge graphics [ In reply to ]
On 01/15/12 19:35, Matt Garman wrote:
> On Sun, Jan 15, 2012 at 04:49:12PM -0600, Keith Pyle wrote:
>> > There has been frequent discussion about the state of Intel graphics
>> > and their Sandy Bridge chip capabilities, so I wanted to share our
>> > recent experiences. Here's the hardware and software being used for
>> > the MythTV front-end system.
> I am working on a similar system. Here are my specs:
> ...
> I have very slight horizontal tearing across the top 1/10th of my
> screen. It's not constant; exists in some scenes, and not in
> others. I can *almost* ignore it.
>
> I tried enabling the XvPreferOverlay setting in xorg.conf that was
> suggested here; that had no effect. Same goes for the
> i915.powersave=0 module param that I read about somewhere else.
>
>> > We tried switching to ffmpeg & opengl, but Myth would crash on
>> > starting any playback. A bit of debugging identified a
>> > null-pointer dereference in the mesa video driver, which we
>> > reported to freedesktop.org. They fairly quickly posted a
>> > proposed (simple) patch which we have now incorporated. With the
>> > patch, playback works *very* well with ffmpeg & opengl. While I'm
> Do you happen to have a link to that patch? Or if it's small enough
> can you post it? I too have the "crash when using opengl" issue.
>
They haven't committed the fix yet, but we picked up the proposed patch
by following the bug report updates and the mesa-dev mailing list. The
patch can be seen here
http://lists.freedesktop.org/archives/mesa-dev/2012-January/017412.html
(Because of indenting changes, it looks like more than it really is.
The effective change is one line to exit the function if the pointer
"intel" is null.) Until the patch is released for your distribution,
you'll need to manually patch mesa/drivers/dri/intel/intel_screen.c as
shown at the above link.

If you need the mesa source, it can be obtained at
git://anongit.freedesktop.org/git/mesa/mesa

Keith
Re: MythTV experience with Intel Sandy Bridge graphics [ In reply to ]
On Mon, Jan 16, 2012 at 05:39:05PM -0600, Keith Pyle wrote:
> They haven't committed the fix yet, but we picked up the proposed
> patch by following the bug report updates and the mesa-dev mailing
> list. The patch can be seen here
> http://lists.freedesktop.org/archives/mesa-dev/2012-January/017412.html
> (Because of indenting changes, it looks like more than it really
> is. The effective change is one line to exit the function if the
> pointer "intel" is null.) Until the patch is released for your
> distribution, you'll need to manually patch
> mesa/drivers/dri/intel/intel_screen.c as shown at the above link.
>
> If you need the mesa source, it can be obtained at
> git://anongit.freedesktop.org/git/mesa/mesa

Thank you for posting that link. I saw that discussion thread, but
wasn't sure whether or not it was the right one.

Using that patch, I was able to rebuild the mesa package for my
distro (centos6, using elrepo-testing repository). I only tested
briefly, but so far my experience is the same as what you originally
posted: defect-free, smooth playback (as far as I can tell anyway).

How long have you been running with this setup? I've only been
testing SNB in general for a couple days, and only got the mesa
patch going about an hour ago. But I am very encouraged by what
I've seen and your report. I've been waiting for SNB graphics on
Linux to "just work" for quite a while now. If I ignore the need to
patch and rebuild mesa, looks like we might be there.

I'm keeping my fingers crossed that, going forward, everything
continues to work as well as my nvidia/vdpau-based frontend!

Thanks again!
Matt

_______________________________________________
mythtv-users mailing list
mythtv-users@mythtv.org
http://www.mythtv.org/mailman/listinfo/mythtv-users