GMA 3650 aka. PowerVR sgx545 and Linux

[idata]

Hello,

Correct me if I'm wrong, but it seems there is absolutely no support for Intel GMA 3650 chipsets on Linux ?

There's a lot of similarities with the GMA500 case, a few years ago for which there is still only very poor

drivers under Linux. The GMA500 was also a chipset from PowerVR painted with Intel colors.

Will it be the same fiasco with this GMA series ?

Or maybe I missed something ?

By the way, I tryed to ask the developpers of PowerVR drivers and they told me to ask Intel support.

[idata]

At least they tried, I guess.

[idata]

Hello,

How will this involve when we will have to upgrade to quantal (12.10) as Robert drivers are and will always be only for precise 32bits.

Thanks

[idata]

I have already recommended to 2 friends NOT to buy Atom N2600 machines due to lack of drivers for Linux.

[idata]

No, you're absolutely right. If they have to go to you for technical advice, they couldn't possibly install something like MeeGo on their system.

[idata]

my dn2800mt server is down the ssd disk failure, did only 3 months runing, at last i was using the intel linux video driver for fedora bee.

who knows what gpu and cpu with this driver did over nm10 chip to my new ssd disk drive.

s..t happens ,, forest gump ,,

peter

[idata]

Since there is no way to start a thread on this web site page, I am responding with a general statement: I have all kinds of Linuxes running on separate partitions of my DN2800 desktop board, including a partition devoted to MeeGo. One thing I have found out is that available benchmarks (e. g., the Peacekeeper) are in a narrow range, with or without specific pvr powervr graphics, with or without one or another adaptation-pc kernel. You can tell me that MeeGo has superior 3d graphics, but any other Linux using the adaptation-pc kernel and a custom-compiled 'fbdev-drv.so' for graphics produces a satisfactory lumpy-looking desktop using the 3650, with no noticeable penalty in performance. In fact, data indicates that the swrast dri driver for fbdev_drv.so is using the adaptation-pc provided /dev/dri to render directly.

As for tthe entertainment aspect of the 3650, google-chrome has a most recent beta with a hidden surprise: It can be used as a high-definition video player by pointing it at a suitable ogg or mpeg4 file on your system (e. g., URL: /home/tux/Videos/BugsBunny.ogg). All you need to make this happen is Xine and the xine-plugin; or mplayer and the mplayer-plugin; or totem and the totem-plugin. Have one of these installed, and the ffmpegsumo.so chrome browser library comes into play, with at least 720p full screen, full-speed video, and synchronized audio. And it does this without using an accelerator chip. So long as Google continues to provide this surprise generosity, ANY dual-cpu atom in the 1.6 and better cpu speed can be made to reproduce high-definition videos at 50% or less cpu utilization.

I'll tell you from running tests that it is not worth the trouble to install any Linux that tries to use the emgd junk available from Intel. It is most certainly a good idea to get MeeGo and adapt it, or copy its video system and kernel to another Linux.

[idata]

I'm greatly disappointed as well with this return to outdated backward approach we experienced in the past from Intel. I thought I have finally found a box which is exactly something I'm looking for, with low energy consumption and just enough computation power. I guess you guys want to help ARM win the market by not releasing open specifications and source code of these gpu drivers. Why you guys decided to loose customers this way?

And screw you PowerVR..

[idata]

Has this been addressed by the recent commits to fix cedarview support in 3.3/3.4? I'm looking at getting an http://www.asrock.com/mb/Intel/AD2700-ITX/ ASRock AD2700 ITX.

• https://github.com/torvalds/linux/commit/b6195aab9ca63a4f6911365f36eb091666fcb15a# drivers/gpu/drm/gma500/psb_drv.c gma500: Fix Cedarview support (Correct version) · b6195aa · torvalds/linux · GitHub
• https://github.com/torvalds/linux/commit/9aa65a2b9de1b46fcf3a7a623de231663802671d# drivers/gpu/drm/gma500/psb_drv.c cdv: Add all cedarview pci ids · 9aa65a2 · torvalds/linux · GitHub

[idata]

This is experimental support, limited 2D only. Limited, because video buffer size can be set only to 8 MB due to unsolved problems with video memory allocation. Consequently, it does not seem possible to reach 24bpp color mode (only 16 bpp) on higher screen resolutions. Video acceleration is non existing too.

Sadly, nobody seem to work on this driver any more

[idata]

So in the last few weeks its become very easy to install the proprietary drivers on Ubuntu and Mint, but is anyone having screen tearing issues with this driver? Even just moving around windows and playing video I've got screen tearing. Sometimes a video will get pretty close to not tearing and other times the same video will tear a lot.

Is it best to use the open source driver instead if I'm just wanting tear free video?

[idata]

I initially set up Fedora 17 i386 with the gma500_gfx kernel driver and the xorg modesetting driver. It works, but the framerate for things like Flash video was pretty awful, compared with Flash on the Windows 7 install my HP Mini 210-4120 shipped with (unsurprising, as this arrangement is unaccelerated).

To that end, I've decided to try Fuduntu, then attempt to port Robert Hook's cedarview Ubuntu ppa to a bunch of RPMs. Obviously, this requires starting with Fuduntu 2012.2, then upgrading the kernel specifically to 3.2, then disabling further updates to kernel, kernel-devel and kernel-headers packages before performing any more updates from the Fuduntu repos. I'm hoping that kernel 3.2.18-2 is sufficiently close to Ubuntu 12.04 LTS's 3.2 kernel that the drm drivers port fairly easily.

Hopefully, the performance will be worth the effort...

[idata]

Actually, if you use any number of Slackware derivatives, the stock Linux kernel will give you a psbgfx framebuffer. If you go into the kernel modules, say,

/lib/modules/3*/kernel/drivers/gpu/stub

you will find the poulsbo.ko module, which does nothing except crash the psbgfx framebuffer. So, rename the poulsbo.ko to prevent it from loading, and you have a working psbgfx framebuffer that works with the same kernel command line on both this gma500 fit-pc2 and my Cedarview DN2800 desktop. Remember, leaving poulsbo.ko operational crashes the system; otherwise, it starts up with all kinds of salutary comments about setting up the fbcon framebuffer and the /dev/dri in the dmesg buffer.

Assume you get this far. You have a framebuffer that talks to the mesa driver. Now all you need is an fbdev_drv.so in /usr/lib/xorg/modules/drivers, and you have a working system.

To get kernel modesetting to work, you add 'video=1366x768' to the command line, and you get a minimal, yet effective, 16:9 resolution that works. One way to do this is to install the RIPLinuX Slackware that has its own fbdev_drv.so and simply update the kernel to a more recent Slackware kernel. Or use the original kernel, with poulsbo.ko renamed and the added resolution statement. Or you can omit the statement on the command line and you will get an automatic 1024x768 which works fine also.

How does one say this? The acceleration features of both kernels are scarcely used nowadays, so running fbdev as your graphics driver will not ruin anything. On the gma500, using a 'performance' frequency ladder gets rid of all kinds of complaints caused by bad video streaming problems. And, the full Xine package from recent Slackware runs full screen and full motion and full synchronization on the gma500 1100 mhz. You can even get a xine plugin for firefox from some Ubuntu source that works.

[idata]

Please, email me I have a question about cedar from meego, I registered special for your reply.

mailto:ginzaki@myopera.com ginzaki@myopera.com

[idata]

So still no official response from Intel?

Just ordered return for 83 ITX boards using GMA 3650. You can just imagine how pissed our customer were.

Seems like Linux users are just 2nd grade citizens and not important to Intel.

[idata]

Tested windows with one of the boards but drivers are buggy even on windows. And no 64bit driver...

We switched to some Fusion boards and can't say a bad word.

Intel & Imagination should leave gfx chip designing to someone who really knows what to do. or stick with mobile phones.

[powerarmour]

Some new Linux drivers are up :-

Provides Linux* PowerVR graphics/media drivers (32-bit). These graphics/media drivers are ported from Imagination Technologies* (IMG) Device Driver Kit (DDK) version 1.7 ED862890.

http://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&DwnldID=21938 http://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&DwnldID=21938

[idata]

I downloaded cdv-gfx-drivers-1.0.3_bee

Is that driver compatible with 3.3.4-5.fc17.x86_64 ?

If yes, can anyone help me, how install driver, step by step?

I reading file 'INSTALL' but have many quastions.

For example:

>* Apply this kernel patch to avoid a known video playback issue:

> * kernel-ttm-clear-high.patch

What does mean 'apply'? What I must type in console?

>* Apply the kernel driver patch for your desired target:

> * For "Ant": cedarview-kernel-_ant.patch

> * For "Bee": cedarview-kernel-_bee.patch

What is 'Ant' and 'Bee' ?

[idata]

Tried the latest driver from Intel, the graphic performance become utterly slow when I rotated my screen to portrait (vertical). Does anybody have a solution to this? Highly appreciated!

[idata]

There is a typo near the top of page 12 of the instructions given at this link :

http://download.intel.com/support/motherboards/desktop/sb/enabling_hardware_accelerated_playback_fedora_16_v.pdf http://download.intel.com/support/motherboards/desktop/sb/enabling_hardware_accelerated_playback_fedora_16_v.pdf

Instead of :

patch -p1 < ../cdv-gfx-drivers-1.0.3_bee/cedarview-kernel-v1.0.3_bee.patch

one should type :

patch -p1 < ../cdv-gfx-drivers-1.0.3_bee/src/cedarview-kernel-v1.0.3_bee.patch

Moreover, I found that the instructions in section 4.2 (pages 7-8) of the same document regarding removing and connecting back the LAN cable did not work for me. The Fedora 16 installer stipulated that the network connection be available during installation.

[idata]

Just to report that compilation of mplayer fails, at least if attempted as per the instructions given in the Intel document cited earlier. The good news is that, thanks to the CDV driver, the display shows OK. The overall impression one gets is that getting the GMA 3650 graphics right for Linux is still some way off. I wonder if anyone has managed to get hardware accelaration to work properly by following the Intel instructions in the document cited above.

[powerarmour]

Some potentially good news regarding Linux PowerVR drivers :-

http://www.phoronix.com/scan.php?page=news_item&px=MTIzNTc [Phoronix] Linux Users Might See A PowerVR Holiday Surprise

Not going to hold my breath much, but since Intel has burned everyone on Cedar Trail regarding Windows 8 support, at least we may have a better chance of decent performance on Linux in future.