Years Later, Intel Poulsbo Remains A Bloody Mess

[droidhacker]

I bought the Fit-PC2 when it was brand spanking new and to be honest I was very disappointed (and a bit pissed) when I realized there was a black box GPU inside. I don't blame Intel for this, it's clearly PVR's fault.

How exactly do you claim it to be PVR's fault? Intel had every option to give them the finger and use their own GPU instead of this substandard pile of shit.

[patrik]

How exactly do you claim it to be PVR's fault? Intel had every option to give them the finger and use their own GPU instead of this substandard pile of shit.

Intel had nothing that could compete with PVR's performance / power ratio at the time. Using their own technology would have put their product outside the TDP requirements for their intended use. If I get to speculate I believe they tried to bundle their in-house GPU technology with the Zxx Atoms but just couldn't get the TDP low enough and had to toss in some external IP at the end.

The Intel part of the Poulsbo GPU is pretty much a 945gm and the docs are available at intellinuxgraphics.org. They also released source code but as you might know, it never got accepted into the Kernel. At least not until Alan Cox, who is employed by Intel (though I'm unsure whether he is paid by Intel for this) cleaned it up and took out the PVR blob glue.

Sure, they could have handled it much better by supplying their part of the driver straight up at launch. This is what they have been trying to do with Cedarview. Specific documentation on the GPUs that gma500_gfx supports is something we'll probably never see because of the gray area where Intel IP ends and PVR IP begins, but those parts can be figured out from the source code.

In retrospective, what they could have done was to never release Poulsbo at all. But then we wouldn't have anything to argue about.

[droidhacker]

Intel had nothing that could compete with PVR's performance / power ratio at the time. Using their own technology would have put their product outside the TDP requirements for their intended use. If I get to speculate I believe they tried to bundle their in-house GPU technology with the Zxx Atoms but just couldn't get the TDP low enough and had to toss in some external IP at the end.

The Intel part of the Poulsbo GPU is pretty much a 945gm and the docs are available at intellinuxgraphics.org. They also released source code but as you might know, it never got accepted into the Kernel. At least not until Alan Cox, who is employed by Intel (though I'm unsure whether he is paid by Intel for this) cleaned it up and took out the PVR blob glue.

Sure, they could have handled it much better by supplying their part of the driver straight up at launch. This is what they have been trying to do with Cedarview. Specific documentation on the GPUs that gma500_gfx supports is something we'll probably never see because of the gray area where Intel IP ends and PVR IP begins, but those parts can be figured out from the source code.

In retrospective, what they could have done was to never release Poulsbo at all. But then we wouldn't have anything to argue about.

Nonsense, its a weak-ass chip, they could get the same power envelope and BETTER PERFORMANCE by downclocking a 945gse.
It was a shitty decision to go with that PVR crap.

[hwertz]

"As a side note, that bug is a damn fine example of what is wrong with Ubuntu."
Sort of. I've seen problems with other linux distros, with macosx, and especially with windows, where the solution is non-sensical like "turn this option off, then right back on" or whatever.

"How exactly do you claim it to be PVR's fault?"
Although it was Intel's choice to use this chip, it's PVR that produces absolutely no info on the chip. So I blame PVR for this.

"It isn't just a PVR stuck on with chewing gum, its a hybrid of PVR+Intel IP."
Well, actually, it *is* a PowerVR SGX535 stuck onto the chipset. To be honest I think it's just stuck onto an on-chip bus.

"Intel had nothing that could compete with PVR's performance / power ratio at the time."
This. And I'm not sure if they STILL have a competitive chip in this regards. The Z530 has a 2 watt TDP, the US15W (which is where the PowerVR sits) is 2.3 watts TDP. The 945 used like 6-8 watts all by it's lonesome.

That said.. I haven't upgraded to Ubuntu 12.04 yet, I'm running 11.04. EMGD is a bitch and a half to install, but it does work OK... I'm using two PPAs:
http://ppa.launchpad.net/gma500/emgd-1.8/ubuntu natty main
and
http://ppa.launchpad.net/jools/emgd-xorg1.9/ubuntu natty main
This runs the current EMGD. What can I say? This is a slow chip, the OpenGL is pretty slow, the 2D only supports solid fill, bitblt (X calls it "copy") and render extension. (I doubt this is a driver fault, I'm guessing there's just not an extensive 2D accelerator.) *BUT*.... the video playback works great! Vaapi lets me play an HD video at 20% CPU usage, that would otherwise be between 80% and 110% (i.e., 110% would mean an action scene could 'fall behind' in mplayer then catch up afterwards.) I found playing back 60FPs stuff requires framedrop (it seems to manage about 55FPs.) I wonder if it tries to frame-sync, my panel runs at 60hz? Anyway, though, besides just playing full-screen, this gives me plenty of power to play a video in a window while doing something else.

[patrik]

"Intel had nothing that could compete with PVR's performance / power ratio at the time."
This. And I'm not sure if they STILL have a competitive chip in this regards. The Z530 has a 2 watt TDP, the US15W (which is where the PowerVR sits) is 2.3 watts TDP. The 945 used like 6-8 watts all by it's lonesome.

Yes the US15W uses about 40% (2.3 / 6) of the power that 945 does and transistor power consumption isn't linear to the clock speed either. Down clocking the 945 to match that would make simply make it useless.

[locovaca]

Sort of. I've seen problems with other linux distros, with macosx, and especially with windows, where the solution is non-sensical like "turn this option off, then right back on" or whatever.

No, the WTF is the progression of the bug. Let me summarize it for you:

Act I - A cold day somewhere in the Northern Hemisphere
User 1: I have a problem
Cron Job: Please run this command.
User 1 runs command

Act II - Two and a half months later
User 2: Uhm, is anyone going to look at this?
Developer 1: Oh, look, this is an old issue that nobody has touched. Look, reproduce on the beta version and let us know if it's an issue. Oh, and run this command. And btw, this isn't confirmed.
User 1: I already ran the command, and other people have indeed confirmed this is a bug.
Developer 1 passively-aggressively changes bug back to unconfirmed
User 1: I've done what you've asked, WTF do you need, I'm willing to give it to you.
Developer 2 changes bug back to confirmed
Cron Job: Have we mentioned that you could run beta kernels? Why don't you try that. By the way, this bug is unconfirmed.
Cron Job changes bug back to unconfirmed

Seriously annoying shit that happens on most Ubuntu bugs you'll find. I gave up submitting bugs for this reason solely. Having cron jobs f up my bug reports and giving a blanket "test this on the bleeding edge version that isn't released yet even though we have no friggin clue what would've fixed it" is more annoying and demoralizing than just not submitting a bug report and doing exactly that- blindly hoping the next version magically fixes something.

[grege]

Xubuntu on the other hand - functions. I have an old Dell Mini 12 and have, until recent times, successfully used the hacked EMGD driver as provided in the gma500 PPA. I tried a clean install of Xubuntu 12.04 and it "just worked" for the purposes for which this netbook is used.

2D applications work fine. Chrome web pages scroll smoothly and LibreOffice documents have no issues.

You can turn on composting effects and they will work, but they are apparently using LLVMPipe. With a single core Atom and 512mb of ram I thought this is not a great idea, so I set composting to off.

No 3D, but with old netbooks running Xfce4 that is no loss.

Windowed YouTube videos play smoothly

[grege]

Xubuntu on the other hand - functions. I have an old Dell Mini 12 and have, until recent times, successfully used the hacked EMGD driver as provided in the gma500 PPA. I tried a clean install of Xubuntu 12.04 and it "just worked" for the purposes for which this netbook is used.

2D applications work fine. Chrome web pages scroll smoothly and LibreOffice documents have no issues.

You can turn on composting effects and they will work, but they are apparently using LLVMPipe. With a single core Atom and 512mb of ram I thought this is not a great idea, so I set composting to off.

No 3D, but with old netbooks running Xfce4 that is no loss.

Windowed YouTube videos play smoothly

BTW - there is no video playback that works.

[DanL]

I really don't give a shit who came up with the design.

I know, you just want something to rage at because you couldn't possibly be responsible for your terrible hardware purchase decision. If you want something to take out your frustration on, try a mirror...

[droidhacker]

I know, you just want something to rage at because you couldn't possibly be responsible for your terrible hardware purchase decision. If you want something to take out your frustration on, try a mirror...

Oh go pull your head out of your ass.
The purchase decision was based on the device's overall capabilities, and the fact that it said INTEL, which EVERYONE at that time assumed (wrongly, as it turned out) meant that the thing would be properly supported by open source drivers. INTEL FUCKED EVERYONE OVER with ***THEIR*** decision to ship PVR crap.

Do you not notice how computers you buy in the store have labeles on the boxes and stickers below the keyboards identifying WHO MADE various critical components? A sticker that says AMD, a sticker that says NVIDIA, a sticker that says INTEL. NO STICKER SAID POWERVR!!!!

Intel boxed up this SHIT as their own and fucked everone over.

AT LEAST IF IT SAID POWERVR, I WOULD HAVE ANTICIPATED DRIVER PROBLEMS AND ASSESSED ACCORDINGLY.

The reason why this is NOT my fault, is because Intel had NEVER done this previously. Previously ALL graphics parts were supported by their own open source driver.