Intel Releases 965/G35 IGP Documents

[phoronix]

Phoronix: Intel Releases 965/G35 IGP Documents

At the Linux.Conf.Au conference today, Intel has announced NDA-free programming documentation covering the 965 Express and G35 Express integrated graphics processors (IGPs). Intel's display driver has long been open-source, but up until now, they have not been releasing the programming documentation for these products to the public. This move comes months after AMD announced their new open-source strategy and began releasing register documentation on their R500 and R600 GPUs.

http://www.phoronix.com/vr.php?view=11765

[yoshi314]

one could say "two down, one to go" when speaking of those major video card companies ;-)

[remm]

I can already assert my next video card will not be an NVidia :)

[ivanovic]

Thank you AMD/ATI for making this possible. Without you opening the specs, this would probably never have happened...

Looks like this year might become another great year for open source. The success of the EeePC will increase Linux usage numbers in all the statistics (since those tend to mainly count "copies sold", which is generally not a way prefering open systems like linux and bsd). And those statistics might lead to more companies considering cross platform development.

[chrisr]

Are we talking equivalent R200? R300? R400? Similarly for NVIDIA?

Which version of OpenGL can it support?

[val-gaav]

Thank you AMD/ATI for making this possible. Without you opening the specs, this would probably never have happened...


true ... but guys at AMD should really speed up this process, as intel right now is taking a lead (it has docs and good opened driver) ...
It would be really bad if radeonhd wasn't completed and Nvidia did something alike intel ... bad for AMD.

[moonboots]

I can already assert my next video card will not be an NVidia :)

Don't be too quick to count NVidia out, it depends wether they adapt to these changes or not. The ball is in NVidia's court ...

[Svartalf]

I can already assert my next video card will not be an NVidia :)

I will be...hesitant...to make that remark just yet.

Intel's still not QUITE credible for the high-end (Though they ARE getting closer to that...), and if you're honest about it, neither is AMD right at the moment. The future's bright for both parties- very bright. And if your plans are 6-12 months out for purchase, you can probably make that claim without worries or concerns. But if you're buying right now, you're buying to reward their show of faith, not buying something that'll be truly usable yet if you're gaming or doing 3D visualization work. The GMA X3500 shows some real promise of being usable on the low-end for that and light duty gaming. On paper the R600 rocks.

The reality, though, is that NVidia's kind of largely the only robust play on the mid-to-high end right now.

I don't exactly like that any more than anyone else here- but that's how it is. I'm keen on seeing what comes of the next 6-12 months on the front now that Intel's stepped up to the plate along with AMD. No good reason for NVidia to not do the same thing, to be honest. If only there were more good things showing on the fglrx driver story or there were some solid 3D answers in the open source solutions for AMD past R300/R400, then I'd be leaning the same way you are.

[val-gaav]

If only there were more good things showing on the fglrx driver story or there were some solid 3D answers in the open source solutions for AMD past R300/R400, then I'd be leaning the same way you are.

IMHO AMD should drop the development of fglrx for some time and focus on doc and code releasing. Last fglrx driver seems to be quite nice already so people shouldn't complain if they know that AMD for the time being is focusing on the docs and open drivers...

In the long run AMD should drop fglrx for good and just develop some blob module for open drivers for things that couldn't be opened like h.264 hardware decoding or other DRM things... It's a waste of resources to develop both open and closed drivers.

[Svartalf]

Are we talking equivalent R200? R300? R400? Similarly for NVIDIA?

Which version of OpenGL can it support?

According to Mesa's site, they're advertising 2.1 support where it's available and since the bulk of support isn't in the driver layer but in the API layer...

I'd be positive of 1.5 support, I'd suspect strongly of 2.1 support with provisos on what ARB extensions they advertise being available.

[Svartalf]

IMHO AMD should drop the development of fglrx for some time and focus on doc and code releasing. Last fglrx driver seems to be quite nice already so people shouldn't complain if they know that AMD for the time being is focusing on the docs and open drivers...

Yeah. The unrest in the AMD user camp (raises hand...got parts, I do, I do...) is, I think, due to the lack of truly quality results on the proprietary front along with a glacial pace for the documentation and the honestly opened stuff.

It doesn't help that even IF the driver works well enough in 3D and doesn't boggle your system that the 2D performance is seriously subpar and things like FBOs are just simply not implemented in the new codebase. Seriously. Everyone's trying to get off of the PBuffers wagon (which is a dead mess to work with and horribly non-portable...) and onto the FBO wagon to do programmatically generated texture surfaces for rendering. FBO's make it part of the OpenGL API and make it really easy to use, less resource consumptive and doesn't need any UI environment to do it.

I could live with and even maybe tolerate that if the pace of opening up were moving faster. Because there'd be a concrete, definite time that we'd see being able to honestly use the hardware.

[bridgman]

Mixed open/closed source driver discussion continued here : http://www.phoronix.com/forums/showthread.php?t=7647

Intel document discussion, carry on...

[Saist]

Wanted to comment on the Project Larrabee mention. Unless something has radically changed, Intel isn't using that project to get into competition with AMD and Nvidia. According to a report posted by ArsTechnica here : http://arstechnica.com/news.ars/post/20070417-intel-officially-owns-up-to-gpu-plans-with-larrabee.html : The Larrabee Project isn't exactly shaping up to be a killer GPU to compete with the Shader Model based GPU's.

While Intel might indeed be aiming for the high performance GPU market... keep this article I wrote back in 2006 in mind : http://zerias.blogspot.com/2006/12/firefly-mmo-flop-in-progres-or-intel.html

[curaga]

Well, the Larrabee will be a huge gift for video encoding/decoding.

[drag]

Intel isn't using that project to get into competition with AMD and Nvidia.

Of course they are. Those are Intel's main competitors in most everything they do.

It's just that it's confusing that Intel is not aiming for peak DirectX 10/11 gaming performance. Which is fine with me, that's not something I particularly care about.

The GPU is not just for accelerating games anymore. It's a co-processor that can be used to augment the overall performance of your machine. What runs on top of them is software like anything else... and Intel seems to be aiming at making their GPU easy to use for lots of different tasks in addition to gaming.


According to Mesa's site, they're advertising 2.1 support where it's available and since the bulk of support isn't in the driver layer but in the API layer...

Ya.. OpenGL does not work like DirectX. If Mesa supports 2.1 then all the drivers that are based on that version support 2.1 also, more or less.

What matters is how much of the API the graphics card accelerates and how well it can do it.

The GMA X3000 and GMA X3100 are the most advanced IGP that Intel is offering at this time.
It supports pixel and vertex shading model 3.0. It can do Anisotropic filtering up to 16 times. It has a theoretical fill rate of 1067 megapixels/s and 2133 megatexel/s at 667mhz.

This puts it roughly on par with NV40 (Geforce 6) in terms of hardware features.

Does this mean that it will perform on par with Geforce 6 stuff? Nope. 'fraid not. Gaming performance of the GMA X3100 on Linux right now is best described as 'lousy'.

If your a open source purist (I am a dirty purist ;) ) then that means that if you restrict yourself to open source drivers then Intel will outperform Nvidia at any time. But if your going to use Nvidia's proprietary drivers then Nvidia will provide a night and day performance increase over any of nvidia's offerings for any remotely modern Nvidia card.

If your gaming requirements are light (Intel GMA X3100 can drive 'Return to Castle Wolfenstein' comfortably), you only want a 3D desktop, or your aiming for best power management features (for a laptop) and that sort of thing then I'd recommend getting a Laptop with Intel IGP.

Otherwise if you want gaming performance then a low-end Nvidia card will serve you much better.

[Svartalf]

This puts it roughly on par with NV40 (Geforce 6) in terms of hardware features.

Does this mean that it will perform on par with Geforce 6 stuff? Nope. 'fraid not. Gaming performance of the GMA X3100 on Linux right now is best described as 'lousy'.


Heh... On paper, they're nice parts. The biggest problems with their real use in gaming is partly that we've not had a lot of stream processor development under our belts to know how to really optimize the shader operations needed for 3D rendering coupled with the hamstring of UMA operation. (Even WITH good drivers, the X3000/X3100 have been still lackluster- though vastly better than the past offerings. The X3500 is the first "real" contender in the IGP space from Intel- and we're still going to be fighting the uphill battle of the stream processor execution optimization for a while yet.)

In the end, I think having the tech data, including how to drive those shaders through their opcodes, will end up being a boon. Right now, I'm seriously considering getting a G35 chipset motherboard to play with and see what all I can tweak. If I had AMD's stuff right now (hint...hint...;)) I would already be DOING it with machines in hand since I've got several R300/R400/R500 boards in hand. Heh... It's probably better, short term, though, as I've got too damn many irons in the fire already... :D

[curaga]

I'm going to get a mobo with G35 soon too. It would've been G33, but the model I want has been out of stock for months. So I'll just wait for the improvement of X3500 :)

Hmm. Didn't Intel claim the X3500 will archieve double 3dmark score compared to X3100?

[R3MF]

does the document release include thespec for the Clearvideo hardware, so the FOSS community can make use of hardware video acceleration?

[Svartalf]

I'm going to get a mobo with G35 soon too. It would've been G33, but the model I want has been out of stock for months. So I'll just wait for the improvement of X3500 :)

Hmm. Didn't Intel claim the X3500 will archieve double 3dmark score compared to X3100?

Yes. It's their first relatively credible IGP part. The X3000 and X3100 fell a bit short of that. I wish I had a better budget for things- because I could get one right now if it weren't for all the other things I need to be buying, including getting my roof fixed.

[Extreme Coder]

Does this mean that the drivers for the 965 IGP will improve? Because I have a friend with that mobo, and a lot of games crash or can't run (for example, Regnum Online), while with his previous mobo(also an Intel, but it was 850 I think) it was working fine..
I was really disappointed at Intel :P

[DavidC]

I'm going to get a mobo with G35 soon too. It would've been G33, but the model I want has been out of stock for months. So I'll just wait for the improvement of X3500

Hmm. Didn't Intel claim the X3500 will archieve double 3dmark score compared to X3100?

Eh, no... It's the one AFTER the X3500 that'll be 2x faster. Actually it'll be 1.7x faster Intel says. X3500 has no perceivable architectural difference from the X3000. Hell, can anyone tell me what advantage X3500 has over the X3000 other than that it supports DX10?? I think DX10 support is just disabled on the X3000, so there's really no difference at all.

X3100 is laptop, btw.

[curaga]

Oh yes. Well, I ended up getting the Gigabyte P35-DS3L without integrated graphics, and putting my best two pci cards in for a dual screener. Saved 40$ when compared to the G33 board.