Winds of freedom for graphic drivers!!!

[ecastro]

I've read with pleasure the news about Intel releasing drivers for one of it's graphic cards open source! And almost at the same time AMD announced that it is seriously thinking of releasing ATI drivers to the open source community... might the era of open source have arrived to the graphic drivers?! I hope so...

[JaXXoN]

Originally Posted by ecastro

I've read with pleasure the news about Intel releasing drivers for one of it's graphic cards open source!

These are actualy not "graphic cards", but graphic engines
integrated into their chipsets:

http://intellinuxgraphics.org/documentation.html

I certainly welcome intel's decision to provide (open) sources
for their chipsets, but although intel has the biggest market
share concerning "graphic chips", i wouldn't say they are the
technology/performance leader in this field. Means: for the
average Linux user, i guess intels on-chip graphics will be
probably good enough for daily use, but for professional 3D
applications or "serious" gaming, i think the competition is still
in a much better shape.

Originally Posted by ecastro

And almost at the same time AMD announced that it is seriously thinking of releasing ATI drivers to the open source community

Could you please provide a link for that announcement?

regards

Bernhard

[ecastro]

I read it here...

http://www.infoworld.com/article/06/...OPcurve_1.html


I know Intel are not like nvidia or ati... but it is a great news anyway... and if amd follows this steps and delivers open sources drivers the nvidia would be in real trouble... and would have a lot of pressure to deliver open sources drivers as well....

[morganth]

Dear Bernard,

I certainly agree with you that as of now, ATI and Nvidia, and Nvidia in particular, produce better graphics cards than Intel does. Intel's cards are only meant for "general workstation" use -- serious 3D requirements mean going with one of the big two.

However, I still can't really understand the argument for keeping these drivers binary. If there is one piece of software that shouldn't be closed source, it's hardware drivers. Keep your DVD player and your desktop software closed, but drivers should be open.

I just shopped around for a laptop, and chose one with an Nvidia chipset because I knew that Nvidia's binary drivers are better than ATI's. (That and, I've always had a better experience with Nvidia cards on my desktop.) My same laptop, however, was offered with the Intel 955 chipset. If I had waited just two weeks to make my order, I would have chosen Intel over Nvidia without question, given this news on their decision to open source. I'm not a Free Software or OSS zealot, and I don't believe software "needs" to be free or any of that garbage. But I do believe that _hardware_ that I purchase (or, as is increasingly the case, am forced to purchase due to lack of choice) should be defined by OPEN standards, and should be supported by OPEN drivers. This is the only way to ensure that my hardware will continue working into the future, even if NVidia dies, and it's the only way for me to know that my operating system of choice (Linux) will always support my hardware of necessity.

A lot of people think the way I do. I won't say millions of people, since most people don't even know what a driver is, nevermind distinguishing between closed and open ones. But definitely a serious number of technologically-savvy individuals want to buy their next computer so that it works with Linux. Even my friend who is a law student and couldn't care less about computers and software, when asking me about my current laptop, said to me, "Do you think that laptop will run Linux properly -- I'm thinking of trying it out." Binary blobs for drivers is one more way to give the average user headaches in an alternative OS.

By the way, the news about ATI is reported here:

http://www.infoworld.com/article/06/...OPcurve_1.html

Andrew Montalenti

[_tf_]

nicety!!!

thanks to intel and (soon?) amd!
i think this will give the X development a *great* push to develop really cool new graphics code!
... and the end user that "yay, i just compiled thie new kernel and there is *no* need to reinstall that nvidia binary driver!" effect! ;-)


again, thanks!
rSl

[JaXXoN]

Originally Posted by ecastro

I read it here...
http://www.infoworld.com/article/06/...OPcurve_1.html

FTA:

Lastly, and remember you heard it here, AMD is strongly considering 
open-sourcing at least a functional subset of ATI’s graphics drivers.

This doesn't necessarily mean a lot :-)

Originally Posted by ecastro

if amd follows this steps and delivers open sources drivers the nvidia would be in real trouble...

No, they are not: although NVIDIA may have 95% market share
amongst "serious Linux 3D user", this probably only represents
2 percent of their overall business (if at all). So if they would be forced
to open source their drivers by competition, then nvidia might simply
consider dropping linux support altogether ...

regards

Bernhard

[Thunderbird]

One comment regarding Intel graphic cards. Their latest GPU isn't bad at all though I haven't seen benchmarks yet. Intel has released two new integrated gpus for their i965 chipset the GMA 3000 and the X3000. The X3000 model supports pixel shaders 3.0, vertex shaders 3.0 and also geometry shaders (dx10) and more stuff. It is a very advanced GPU but I don't know how fast it will be but for 3d features it looks good. The plain GMA 3000 is less advanced it does pixel shaders 2.0 (or perhaps 3.0) in hardware but vertex shaders in software and it is less advanced in other areas.

So their latest integrated graphics might not be as bad as 'traditional integrated graphics'.

[JaXXoN]

Originally Posted by morganth

However, I still can't really understand the argument for keeping these drivers binary.

One oftenly repeated argument is that parts of the "inner" driver
sources are copyrighted by third parties. So, we'd need to chase
this third parties to allow nvidia to open source their sources,
but who are these third parties? AFAIK, nvidia was founded
by a bunch of SGI graphics gurus, so I can imagine they licensed
the SGI OpenGL reference implementation for a starter (pure
speculation!). BTW: the SGI OGL RI is OSS since a couple of years.

Another subject is "patent issues": somebody at the competition may
be able to understand how the chip works inside by analyzing the driver
sources. Then he/she may recognize that a patent has been infringed
and all hell breaks loose. Means: whoever shows his code first may
fear to be sued. Even if nvidia and AMD (ATI) would agreed to open
source their drivers at the same time (mutual patent licensing
agreement), then there might be still some patent trolls sitting around
waiting for their opportunity. (BTW.: from that point of view, i think it's
very brave from Intel to open source their drivers :-)

Originally Posted by morganth

I'm not a Free Software or OSS zealot, and I don't believe software "needs" to be free

Me either: as long as the damn thing works as i expect, i basically
don't care if its OSS or not - however from my experience with the
nvidia driver i can say that it's always a PITA to get closed source
drivers working with non-standard kernels such as realtime preemption
or xen (check this forum for details).

nvidia already made developers life much easier by abstracting
the driver into an os-dependend part (source code available) and
an os-independend part (object code) so that (in theory) it should
be possible to adopt the driver for any enviroment, but there's
not much you can do when you see that the driver got stuck
in function __nvrm000072() while in interrupt context or whatever.

I think the stability issues observed are caused in the 2D part
(i.e. console switching) and DMA related (AGP). I think it would
extremly helpfull if nvidia would further abstruct their driver so that
the 2D part and the DMA interface code are also available in source
code. (not necessarily "open source" - i can life with a couple of
OSS zealots complaining that non-GPL kernel modules are illegal)

Technolgy/Copyright/Patent wise, 2D and DMA are definitly no
rocket science! The "precious" 3D stuff could still be object code.
Idealy the driver could even work without the 3D code (optionally
loadable firmware). This "infrastructure" would make it *much*
easier to isolate bugs. But i for one don't hold my breath for this
to happen :-)

regards

Bernhard

[JaXXoN]

Originally Posted by Thunderbird

Their latest GPU isn't bad at all

My guess is that the 3D performance is somewhat comparible to
the GeForce 6150 chipset. I can imagine that lots of "average"
Linux users may consider purchasing intel HW in near future
(Core2 Duo is also not bad at all, BTW.)

Originally Posted by Thunderbird

though I haven't seen benchmarks yet.

A good benchmark might be not that easily feasible in this case, since
the 965 requires an intel CPU and the 6150 requires an AMD CPU. Means:
it's virtually impossible to compare the 3D performance directly.

regards

Bernhard

[MamiyaOtaru]

I saw that infoworld article and was excited. Since then, news blurbs have popped up all over about AMD opening up specs, and I check them to see if they provide more concrete information. But they all link back to that infoworld article, which is nothing but speculation. Disappointing.

Nevertheless, the possibility is enough to keep me from buying nVidia right now. In the meantime I can hold out with intel, whose stuff will be open.

I keep hearing how people just want their cards to work, and all this open sources stuff is overblown; binary is fine. That's a wonderful selfish attitude. It ignores the wishes of the people who built the kernel you are using. It hijacks their efforts. If a user wants to ignore the intent of the kernel's developers, perhaps the user should use a different kernel. There is no place for closed source in the kernel. If closed source is so cool, why don't you marry it?^H^H^H^H use Windows or OSX?

In nVidia's defence, we've heard that intel holds IP rights for stuff relating to PCI express, and would throw the book at nVidia if they released anything to do with that (and that intel already did such a thing in the past). All I can say to that is that nVidia should be hard at work coming up with a hypertransport solution with AMD instead of being complacent and letting intel control them.

Hypertransport is possibly what will allow AMD/ATI to finally open up, and I'll be happy to wait and see if it happens before buying my next GPU. In the meantime I can stick with intel graphics, even if they are the villains in the story. This is out of the pragmatism so often touted here: I just want my stuff to work. I don't want to wait months for my card to work with xorg 7.1, with xgl, with a new kernel, whatever. Give me a driver that can be modified as needed by people who seem to actually care.

So, Intel may control the method by which the GPU is connected to everything else, allowing them to be open while preventing others from doing the same, but nothing stops others from developing their own way to connect (see hypertransport). Because they choose not to do so I'm supposed to feel sorry for them and accept the suboptimal solution of binary blobs? Thanks, no. Intel for now, and AMD/ATI if they can ever come up with the gumption to go their own way and open stuff as well. nVidia, who seems to have no intention of doing so, can count on my 5500 being my last purchase from them until they get off their butts and do the same.

Can't tell me AMD wouldn't have been happy to work with nVidia on HT, and that nVidia couldn't have done it, what with nVidia producing everyone's favourite chipset for AMD. Oh well, opportunity lost.

[JaXXoN]

Originally Posted by MamiyaOtaru

If closed source is so cool, why don't you marry it?^H^H^H^H use Windows or OSX?

Please choose:
[ ] binary only 3D drivers
[ ] no 3D drivers at all.

Originally Posted by MamiyaOtaru

In nVidia's defence, we've heard that intel holds IP rights for stuff relating to PCI express

Can you please provide a URL? I'm not sure where PCIe might be
an issue in this case: from what i know, you can get the whole spec
from the PCI SIG for a nominal fee. There might be patented
HW implementation details, but then this information is public
by definition :-)

Originally Posted by MamiyaOtaru

So, Intel may control the method by which the GPU is connected to everything else

I'm pretty sure that PCIe is - if at all - a very minor issue: if this
would be the only obstacle, then nvidia could easily open source
some drivers for AGP :-)

I think nvidia truely fears patent infrigements: in the USA patent system,
you can basically get anything granted, even if it's prior art - you just
need some good (bad?) patent lawyers obfuscating the patent so that
almost anything in the universe my apply to the claims :-) A good
(actually bad) example is RIM (Blackberry) who payed ~ $500 million
to NTP for something that has later been deleted by the USPTO.
Funny thing is that the seattlement occured outside the court room,
so there is no way for a refund :-)

regards

Bernhard

[MamiyaOtaru]

Originally Posted by JaXXoN

Please choose:
[ ] binary only 3D drivers
[ ] no 3D drivers at all.

In light of recent announcements from Intel and potential developments at AMD, this is a false dichotomy. I no longer have to choose either (which unfortunately means not choosing nVidia).

Can you please provide a URL? I'm not sure where PCIe might be an issue in this case

http://lists.suse.com/archive/suse-l...-May/3373.html
and more detail in a followup:
http://lists.suse.com/archive/suse-l...-May/3517.html

I think nvidia truely fears patent infrigements

Of course. So out of the oft championed pragmatism and a desire to run 3d apps with Linux, I'd have to go with a vendor that doesn't have those crippling fears (ie Intel, maybe AMD later). Part of that fear seems to have to do with PCIe (which is why I brought it up), at least according to the links provided, which is why I lament the current lack of action with hypertransport GPUs and look forward to possible future such action from AMD. I am of course not an industry insider, so I cannot vouch for the correctness of the information provided, but it certainly would explain a lot.

I'd love to spend hundreds of dollars to get faster 3d performance, but I'm no longer willing to do so if it comes with a binary blob, and flies in the face of the spirit of the OS I choose to use. If someone else is willing to do so, I postulate that Linux is the wrong choice for that person. Linux can hardly be said to be easier than Windows (though it is for me, even if closed drivers make it worse), and a carefully tended Windows box would be far better for gaming anyway. I just don't get why some feel pragmatism should trump idealism in these debates. If one pragmatically wants his games to just work, openness be damned, one simply has to use Windows. Anyone who can understand Linux can figure out how to keep Windows clean. Idealism is all Linux has going for it in this case, and shouldn't be dismissed and mocked so quickly.

In short: if one must have 3d drivers, and binary drivers were the only option, Linux would have no point, so I'm very glad we have more options than were given in your artificially binary choice. I just wish one of those options was nVidia.