Intel releases graphics programming manuals [LWN.net]

Intel releases graphics programming manuals

Posted Feb 1, 2008 4:04 UTC (Fri) by allesfresser (subscriber, #216) [Link]

Seconded, in spades.

Intel releases graphics programming manuals

Posted Feb 1, 2008 8:14 UTC (Fri) by kripkenstein (guest, #43281) [Link]

Thirded. I'd buy such a product right now if it were available.

Intel releases graphics programming manuals

Posted Feb 1, 2008 8:57 UTC (Fri) by einstein (subscriber, #2052) [Link]

Fourthed. I'd also vote with my wallet.

Intel releases graphics programming manuals

Posted Feb 1, 2008 9:55 UTC (Fri) by smitty_one_each (subscriber, #28989) [Link]

Fifth, votes with bottle. *ba-dump-bump*
Seriously, I specified the Intel chip when I bought the current laptop from Dell.  One hopes
that this splendid announcement is related to people avoiding evil lock-in, and that other
vendors echo the clue.

Intel releases graphics programming manuals

Posted Feb 1, 2008 12:48 UTC (Fri) by pheldens (guest, #19366) [Link]

6theth
I'd replace my r300 (x700 pcie)

Intel releases graphics programming manuals

Posted Feb 1, 2008 13:44 UTC (Fri) by daniels (subscriber, #16193) [Link]

Why? AMD's specs are completely open, and that driver is also supported by AMD developers.

Intel releases graphics programming manuals

Posted Feb 1, 2008 15:32 UTC (Fri) by ajross (subscriber, #4563) [Link]

No they aren't.  They've released only 2D specs so far.  More have been promised (for going on
6-7 months now), but none have been delivered.  Intel's release is the very first
documentation on the 3D pipeline of any hardware of which I am aware.

Intel releases graphics programming manuals

Posted Feb 1, 2008 15:55 UTC (Fri) by intgr (subscriber, #39733) [Link]

If you've been paying attention to the ten-year timeline series then you should note that:

October 21, 1999: [...] ATI announces that it will be releasing 3D programming information for its video adapters - the good news here is that it's finally getting around to doing that.

ATI, and later AMD, have been promising to release hardware documentation repeatedly, for over 8 years. All this time they've been mocking the open source community with empty promises, and have finally gotten around to releasing fairly straightforward 2D specs that have been reverse engineered anyway. They've also been mocking us with closed source drivers.

Intel, on the other hand, has been developing open source drivers for a while now — though without open documentation. And instead of making big announcements about releasing 3D documentation, they just did it! Don't give any credit to ATI/AMD when it comes to graphics hardware, they don't deserve it.

Intel releases graphics programming manuals

Posted Feb 1, 2008 18:55 UTC (Fri) by rahvin (subscriber, #16953) [Link]

AMD released full spec's for the hardware (right down to register information), a base 2d driver and hired Novell to write a full driver. They stated at the time that they can't release the 3d driver because it contains licensed code. They said they MIGHT release parts of the 3d driver if it's useful once development reaches that stage but there will be major and significant holes because of the licensed code they can't release.

Ending off the X Developer Summit this year, Matthew Tippett handed off ATI's GPU specifications to David Airlie on a CD (as reported by Daniel Stone). However, the specifications are also now available on the Internet! At http://www.x.org/docs/AMD/ is the location of the documentation where you can freely download the files. Right now there is the RV630 Register Reference Guide and M56 Register Reference Guide. The RV630 Reference Guide is 434 pages long while the M56 Guide is 460 pages. Expect more documentation (and 3D specifications) to arrive shortly. The new open-source R500/600 driver will be released early next week. More information to come soon. Tell us what you think. For more information, read our ATI/AMD's New Open-Source Strategy Explained article.

http://www.phoronix.com/scan.php?page=news_item&px=NjA1Mw
http://www.phoronix.com/scan.php?page=article&item=826&num=1

Although AMD didn't release a full OSS driver along with it they did hire Novell to develop an OSS driver. This is FAR better than nVidia and only slightly a step down from Intel IMO.

Intel releases graphics programming manuals

Posted Feb 1, 2008 22:20 UTC (Fri) by daniels (subscriber, #16193) [Link]

Er, you do remember that the r200 3D driver was developed by Red Hat/The Weather Channel with
ATI's contribution and support, yes?

Intel releases graphics programming manuals

Posted Feb 1, 2008 23:44 UTC (Fri) by drag (subscriber, #31333) [Link]

R200 was a _long_ _long_ time ago. And it was obsolete before the drivers actually got stable
enough to use for desktop work.

I don't think Redhat had anything to do with it either, but I don't remember. The Weather
Channel sponsored it and Tungsten Graphics were the ones to originally develop the driver.

ATI only released documentation to a very select group of people under NDAs. They also did not
release full documentation on the 3D hardware.. what they did release was stripped down and
lacked information about many 3D interfaces and the card's debugging facilities. For the R200
to end up being stable a great deal ended up having to be reverse engineered.

Intel took the same path with their driver development also. They originalled released
documentation to Tungsten Graphics under NDA and had them develop the original driver for the
915 series chipsets before they moved driver development in-house.

(Nothing against Tungsten Graphics; they are responsible for a hell of a lot good things for
Linux in terms of 3D graphics and deserve a lot more credit then they get)

Intel started off doing more then ATI ever did in terms of providing documentation with the
first GMA stuff.

What we have now is better then what ATI has promised and much much much better then anything
they ever delivered. There is realy no comparision. This is fantastic.

I do realy realy realy hope that ATI/AMD follow Intel's lead in supporting Linux in a truly
open manner. And, hell, if Intel comes through with it's discrete GPU plans it may make a big
of enough impact on to crack Nvidia's tight lips.

Intel releases graphics programming manuals

Posted Feb 2, 2008 0:13 UTC (Sat) by daniels (subscriber, #16193) [Link]

Sorry, Tungsten Graphics, not Red Hat.

The main problem I have is just with presenting AMD's work as done.  Intel got everything done
silently inhouse and dumped it all in one big load.  AMD presented documents as soon as they
were ready (John Bridgman started burning the CD around thirty seconds after his Blackberry
beeped to tell him that final signoff occurred), and it's still a work in progress -- the
easiest and most crucial part (modesetting) first, and then moving through the rest.  By
contrast, Intel just surprised people one day.

Intel releases graphics programming manuals

Posted Feb 1, 2008 22:28 UTC (Fri) by arjan (subscriber, #36785) [Link]

Yes this was the announcement...
... what they actually released was a 2D register dump without even 2D acceleration info..

Intel releases graphics programming manuals

Posted Feb 1, 2008 22:55 UTC (Fri) by daniels (subscriber, #16193) [Link]

And it turned out to be a very useful set of PDFs, too.  (Bear in mind that they're not
secreting away documentation: the only stuff they have is what they've already released, and
what they're currently working through the legal department to release.)

I'm sure you appreciate that it's extremely difficult to release these things -- the Intel
release didn't exactly happen overnight -- so why not cut your competitor some slack?

Intel releases graphics programming manuals

Posted Feb 2, 2008 0:12 UTC (Sat) by scottt (subscriber, #5028) [Link]

I think Intel deserve some bragging rights after investing in and releasing such a set of
in-depth technical documentation. That they were willing to pay technical writers to write
these docs then release them to the community surely deserves praise over AMD?

Intel releases graphics programming manuals

Posted Feb 2, 2008 0:27 UTC (Sat) by daniels (subscriber, #16193) [Link]

Absolutely -- what they've done is fantastic and they deserve no end of praise.

Intel releases graphics programming manuals

Posted Feb 2, 2008 1:11 UTC (Sat) by rahvin (subscriber, #16953) [Link]

I agree, Intel deserves immense amounts of praise, had they not made their move to open up
their hardware many others would likely not have followed. ATI doesn't deserve any praise.
They never delivered on their promises. AMD on the other hand deserves a round of applause.
They're the ones that released the hardware information and they are the ones paying Novell to
flesh a driver out that hopefully the community can take over.

The only point I tried to make was that AMD did release their information, not in the best
form, but I got the impression is was a "hey we better get this stuff out there!" after they
completed the acquisition of ATI. Rather than doing what Intel did and package everything up
nice and pretty before putting it out there they didn't want to wait six months and have
everyone disappointed in them. That's why IMO they wanted everyone to know that the ATI lies
of the past were over, so they released what they had available and are paying to get a driver
written (I think the guys at Novell also have direct access to the ATI hardware guys to ask
questions). I think that's commendable.

Given that nVidia is the only major vendor without an OSS driver (or at least the only one
without one under construction) and with manufacturers like DELL beginning the process of
requiring that their vendors have OSS drivers we will hopefully see some intense pressure and
possibly OEM losses that will force them to open up their cards as well. That pressure
wouldn't be possible had AMD not released the ATI specs. With only Intel in the OSS camp
nVidia had no pressure on the high end to release OSS drivers, they do now.

Intel releases graphics programming manuals

Posted Feb 2, 2008 1:50 UTC (Sat) by drag (subscriber, #31333) [Link]

Ya AMD has always been very good about their stuff. Better then Intel in the past. 

But ATI has sucked very much in the recent history, unfortunately. (as a seperate entity) It
seems that there is people in ATI that want to do the right thing, but corporations being what
they are, don't have enough support in the company to get it out the door. 

I understand better now the sort of stuff that is going on, now that I work for a company that
does some hardware/software development. All the different fractions, pressures, and divisions
in a company.  It can be quite a mess.. 

Both ATI and AMD were large, mature companies. Combining them probably performed the
equivelent of dropping a nuclear bomb on their corporate structure. I bet lots of people
inside that company now have little idea on what who they need to talk to and how to get the
hardware openned.. Legal folks, software folks, corporate folks, customer folks, hardware
folks, all are going to have a stake in it and have their own massive concerns.

For example Microsoft could (in a very sneaky manner) design the API for DirectX 11 to favor
NVidia's innovations and API extensions rather then ATI's. Could cause a significant boost to
Nvidia's performance and stability were ATI's stuff is going to seem 'old generation' and may
even require licenses from Nvidia to be compatable. This sort of thing could realy nail ATI to
the wall.

Intel releases graphics programming manuals

Posted Feb 13, 2008 11:00 UTC (Wed) by daenzer (subscriber, #7050) [Link]

> ATI doesn't deserve any praise.

Right - providing documentation, financial and other support to the relevant people for
developing free drivers (that are still being maintained in the open and widely used), almost
ten years ago - what a horrible thing to do. Shame on them.

</sarcasm>

P.S. Congrats to Intel on this release.

AMD releasing 3D specs

Posted Feb 3, 2008 10:54 UTC (Sun) by Felix.Braun (subscriber, #3032) [Link]

Does anybody know, whether and in how far AMD has actually followed through on the promise
that "more documentation (and 3D specifications)" were to be released "shortly"? It seems to
me that they'd have staged a big press conference for the occasion. So, because nothing of
that sort has ocurred I'm left with the impression that this promise was worth about the same
as the ones they made 1999.

AMD releasing 3D specs - not yet

Posted Feb 3, 2008 13:47 UTC (Sun) by khim (subscriber, #9252) [Link]

You can read about second dump here. Looks like they were too optimistic. They have published more info and they do have right person dedicated to the task, but no, 3D acceleration is not there yet...

Intel releases graphics programming manuals

Posted Feb 1, 2008 10:54 UTC (Fri) by danielhedblom (guest, #47307) [Link]

I would switch every machine i have in possession at an instance. I cant understand why Intel
doesnt do this.

Intel releases graphics programming manuals

Posted Feb 1, 2008 14:08 UTC (Fri) by drag (subscriber, #31333) [Link]

They are going to. Yes, indeed, Intel says they are coming out with a discrete video card with
advanced GPU on it.

It's currently called 'Larrabee'. It's a new GPU architexture designed to be easy to program
for. It will consist of 16-24 'x86-like' in-order proccessor cores and a completely
programmable pipeiline.

It should provide a _massive_ amount of floating point performance. Blowing the doors off of
anything coming out of Nvidia or ATI.. however due to it's design it will not be as highly
optimized for gaming as a traditional GPU and that will cause it to lag behind other video
cards in DirectX 10/11 performance. It would require gaming engines to be modified to take
full advantage of it. (At least this is the speculation on it right now).

But the cool thing about it is that it would be easy to program for, I think. You can take a
software-only OpenGL stack and only with a few modifications compile it to take advantage of
the GPU. This goes the same with any program or other API you can think off. Graphics
acceleration, realtime raytracing, media encoding, etc. Sort of like programming for SSE or
something like that. CPU++

All this stuff is a bit over my head. But I don't care.. if it's going to be open like this I
am getting one. Damn the torpedoes, full speed ahead. If it's open then Linux folks should be
able to embrace it so fast and turn Linux into a Floating Point powerhouse so fast it will
make Bill Gate's head spin.

I am speculating that since Windows is very slow in adopting new hardware architectures that
Intel is going to use Linux to provide a platform for developers who want to take advantage of
it early on. Similar to how AMD used Linux to provide advanced support for their 64-bit CPUs
long before MS released a stable 64bit version of their OS for it.

It's suppose to be out by late 2009 or early 2010.
Latest news about it that I know about:
http://arstechnica.com/news.ars/post/20080117-larrabee-be...

Here is a bunch of GPU geeks giving their insightfull 2 cents in on the subject:
http://forum.beyond3d.com/showthread.php?t=46393

Oh, and Tungstun graphics is working on a API that should come in handy for a very flexible
GPU. It's designed to replace the complexity of DRI drivers with minimal code and maximum
flexibility. They claim that they can take the a driver and reduce it down to a couple
thousand lines of code and not loose functionality. Something like that. They claim to have a
mostly working driver for the i915 chipset and have a driver for i965 and the Cell proccessor.
Cool stuff.
http://www.tungstengraphics.com/gallium3D.htm

Too late

Posted Feb 1, 2008 15:03 UTC (Fri) by jreiser (subscriber, #11027) [Link]

It's [Larrabee] supposed to be out by late 2009 or early 2010. Two years from today? Did the Itanic team "rescue" them?

Too late

Posted Feb 1, 2008 15:52 UTC (Fri) by drag (subscriber, #31333) [Link]

No. The Larrabee stuff is made by a seperate group of folks from within Intel setup specificly
for working on this GPU stuff. 

They called it a 'internal venture company' or something like that.

Remember CPUs are not like software. Intel and AMD and friends are always working on tech 2-3
generations out. The latest and greatest cpus you can buy right now from Intel were probably
designed in 2003-2004 or so.

How capable are these graphics chipsets anyway?

Posted Feb 1, 2008 14:16 UTC (Fri) by drag (subscriber, #31333) [Link]

The documentation is for Intel IGP video cards found in the majority of newer laptops and
desktops. Intel produces the majority of desktop-oriented video devices. It's a normal chipset
on-board memory-sharing video chipset.

The equivelent for Nvidia or ATI would be whatever they have integrated into their
motherboards. 

For raw performance pretty much everything ATI and Nvidia have outperforms the Intel stuff.
The advantages of Intel are open drivers and very good power management. I would not buy a
laptop without one. 

In terms of performance these video cards can be used very easily to run Compiz-fusion and
perform well up to about 'Return to Castle Wolfenstien' levels. You can probably get Doom3 to
run on one in Linux, but it would not be any fun to play. 

They are essentially crippled by the shared memory scheme. If Intel provided the same chipsets
with dedicated video ram then their performance would be on par with the lower end side of
Nvidia/ATI's offerings.

But if you want 3D desktop and have good power management performance then Intel is what you'd
generally want.

How capable are these graphics chipsets anyway?

Posted Feb 1, 2008 17:13 UTC (Fri) by johnkarp (subscriber, #39285) [Link]

Shared memory isn't necessarily a bottleneck with appropriate RAM. SGI 
graphics workstations used the scheme. Among other things, video input 
could be written directly into texture memory, if I recall correctly. 
Though it did help that they used dual-ported memory....

How capable are these graphics chipsets anyway?

Posted Feb 1, 2008 17:34 UTC (Fri) by drag (subscriber, #31333) [Link]

What I would personally like to see is the GPU integrated as a specialized CPU core into the
main proccessor and 512MB-1GB of very high speed ram integrated very close to that.

I think the high-bandwidth and low-latency memory that is typically spent on video cards would
go a lot better next to the CPU rather then sitting on the far end of a PCIE bus. Like a
'Level 3 cache' or something like that, then have cheaper main memory in it's normal position.


I think ultimately discrete video cards are going to die off, except as just I/O ports for
video connections. With 16+ cores avialable in desktop PCs in a few years it's not going to
make a lot of sense to have them all be the same type of core. 

Oh Well with what is currently being used with this Intel onboard the main memory is shared
over PCIe and utimately it takes away bandwidth from the CPU in order to refresh the video
display. Having dual channel, low-latency memory can help, but that only goes so far.

How capable are these graphics chipsets anyway?

Posted Feb 1, 2008 19:05 UTC (Fri) by rahvin (subscriber, #16953) [Link]

What I would personally like to see is the GPU integrated as a specialized CPU core into the main proccessor and 512MB-1GB of very high speed ram integrated very close to that.

I believe you just described both Intel's and AMD's plans for mid next year. I know for sure AMD has that on the roadmap, and I'm sure Intel does as well. The only thing I would say is likely different is that it's not going to be a single GPU core, AMD calls them XPU's and they'll apparently be processing units that are highly specialized to do the work that is typically done with graphics. The roadmap concept graphics actually show as many xPU's as there are CPU cores on the AMD side. I imagine these xPU's will more than likely replace the floating point pipelines in the CPU.

How capable are these graphics chipsets anyway?

Posted Feb 1, 2008 23:48 UTC (Fri) by drag (subscriber, #31333) [Link]

Ya, I am told that If you look at the current generation of GPUs on the market they are made
up of a many 'cores'. Graphics proccessing is something that naturally lends itself to highly
parrellel programming so it wouldn't suprise me if that was entirely true.

How capable are these graphics chipsets anyway?

Posted Feb 2, 2008 8:48 UTC (Sat) by Cato (subscriber, #7643) [Link]

It will be very handy for lower-end PCs and mobile devices to have the GPU and CPU on a single
chip, with a smarter memory hierarchy.  I don't claim to understand all this but I believe the
only thing stopping such integration is that CPUs are on a 12-18 month cycle time and GPUs on
about 6 months, so it may be that the integrated GPU-CPU combination will evolve at CPU rates,
i.e. taking a snapshot of the latest GPU technology and delivering in 12-18 month cycles.
There's also a commercial issue in that gamers buy a new graphics card every year or so -
non-integrated graphics will survive a long time in that market, until the rate of GPU speed
increases slows down quite a bit.