AMD Releases 900+ Pages Of GPU Specs

[phoronix]

Phoronix: AMD Releases 900+ Pages Of GPU Specs

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.

http://www.phoronix.com/scan.php?page=news_item&px=NjA1Mw

[mlau]

Wow, that's some dry reading...

[Thetargos]

Wow, AMD IS delivering... interesting, most interesting indeed!

[yoshi314]

nice. can't wait to get my hands on the driver. the new fglrx instantly dissapointed me - it might be new and "magical", but it still has all the shortcomings of a binary driver.

i'll keep waiting for this one instead.

/me goes to look at the specs :]

[Nicolas]

That's nice... Now we can hope for a decent driver...

I haven't tried 8.41 as I left my notebook at home today.

Will probably wait for 8.42 with AIGLX support and more bug(fixes hopefully).

[Xipeos]

I would be tickled beyond words for opened sources, but that's NOT likely.
Tickled enough, Svartalf? :)

[taipan]

I just registered to say I love you
I just registered to say how much I care
And I mean it from the bottom of my heart
:D

[Chandon]

Are these documents the complete specs that a programmer would need to implement good quality free drivers? Are they real? Do they work?

Can I really buy ATI hardware now and expect free drivers for it reasonably soon, or is there going to be a post in two weeks saying "ATI Specs Incomplete / Useless"?

[Michael]

Are these documents the complete specs that a programmer would need to implement good quality free drivers? Are they real? Do they work?

Can I really buy ATI hardware now and expect free drivers for it reasonably soon, or is there going to be a post in two weeks saying "ATI Specs Incomplete / Useless"?

These specs document the 2D portion of the RV630 and M56. More documents will be out in the coming days and weeks (including the 3D portion).

The first OSS driver will be released is Monday.

[Thetargos]

Alas a working quality driver (full driver) will probably be still at least six moths off.

[koolmanoncampus]

Alas a working quality driver (full driver) will probably be still at least six moths off.

Even if that is true, at that rate we should expect a driver more functioning than fglrx in about 2 months. Even if it lacks 3d initially; I'd love a driver that has randr dynamic monitor support and can have decent video playback.

[Thetargos]

Most certainly. At least we can be certain of a few features that will be there in the OSS driver, and most likely it'll feature pretty fast 2D rendering.

[Nicolas]

Most certainly. At least we can be certain of a few features that will be there in the OSS driver, and most likely it'll feature pretty fast 2D rendering.

It's a good start :).

[joshuapurcell]

I just registered to say I love you
I just registered to say how much I care
And I mean it from the bottom of my heart
:D

I wonder how many people on this forum actually know what you are referring to?


This is great news... anyone want to buy my Nvidia video card (specs in sig)? :D

[Xipeos]

This is great news... anyone want to buy my Nvidia video card (specs in sig)? :D

I'll give you 20$ for it. Honestly, it's just a heap a metal that's soon to be obsolete (as soon as directx 10.1 comes out, if you even care about directx). On linux, it's already obsolete.

What do you say? :)

[Thetargos]

I wonder how many people on this forum actually know what you are referring to?
What are you talking about?!?! Stevie was a Wonder!!

This is great news... anyone want to buy my Nvidia video card (specs in sig)? :D
I'd hold on to it for a bit longer until we have the first fruits of this ;)

[slumbukken]

Quick scan shows no definition of the internal programming model - is that available already elewhere?

[Svartalf]

Tickled enough, Svartalf? :)

I think I might be- the driver's story is leaving me...underwhelmed...yeah, that'd be the best, most diplomatic word for it. I'll see in a bit when I download and start perusing it. If it is good and useful (and I'll know pretty quick if it is or not) I'll be much more tickled than I would otherwise be.

[TechMage89]

Huh, I wonder if even the amd driver developers know what the internal programming model is. Graphics chip manufacturers are notoriously paranoid about people finding out how their chips work. At least now we have the barebone specs. Once it's working it can be experimented with and performance-tuned.

@Svartalf
What more do you want? Access to the fglrx tree? We've always known that would never happen: amd couldn't do that if they wanted to for the previously explained legal reasons. Now we have an excellent chance for a really quality graphics driver for Linux. Granted, it will take a year before it's even close to feature-complete, but as long as people want to develop it, the specs are there (I have them on my hard drive, at least the 2d ones that have already been released) and there won't be the r200 issue this time around, what exactly do you want? It to cure hunger, AIDS and cancer?

[joshuapurcell]

I'll give you 20$ for it. Honestly, it's just a heap a metal that's soon to be obsolete (as soon as directx 10.1 comes out, if you even care about directx). On linux, it's already obsolete.

What do you say? :)

I guess I'm going to have to try and sell it somewhere other than a Linux-related forum when I get serious about it. It's not worth $20 bucks to take it out of my case, box it up and drive to the post office :D .

[joshuapurcell]

I think I might be- the driver's story is leaving me...underwhelmed...yeah, that'd be the best, most diplomatic word for it. I'll see in a bit when I download and start perusing it. If it is good and useful (and I'll know pretty quick if it is or not) I'll be much more tickled than I would otherwise be.

I hear what your saying and I understand why. But as you can see from the amount of posts about this news here and on other sites, it's obvious that people want to believe this means major changes for Linux gaming. I'm willing to buy in completely to the optimism though... it can't hurt.

What can hurt though is if AMD/ATI screws this up somehow, and I don't mean by releasing the needed specs along with an open-source driver that blows... I mean releasing an open-source driver that blows and not following that up with the specs for the Linux community to do something to fix it. The only thing I really hope comes out of all this is for open source developers to have the tools necessary to improve (or create if needed) a 3D driver. If that doesn't happen eventually then this really isn't something to get all that excited about. So far we don't have any reason to believe this won't happen.

[deanjo]

It will be a interesting year in review next Sept. Opensource... you now are getting what you need..... time to put up or shut up.

[yoshi314]

take a look at the comments at digg/slashdot. looks like nvidia became the "linux bad guy" now :).

what a sudden turn of events :D

maybe they (nvidia) are already concerned about this.

[deanjo]

Could also backfire with Nvidia saying "screw linux development. They are a small part of our revenue and it's cheaper to let ATI concentrate on the small niche markets then to worry about it ourselves"


If after a year a person can't play UT3 at faster framerates then the closed source alternatives it actually adds fuel to closed source advocates saying "See, we gave you what you need, performance still is superior in closed source solutions." If that happens the entire FOSS movement gets a HUGE black eye that squanders future FOSS attempts.

[greg]

Maybe the expectations of many people are just too high. Developing OpenGL drivers isn't easy and there are very few developers working on it, while ATI/AMD and NVidia have lots of real experts and a lot experience at hand - which are working full-time on the drivers.

I don't think we'll have a driver in a year that outperforms ATI's closed source driver, we'll be lucky if it can keep up with it. Open Source doesn't magically solve all problems.

However, this driver will probably much better in other aspects, like Xrandr (hotplug) support or 2D acceleration.

[dkasak]

This hasn't changed my mind on buying only Intel systems. ATI have repeatedly lied to their customers about supporting Linux, only to drop things half-way through, with no explanation. R100 users were swept under the carpet. R200 users were told they'd be supported by fglrx, then later told that the open-source driver was their only option. R300 users have had a very buggy fglrx, or the completely reverse-engineered open-source driver.

I own all 3 of the above. My current laptop has a Radeon X700 mobility - not the latest hardware, but certainly not to be snorted at. Nothing that ATI has done - EVER - has helped the open-source driver for this card.

So while people with rose-coloured glasses proclaim their undying love for ATI, I'm sorry, but I've been around too long to be fooled by this. When ATI release full specs for non-current hardware, ie R300, I may concede that things have changed. Until then ... I've heard it all before - this is a marketing exercise to drum up support for their latest cards. In other words, ATI feels that owners of older hardware can go to hell ...

[joshuapurcell]

When ATI release full specs for non-current hardware, ie R300, I may concede that things have changed. Until then ... I've heard it all before - this is a marketing exercise to drum up support for their latest cards. In other words, ATI feels that owners of older hardware can go to hell ...I don't expect this to be high-priority for AMD unless one of their old chipsets is used to show how well this new process can work. In other words, this may happen if an older chipset turns out to be so much easier to get a good 3D driver working compared to the latest chipsets.

Even if they don't provide tons of quality support for older chipsets it won't change my outlook on this situation. AMD/ATI could even announce that they will be providing no information on any card later than their upcoming XXXEliteX3000XXX card next year, and as long as the support they do offer on that upcoming card is quality then I'll be a happy customer. I'm interested in getting good quality 3D working on my Linux box, and my definition of 'good quality' gets better and better with each new generation of video hardware and new games released. I'm sure I'm not the only one with this outlook.

Older hardware is not nearly as capable as newer hardware in running the latest games, and getting the hardware support into Linux for new games can help sway the gaming industry towards adopting Linux-friendly development practices. That means more games for Linux users, and in turn a greater need for more 3D support in Linux. If AMD is faced with choosing between releasing documentation on more of the latest features of upcoming hardware vs. getting all relevant information together for older chipsets, then I hope they will go with the new hardware every time.

[Thetargos]

Could also backfire with Nvidia saying "screw linux development. They are a small part of our revenue and it's cheaper to let ATI concentrate on the small niche markets then to worry about it ourselves"

You are looking at the Linux market from the wrong perspective: It has never been about the consumers! Not for ATi, nor for nVidia! It has always been about the professional market. Initially consumer support was a "by product" simply because in many instances consumer-grade hardware was being utilized for Professional-grade tasks. This is how both companies started off, and since then the consumer market in Linux has grown (still not to the point of being the sole justification of its existence).

They have acknowledge that and the Linux consumer market is expanding at a steady pace. AMD has many interests in this sector, not only for Professional or Consumer hardware, but for future developments. I'd venture to guess that one of the main applications of the upcoming GPU/CPU products will be precisely one of the most talked about projects in the IT world: The OLPC. Currently it is powered by AMD's Geode chips, but in the future I'm sure it'll be powered by a more powerful CPU with a GPU on the same die (allowing for even better power consumption performance), and guess what does the OLPC runs? Here's a hint, it ain't a M$ OS.

[lenrek]

... Currently it is powered by AMD's Geode chips, but in the future I'm sure it'll be powered by a more powerful CPU with a GPU on the same die (allowing for even better power consumption performance), and guess what does the OLPC runs? Here's a hint, it ain't a M$ OS.

My company build appliance system for end user, now we are using VIA processor. If this ever happens, I will definite recommend this to my management. :D

[Malikith]

Alright I just flipped through that thing, god damn that is a big book. I could never write a driver off that, I'd go insane haha.

[MrPeach]

Are these documents the complete specs that a programmer would need to implement good quality free drivers? Are they real? Do they work?

Can I really buy ATI hardware now and expect free drivers for it reasonably soon, or is there going to be a post in two weeks saying "ATI Specs Incomplete / Useless"?

The specs are incomplete/useless.

(at least as far as doing any VIVO or accelerated gfx is concerned - the best these specs will do is get an image on the VGA display. After that it's basically framebuf time.)

[yoshi314]

The specs are incomplete/useless.almost everyone is missing the point.

these are just initial docs from ati/amd. more will follow. of course you can't expect to see avivo or tv-out or full 3d specifications explained in them, for obvious reasons.

[Michael]

These initial specs are useful at this point for the driver... Look at the recent Avivo activity with all of the changes attributed to the specs.

[libv]

Are these documents the complete specs that a programmer would need to implement good quality free drivers? Are they real? Do they work?

Can I really buy ATI hardware now and expect free drivers for it reasonably soon, or is there going to be a post in two weeks saying "ATI Specs Incomplete / Useless"?

These docs are not complete, even for those parts of driver development they should cover.

They are, however, and this is an important distinction to make, all that can currently be dragged out of ATI. You have to understand that this is completely new for a company like ATI, and that it is very non-trivial to get this documentation and get it approved for release.

They are also not overly useful on their own, a lot of what my colleagues and i have been doing is poking and guessing until it somewhat worked. We just have these bare register specs and no actual programming documentation like there used to exist in the 90s and earlier.

[djdoo]

Hmm...
Let's hope that they will not stay there for a long time and release soon some "hotter" info...

Maybe they just prepare the new package of info right now, or discussing what info this will include...

Who knows?... Maybe Michael??

But I certainly have a good feeling! If policy changed completely about the approach of the community by ATI-AMD I believe we will see more and soon.

I am really tired of those nvidia guys bragging all the time about their linux advantage empire (full of bugs see Composite, Suspend, not even XvMC works bugless)...

Jim

[chrisr]

These initial specs are useful at this point for the driver... Look at the recent Avivo activity with all of the changes attributed to the specs.

I have just tested the latest avivo driver from git and it still has the "screen is corrupt after you close the lid and then reopen it" bug. (The laptop's screen powers down when the lid closes, of course. However, it doesn't seem to reinitialize correctly afterwards.)

[libv]

Hmm...
Let's hope that they will not stay there for a long time and release soon some "hotter" info...

Maybe they just prepare the new package of info right now, or discussing what info this will include...

Who knows?... Maybe Michael??

But I certainly have a good feeling! If policy changed completely about the approach of the community by ATI-AMD I believe we will see more and soon.

I am really tired of those nvidia guys bragging all the time about their linux advantage empire (full of bugs see Composite, Suspend, not even XvMC works bugless)...

Jim

Well, i hope that the next doc drop will means that the modesetting and basic driver information is at least complete. 2D accel and especially 3D accel aren't even being discussed yet.

[djdoo]

@Luc:

I got your point my friend...
I am sure you know better than me :)

Are you one of the people that develop that open driver said to be developed by Novell??

[libv]

@Luc:

Are you one of the people that develop that open driver said to be developed by Novell??

Yes, SUSE is the linux distribution and service provider that is a daughter company of Novell, and i am indeed one of the developers of this driver :)

[butdie]

I have just tested the latest avivo driver from git and it still has the "screen is corrupt after you close the lid and then reopen it" bug. (The laptop's screen powers down when the lid closes, of course. However, it doesn't seem to reinitialize correctly afterwards.)

I got the same bug also. I use Ctl-Alt-F6 Ctl-Alt-F7 to get screen back.

Another missing feature of the avivo driver is XVideo support for video play back. right now, I have to use mplayer -vo gl2, which is slow

[libv]

I got the same bug also. I use Ctl-Alt-F6 Ctl-Alt-F7 to get screen back.

Another missing feature of the avivo driver is XVideo support for video play back. right now, I have to use mplayer -vo gl2, which is slow

This is no different with our driver yet. We will hook into ACPI correctly soon, when we get the time to do this.

Switching VTs reinitialises all of modesetting, and gives you a correct mode. Lid close events shuts down many things, and the lid open event than reenables everything like the bios thinks everything is being used, but this was of course altered by avivo or our radeonhd driver. Once we capture these events, this problem will be handled by the X server and "It Will Just Work" (famous last words).

[stunted]

What are the chances of the specks enabling us to use the GPU for more general purpose processing? I frequently backup and compress whole harddrives of data, or transcode MPEG2 video captured from my PVR-150 to Xvid, (soon H.264) is it going to be possible to use the GPU for some/which appropriate tasks?

The Folding@Home people only support R500 series GPUs because "The X1K boards were the first to support 32-bit precision and branching/looping on the ATI/AMD side of the world" and they strongly recommend the R580 / X1900 series, as it has 48 pixel shaders, presumably the same will be true of any future GPGPU efforts? What about R600 chips with their Unified Shaders?

What are the chances of H264 playback assistance, or a more general video decompression framework? What about on older (R300) hardware? Presumably this wouldn't require branching/looping and may well enable thousands of older PC's to perform as HD MythTV frontends.

I realise these questions are way ahead of them selves, and there's a LOT of work to do before we get to this point, but I'm just very excited at the prospect.

Thank you to all the developers for your effort on behalf of the FOSS community

http://folding.stanford.edu/FAQ-ATI.html
http://en.wikipedia.org/wiki/GPGPU
http://en.wikipedia.org/wiki/Comparison_of_ATI_Graphics_Processing_Units

[stunted]

I just found www.gpgpu.org and their not mentioning this on the forums there, I guess specifications aren't going to make a big difference to the world of General-Purpose computation on Graphics Processing Units as they already have open APIs to use, though with closed compilers.

I wondered if we might get an open gpgpu framework and compiler, but I suspect it depends on how detailed the rest of the specks are and possibly how capable intels upcoming Larrabee GPUs are as to weather or not the results will be worth someones effort.

[koolmanoncampus]

What are the chances of the specks enabling us to use the GPU for more general purpose processing? I frequently backup and compress whole harddrives of data, or transcode MPEG2 video captured from my PVR-150 to Xvid, (soon H.264) is it going to be possible to use the GPU for some/which appropriate tasks?


I was actually talking to an AMD representative recently, they mentioned this exact functionality in future tools they plan to release.

He mentioned how they want to release tools to allow developers to better harness the GPU's computing power.

[stunted]

I was actually talking to an AMD representative recently, they mentioned this exact functionality in future tools they plan to release.

He mentioned how they want to release tools to allow developers to better harness the GPU's computing power.

That's very exciting, I wonder weather they'll be open source or propriety.

After posting I did some more reading round the gpgpu.org forums and found this thread
http://www.gpgpu.org/forums/viewtopic.php?t=4756
which raises some very interesting questions about the reliability and error rates on GPUs. I don't think for the moment I'd use the GPU for anything like compressing harddrive images, of course a few errors in a video compression probably won't mater but when the integrity of the data is more important than the speed of compression I'll stick with my CPUs and ECC registered RAM.

[koolmanoncampus]

That's very exciting, I wonder weather they'll be open source or propriety.

After posting I did some more reading round the gpgpu.org forums and found this thread
http://www.gpgpu.org/forums/viewtopic.php?t=4756
which raises some very interesting questions about the reliability and error rates on GPUs. I don't think for the moment I'd use the GPU for anything like compressing harddrive images, of course a few errors in a video compression probably won't mater but when the integrity of the data is more important than the speed of compression I'll stick with my CPUs and ECC registered RAM.

I wouldn't know about the license state of such a tool. Hopefully it's along the lines of a gcc toolchain.