AMD Posts A Horde Of New 3D GPU Documentation | Hacker News Reader

AMD Posts A Horde Of New 3D GPU Documentation

phoronix.com

121 pointssamwilliams36 comments

Open articleView on HN

Direct links from http://www.botchco.com/agd5f/?p=58:

AMD R6xx/R7xx 3D Register Reference: http://www.x.org/docs/AMD/R6xx_3D_Registers.pdf

AMD R6xx/R7xx Acceleration: http://www.x.org/docs/AMD/R6xx_R7xx_3D.pdf

AMD Evergreen 3D Register Reference: http://www.x.org/docs/AMD/evergreen_3D_registers_v2.pdf

AMD Cayman/Trinity 3D Register Reference: http://www.x.org/docs/AMD/cayman_3D_registers_v2.pdf

AMD Evergreen/Northern Islands Acceleration: http://www.x.org/docs/AMD/evergreen_cayman_programming_guide...

AMD Southern Islands 3D Register Reference: http://www.x.org/docs/AMD/SI_3D_registers.pdf

AMD Sea Islands 3D Register Reference: http://www.x.org/docs/AMD/CIK_3D_registers_v2.pdf

AMD Southern Islands/Sea Islands Acceleration: http://www.x.org/docs/AMD/si_programming_guide_v2.pdf

AMD HDA audio verbs: http://www.x.org/docs/AMD/AMD_HDA_verbs_v2.pdf

I bought an ATI 5870 a few years ago. At the time, it was a pretty beefy card, and drives my three 23" monitors really well... in Windows.

Late 2012 is when I found that enough bugs were fixed in the [now AMD] proprietary drivers to where I could get serious work done in Linux. The crashes were mostly fixed, but there are still lots of performance issues, as well as a lot of fun bugs (like my cursor randomly turning into a 20x50 block of randomly colored pixels when moving between monitors).

It's been a dreadful experience. The AMD driver releases stopped coming with changelogs for about 6 months, and they still don't always release them (or when they do, they leave a lot out). When a new version of Xorg comes out, it takes them at least two months to add support (while nvidia often has support ready before release, or very shortly after). nvidia, even with their own issues, shows them up enough to make me envious. Intel is really taking things to a higher level with their drivers + OpenGL performance lately, too.

If you have to use proprietary drivers (like I do for 3D/OpenGL), stay the heck away from AMD. They do an awful job with their Linux drivers. Linus likes to bitch at nvidia, but their binary Linux drivers are leagues better than AMD's. They've got issues, too, but are by far the lesser of two evils for higher end GPUs + Linux.

Just use the in-kernel free drivers. They have improved immensely the past few years. The Radeon HD 5870 is called 'Evergreen', you can check out the feature support here:

http://xorg.freedesktop.org/wiki/RadeonFeature

In my experience the free drivers have slightly slower 3d performance and much faster 2d performance.

That matrix is impressively green! Thanks for sharing.

As mentioned in the initial comment, this isn't an option for me.

I don't see that mentioned.

Still hit and miss though. Bought a laptop that even had somewhat decent 3d acceleration, but would horribly lag on 2d terminal output.

Returned it and been using Intel and Nvidia exclusively now - no problems whatsoever.

There are some things that can be done to fix that -

https://wiki.archlinux.org/index.php/Radeon#Slow_performance...

I tinkered around for a few hours and gave up. Not worth my time.

The reason I bought an ATI card _is_ that it had open-source drivers. Is there any reason you "have" to use the proprietary driver? (besides better performance for games, which I assume is not what you are after since you mentioned getting serious work done...)

I spend 99% of my time in Linux, so there is some gaming there as well, especially now that there are a lot of great titles with support on Steam. I rarely ever boot into Windows anymore, mostly for games that I can't play on Linux (or can't play well on Linux).

If I wanted something for work only and wanted to use open source drivers, I'd buy Intel or nvidia now. The ATI 5870 exists to drive up to six monitors over displayport under EyeFinity, so this was as much of a work investment as it was a play investment (it's great for both).

I want to be able to work AND play with this card that I paid good money for. The binary drivers are officially supported, and with that label comes some expectation of quality. Perhaps my expectations are unreasonable, but I've been very disappointed with the quality.

> The ATI 5870 exists to drive up to six monitors over displayport under EyeFinity

Are you saying the open source driver cannot drive size monitors?

It does a pretty poor job with multiple monitors + OpenGL. Lots of pixels to push. I didn't try to get EyeFinity working, but that's probably out of the question as well.

Maybe they're doing CAD, or other 3D modelling of some kind?

Why are you using the proprietary drivers? Is it just the lingering "got to run around finding drivers to install" windows mindset?

If he's doing serious work, then he's probably using OpenCL, and as far as I am aware, the Gallium drivers do not yet fully support OpenCL. When I checked earlier this year, they were getting close, but not yet usable. Has this changed?

Can I finally get rid of the proprietary ATI drivers and still run OpenCL reliably?

"If he's doing serious work, then he's probably using OpenCL"

He doesn't indicate that at all, and I assumed if he were doing "Serious OpenCL work" he'd be using a later-generation card.

5870 has a theoretical peak of 2.7 tflops, a GTX680 has a theoretical peak of 3.1 tflops. even a 7970 has a peak of 3.8 tflops.

you can argue (successfully) that it's much easier to achieve a high proportion of peak on GCN cards, but CL code tuned for a 5870s isn't going to be exactly sluggish.

I get the a similar bug where the cursor gets mangled when swapping monitors on Windows also.

wow, that stinks. I've never ran into it on my Windows 7 partition. I think I'd be even more irate if I did, though. Hate to pay top dollar for a high end card, only to see stuff like that on the platform is was more or less specifically targeted at.

> as well as a lot of fun bugs (like my cursor randomly turning into a 20x50 block of randomly colored pixels when moving between monitors).

This bug also occurs occasionally on Windows with the ATI drivers, it is not a Linux specific bug.

I have the same opinion on ATI, but on windows. I use radeon hd 6630M (switchable graphics), with drivers from 2011, because they never bothered to release an update.

Effect? I'm getting artifacts/BSOD when try to play games. Fortunately I can switch to intel.

> I have the same opinion on ATI, but on windows. I use radeon hd 6630M (switchable graphics), with drivers from 2011, because they never bothered to release an update.

Careful about tagging AMD on that. Mobile manufacturers often repackage the drivers and sometimes mess with the device IDs so you're stuck with their release of the drivers.

If it's an entirely stock chipset, any modern Catalyst driver will be able to talk to that card.

I wonder how much of this is "business as usual" for AMD's GPU team, and how much of it is in reaction to nVidia recently beginning to open up? My limited understanding is that AMD/ATI has been far more more open than nVidia for much longer, but if they see nVidia starting to open up, maybe they're lengthening their stride on these efforts?

Either way, my inner optimist/FOSS fanboy is really hoping that there's a convergence of things happening right now (championed by Valve, I think) that will make GPUs become as open and well-documented as main processors. It's rare to see, but I get all excited any time openness is even a minor basis for competition.

I'd assume "business as usual"; preparing a publication like this takes far, far longer than the time delta to nVidia's announcement allows for.

I think the lead-time issue depends on how they document internally. If they provide 'first-class' documentation for their partners to consume which has already been sanitized of NDA-restricted disclosures, maybe all they need to do is remove the 'CONFIDENTIAL' header.

Though to be fair, I'd be shocked if this was the case. Most companies don't do the IP sanitization bit, which is why NDAs in those types of industries are viral in the same way as the GPL.

> If they provide 'first-class' documentation for their partners to consume which has already been sanitized of NDA-restricted disclosures, maybe all they need to do is remove the 'CONFIDENTIAL' header.

What partners? The only ones writing AMD GPU drivers is AMD, and internal documentation tends to universally be crap.

AMD also supplies GPUs to all major console manufacturers.

> What partners? The only ones writing AMD GPU drivers is AMD, and internal documentation tends to universally be crap.

Don't quote me on this, but it was my impression that vendors AMD's GPUs should have source access to the drivers.

Have you ever tried to get something released as OSS in a large company? Sun was a prime example - inspite of complete buy in from their CEO it took them eons to open source parts of Solaris. And I dont think till the very end they were able to get a completely bootstrapping release of solaris out.

It may have happened for the same reason, but one can not be a reaction to the other.

People are speculating if NVidia started oppening their specs by Valve's request. If so, AMD could have received the same request.

At this time last year, AMD was preparing for a lay off in November (2012) in which they let a bunch of their open source GPU devs go. [1][2] [3] My projection is that this is an attempt to draw in the unpaid open source community. Both a smart and a dick move: they could have released this info while maintaining the paid positions, IMO. [1] "Did AMD shoot itself in the foot by laying off open-source talent? " http://www.networkworld.com/news/2012/110712-amd-open-source...

[2] AMD shutters German Linux lab, gives devs the axe http://www.theregister.co.uk/2012/11/07/amd_closes_german_os...

[3] AMD dismisses numerous open source developers http://www.h-online.com/open/news/item/AMD-dismisses-numerou...

A hoard, surely, rather than a horde.

Hacker News readers are a horde. The newly-released documentation is a hoard.

Assuming this is correct: http://grammarist.com/spelling/hoard-horde/

Anyone else disappointed that the GPU documentation isn't somehow rendered in 3D?

No.

Thanks Valve.

Reply on news.ycombinator.com