Does ATI just not care at all about game developers?

[barkas]

@#1:

You compare the binary Ati driver to the open Intel ones. Why not mention how the open Ati ones run your game?

Having been there, probably better, but the open ones do not support all features of the hardware.

[bridgman]

Are you talking about GL level implemented in the driver or actually "supporting all features of the hardware" as you say ? Most of the work related to higher GL levels does not actually make use of additional hardware capabilities, it's just fancier software.

[Dandel]

@somedev

I don't know about your experience with the opengl bug reports, but the unofficial bug reports at least gets some sort of response quickly.

Anyways, I Have no clue on a few things your talking about.

1) Which game is effected? (Title and link would be helpful so others can confirm this)
2) Which driver Revisions are involved. (This is helpful to the ATI devs so they can help with what appears to be regressions.

3) which radeon Video cards have been tested, and the general system spec on those test machines.

4) Can you provide links to the bug reports you made on the unofficial tracker?

5) have you tried initializing the matrices that are involved on opengl. (Sounds strange, but glsl is buggy as hell on ati cards because the matrix is not initialized)

see this forum post for a bit of info:
http://forum.xbmc.org/showthread.php?t=74572 (patch for glsl on xbmc for single color block video playback)

[bridgman]

Maybe a dumb question, but are barkas and somedev the same person ?

[barkas]

No.

I meant hardware features, in my case volumetric textures didn't work in the open drivers.

[bridgman]

Good point; I don't remember 3D textures even being discussed.

I suspect the issue is that until *very* recently the open source stack didn't have a memory manager capable of using the entire video memory space, so 3D textures would have tended to quickly suck up all of the available VRAM. Now that the memory manager can use VRAM past the BAR limit we should probably revisit volumetric texturing. Thanks for bringing this up.

(for the record, 3D textures could already be supported and I might simply not know about it, but I don't think so )

[bridgman]

Just curious, where are you using volumetric texturing ?

[allquixotic]

Just thought I'd pitch in and say that closed source devs are being bitten by the same issues of working code -> nonworking code, so you are *definitely* not alone, somedev. Take the games Heroes of Newerth and Savage 2 for example. I've already reported that one (as has S2Games staff and many users affected by the bug) but 10.5 and 10.6 are affected. I'll start raising holy Hell again if 10.7 doesn't have that fixed, but before I get *too* critical of ATI, I'll be fair and hold off my nasty-nasty comments (or my elated praise) until 10.7 is released.

The lack of frequent public betas (at least for the Linux version) hurts at least as much as the slow fix turnarounds, though. Since they do parallel development on different feature sets and release weekly candidates to the lucky people who got into the beta program, the rest of the public is left to wait an indeterminate number of weeks until a branch (which may have a needed bugfix) is integrated into the official stable driver.

And don't mention how the beta drivers are often broken and problematic on some hardware. Sure they are. But on the off chance that a weekly candidate fixes an issue I care about, without breaking anything else I care about, it's beneficial to me. (couldn't care less if it breaks XYZ hardware I don't own or XYZ OpenGL extension my games don't call)

Come to think of it, the only hardware-accelerated graphics driver for Linux that does not have public pre-releases -- besides maybe the closed source VIA crap/vaporware -- is the Catalyst driver. Let's see:

-All the Mesa based open source drivers have even better than beta; you can check it with every git commit, every branch, etc. And, in the event that the bug is "fundamental" (does not require specific graphics hardware knowledge to spot what is going wrong), a college-educated programmer who knows C can contribute a fix. Drive-by "oops, I found a NULL pointer" fixes do happen in mesa.

-Nvidia's binary driver publishes public betas appreciably often, both for major feature pushes (as ATI did for GL 4.0) and for minor bugfixes (as they often do leading up towards a new stable release, but ATI doesn't).

-Unless you have a special relationship with ATI, you get (at best) Catalyst updates every 30 days, sometimes a bit more; you only get major feature updates every 3 or 4 months.

Basically, if the drivers are already broken (for things I care about), how is it possibly beneficial to me, as a user, to withhold drivers that might also be broken? There is a chance they may not be broken, and last I checked, one of the goals of perpetually improving software is / should be to get features users need into their hands as quickly as possible.

For binary drivers like this, there is a middle ground between public betas and only stable updates. The middle ground solution is what (ech) Microsoft uses, and it seems to work ok.

That is, deploy stable hotfixes when there are serious regressions that prevent the software from working at all for some users, when it used to work before the update.

The downside of hotfixes is that it adds a huge managerial overhead to ATI... even more than the overhead of releasing your weekly snapshots (which you already make available to testers) to the public. You would have to triage reported problems, analyze their impact, work on a fix, test the fix, and get the fix pushed to the public -- all within the release cycle of a driver. The amount of work required for that can't really justify doing it.

But how hard is it to simply disable the authentication checks for whatever server (HTTP? FTP?) you use to distribute the private betas? It's actually easier for ATI to not employ authentication than to employ it. As Stallman frequently says, this is an anti-feature: you are making people do something exceptional to get you to not do something that users find annoying. And you expend substantial resources on the problem of ensuring that only the users you select are able to access the private betas.

Again, a decent compromise would be to:

1) Implement automated acceptance into the "private" beta program, effectively making it a case of "you can get these betas free and right now, but you just have to go through our registration form first". This still doesn't remove the anti-feature, but at least it will satisfy 99% of the people.

2) In the above situation: make users agree in a terms of service that they will not ask for support with beta drivers, and provide a URL to a place where they can file bug reports (not support requests) if they find regressions. You can even refer them to the unofficial Catalyst Linux bugtracker that you say has been active lately.

I'm sure there's some lame "PR" or "Intellectual Property" reason why you can't release public betas as I propose above, or else you probably would have done it already. This is good, in a way: it disproves people who think that ATI and Nvidia are controlled by a single shadow entity a la Deus Ex. Those silly crazies. It's competitive differentiation, if you will. Or in other words, another reason to go Nvidia.

Disclosure: I was an Nvidia user for many years, but decided to try out the first high-end hardware tessellation card to market, the HD5970. Oops. I'm seriously regretting that nowadays...

[Kano]

You are not the first with the idea of public betas. But still nothing changed... Does anybody at ati think nvidia would test those drivers for new features? Why on earth is a NDA needed just to test a few drivers? Most likely ati wants to hide the bugfixed future drivers from their paying customers because there are no bugfixes in there

[monraaf]

in my knowledge a public beta program is planet after catalyst 10-7 is released.

but yes maybe its a 'fake' but we will see the true after 10-7 .-

Do you by any chance know if they have finally fixed XvBA on Evergeen cards?