"Ask ATI" dev thread

[ThatAlienwareGuy]

I'm running Kubuntu 9.10 with Catalyst 9.10 drivers on my laptop with a radeon hd 3870 (x1).

Works great, except now my laptop now has a black screen and the only way I can use my laptop is by plugging it into an external monitor.

More info here:

http://ubuntuforums.org/showthread.php?t=1315923

[Riotta]

Don't know if that was already but I would like to ask Catalyst devs is people in the dev team personally Use Linux on desktop, what distributions mainly and what they are thinking about the system itself.

[d2kx]

Don't know if that was already but I would like to ask Catalyst devs is people in the dev team personally Use Linux on desktop, what distributions mainly and what they are thinking about the system itself.

I'd assume they're waiting for the opensource drivers to mature because the fglrx is in a horrible state right now.

...

Come on, at least imagine it for a second

[mitch074]

Now for a new question, a real problem

I have a Sapphire RadeonHD 4850 (ont of the first, not tweaked, made-from-specs models); having managed to flash it with a VBIOS that allowed me not to cook eggs on it (90?C on moderate use was rather bad), I had to change the cooler+fan for an aftermarket one (previous one didn't like having to actually work: fan got busted). Silent operation, coolness (less than 50?C on Furmark - under Wine) but... Problem.

The last driver release to work with it is Catalyst 9.8; 9.9 and 9.10 compile, install and load (I can manually load the fglrx kernel module), but as soon as X loads, the system hangs hard: SysRq keys don't work, all disk operations are suspended - I can't get logs. This happens with whatever version of the kernel, glibc, xorg, even distro I try.

Free drivers work rather well with it (both radeon and radeonhd); however, I'd like some 3D support.

[energyman]

well, start with flashing the original bios.

[mitch074]

well, start with flashing the original bios.

I tried. Yes, we have no banana.

[bridgman]

I have a Sapphire RadeonHD 4850 (ont of the first, not tweaked, made-from-specs models); having managed to flash it with a VBIOS that allowed me not to cook eggs on it (90?C on moderate use was rather bad), I had to change the cooler+fan for an aftermarket one (previous one didn't like having to actually work: fan got busted). Silent operation, coolness (less than 50?C on Furmark - under Wine) but... Problem.

The last driver release to work with it is Catalyst 9.8; 9.9 and 9.10 compile, install and load (I can manually load the fglrx kernel module), but as soon as X loads, the system hangs hard: SysRq keys don't work, all disk operations are suspended - I can't get logs. This happens with whatever version of the kernel, glibc, xorg, even distro I try.

Free drivers work rather well with it (both radeon and radeonhd); however, I'd like some 3D support.

One option would be to use the open source 3D driver. It wasn't enabled by default in this round of distros because of Compiz corruption issues that were only fixed a few days before the distros locked down, but it's working well for a number of users.

[mitch074]

One option would be to use the open source 3D driver. It wasn't enabled by default in this round of distros because of Compiz corruption issues that were only fixed a few days before the distros locked down, but it's working well for a number of users.

Well, yes, I know - but I also know that currently, in Free drivers, 3D support is 'limited' to OpenGL 1.5 (at best) with a not exactly stellar performance in dynamic shaders compilation (LLVM is good, but it's not quite there yet, and a lot of work is being directed to Gallium - not Mesa, which is still the default stack). A shame if you consider that R700 chips are OpenGL 3.x capable.

And, dare I admit it? I essentially need 3D support to play WoW on Wine (in OpenGL mode, which works rather well with Catalyst 9.8, with no graphics glitches). Otherwise yes, I'd gladly do away with fglrx and use radeon or radeonhd full time (2D, Xv and compositing performance are ways above fglrx's).

By the same token, it'd be great if, with Gallium, there was a way to use Free/OSS 2D drivers and either proprietary or Free 3D stack; as far as I know, Gallium should allow that. Has it been discussed? What would block it?

[bridgman]

The proprietary 3D stack uses a lot of internal interfaces which are different from those in the open source stack. They wouldn't be able to interoperate without a significant rewrite of the proprietary stack -- which wouldn't be a good idea because the current internal interfaces are chosen to facilitate use on a number of different OSes, not just the X/DRI environment.

re: getting the new Catalyst drivers running, let's start with the obvious questions :

- did you fully uninstall the previous driver versions (using the uninstall script that gets installed with the driver) before installing a newer version ?

- did you run appropriate aticonfig commands before starting X the first time after a new installation ?

Reflashing bioses is always potentially problematic - did you just flash in the BIOS image from another card or edid specific settings on your current BIOS image ? If you flashed in the BIOS from another card then all bets are off, and I really would suggest going back to the original BIOS image. Not sure where the banana comes in.

BTW the devs have pretty much given up on using LLVM in the drivers, although it is still considered a good choice for the softpipe (shaders in software) implementation and may work its way into other parts of the compiler stack.

[mitch074]

The proprietary 3D stack uses a lot of internal interfaces which are different from those in the open source stack. They wouldn't be able to interoperate without a significant rewrite of the proprietary stack -- which wouldn't be a good idea because the current internal interfaces are chosen to facilitate use on a number of different OSes, not just the X/DRI environment.

re: getting the new Catalyst drivers running, let's start with the obvious questions :

- did you fully uninstall the previous driver versions (using the uninstall script that gets installed with the driver) before installing a newer version ?

- did you run appropriate aticonfig commands before starting X the first time after a new installation ?

Reflashing bioses is always potentially problematic - did you just flash in the BIOS image from another card or edid specific settings on your current BIOS image ? If you flashed in the BIOS from another card then all bets are off, and I really would suggest going back to the original BIOS image. Not sure where the banana comes in.

BTW the devs have pretty much given up on using LLVM in the drivers, although it is still considered a good choice for the softpipe (shaders in software) implementation and may work its way into other parts of the compiler stack.

About the obvious questions:
- yes, I uninstalled all drivers properly, using uninstall scripts then manually hunting down any stray file that might have been left behind. I also deleted xorg.conf to regenerate it anew using appropriate tools (aticonfig --initial) on each attempt.

- I reflashed my card using its original VBIOS file, which I had backed up and checksummed previous to flashing it the first time. I also compared the checksum to known original Sapphire VBIOS files, and they matched (just to be sure my card hadn't been customized before I started tinkering with it - backups, backups... I might be a tinkerer, but I'm also a prudent one).

My system is currently running Mandriva Free 2010.0 64-bit. Other distros, other kernels (even a self-compiled vanilla 2.6.31 kernel) exhibited the exact same behavior.

Meaning that, even when restoring my card to its factory condition (minus the fan, which really is busted), I get complete system locks on Catalyst releases more recent than 9.8. I also (since I encountered the case on other OSes and on the FOSS drivers) tried switching to the card's secondary output, using a DVI to D-SUB converter: no luck (in fact, switching to the secondary output would hang even FOSS drivers, in the same fashion, with or without original VBIOS).

I don't think my card is that special; even the Sapphire sticker is merely put on top of the original Ati sample implementation's one. Could it be that the driver hangs because it's expecting data from the fan, but not finding any, gets weird? If that's the case, why didn't it do that with 9.8?

Thank you for your time - I didn't know LLVM had been abandoned (I'll look it up in other, more appropriate sections of the forum). I hope that maybe, with 9.11, it'll be fixed. Otherwise, maybe next release...?