AMD's Legacy Driver Will Not Support X Server 1.6

[bridgman]

The 780 and up should be decently fast - low end gaming only, of course, but fast enough for everything else. The problem with most IGP products is that the graphics requirements are usually "the very best graphics possible for no more than a couple of dollars" ;(

We would probably sell a lot more 3450s if we didn't have the same graphics in a 780

[psycho_driver]

More than nVidia ever did for us; Quit yer yappin'

I beg to differ. nVidia has consistantly provided quality binary drivers for many, many years. They also continue to update legacy drivers to ensure that users with old (oooold) hardware can use new software.

I cannot understand people who continually try to stand up for ATi when they really haven't done a whole lot for the linux community.

They're provided crappy binary drivers and they've released just enough documentation to make subpar open source drivers for their hardware that won't allow it to be used to anywhere near full potential.

I'm sure I come across as an nvidia fanboy on this site because I'm always singing their praises and knocking ATi. In truth I'm just a long-time linux user who has used a variety of cards from both brands (or tried to, in ATi's case). If I was a windows user I might very well advocate ATi cards.

I should also mention I completely gave up on ATi about a year ago. Maybe their drivers have improved since then. At that time, however, there was absolutely no, none, zero, zilch, comparison to be made between nVidia and ATi. Anyone advocating ATi over nVidia for linux gaming or multimedia should, quite honestly, be ashamed of themselves.

[bridgman]

...and they've released just enough documentation to make subpar open source drivers for their hardware that won't allow it to be used to anywhere near full potential.

I'm sorry, but that is just patently false. We have released enough technical information to let an equally sophisticated driver stack run 100% as fast as fglrx or the Windows driver (barring OS differences). The point is that writing an equally sophisticated 3D stack to match fglrx would require more developer effort than the entire Xorg community combined, and none of the developers I have spoken with feel that level of effort is necessary.

A clean, well-written driver of perhaps 1/10th the size and complexity of fglrx (say a straightforward Gallium3D driver, running over a decent memory manager and a carefully tweaked command submission backend) should be able to average 60-70% of fglrx performance across a broad range of apps, with 5-10% of the developer effort. That is what I expect will get implemented, but it's just a guess.

If you want to get 20 or 30 full time developers together for a couple of years and aim for 100% of fglrx performance, the programming information you need is available and we will cheerfully support your efforts. I'm not sure if it would be practical to start with Mesa+Gallium or if you would be better off starting from scratch, but we should know in the next six months.

I should also mention I completely gave up on ATi about a year ago. Maybe their drivers have improved since then. At that time, however, there was absolutely no, none, zero, zilch, comparison to be made between nVidia and ATi. Anyone advocating ATi over nVidia for linux gaming or multimedia should, quite honestly, be ashamed of themselves.

Yeah, that was about the time we started putting the first consumer-related improvements into the driver.

[etnlWings]

I posted this in the, "AMD Dropping R300-R500 Support", thread but I figure I'll cast a wide net and repost it here:

I don't use fglrx for my Radeon 9550 but I've found that I have to install it, if only briefly, to get any acceleration with any WINE-based products.

Once I've installed fglrx, I can then remove it and go back to the xorg-video-ati driver, retaining acceleration in WINE/Cedega but if I can't even install it, come Ubuntu 9.04, the usefulness of my PC is going to be severely hampered.

Anyone else experienced this bug and know a way around it that doesn't involve installing and uninstalling fglrx?

[bridgman]

This is interesting. Do you fully uninstall fglrx before switching back to the open driver, or do you think you are leaving one of the files that fglrx over-writes in place ?

Does the effect survive a reboot ? In other words, do we think this is from fglrx writing registers or changing files ?

[Wyatt]

I beg to differ. nVidia has consistantly provided quality binary drivers for many, many years.

Personally, I'm going to have to beg to differ too. The other way. fglrx has problems; I don't think anyone's going to deny that. To be sure, I had a devil of a time getting it working the first time. nvidia wins that one. But you know, I've had several orders of magnitude more issues with nvidia's terrible driver than ati's terrible driver.

It's one thing to install easily; it's an entirely different matter to be rock-stable for six months at a time. How often should I get an oops because my display driver inexplicably decided that it was a great time to deref NULL while I was taking a simple note in my text editor? It's never happened with fglrx. How often should an app start creating strange green corruptions all over the screen and then kernel oops? So much for "good legacy driver support." Why has performance in nvidia's driver been getting worse with this 6600 since the 180-series drivers hit? Why did it stop properly controlling the fan? Or the recent drivers that seemed to enjoy ignoring signals: I didn't even think drivers could cause that sort of problem! And don't even get me started on the ugly, ugly hack that is Twinview.

There may have been a time where your synopsis was correct. That time is no longer the present.

If we can't defend ATi and AMD for teaching us to fish, please don't defend nVidia for giving us a rotten cod.

[etnlWings]

This is interesting. Do you fully uninstall fglrx before switching back to the open driver, or do you think you are leaving one of the files that fglrx over-writes in place ?

Does the effect survive a reboot ? In other words, do we think this is from fglrx writing registers or changing files ?

I install fglrx, Catalyst and the one or two other packages fglrx demands as dependencies via apt and I apt-get purge them out of there. I also restore my pre-aticonf'd xorg.conf and have tried reinstalling the xorg-video-ati package just for the hell of it.

In the past I've usually left fglrx installed and in-use until something goes wrong (with 8.10 I think I left fglrx installed for about a month before I experimented with a dual-head setup and Catalyst refused to give me my native resolution back after I gave up on my dual-monitor aspirations).

So far I've found no other way to get WINE/Cedega/etc working properly (I use that term loosely) and I've found no way to re-break it, either. Going back to the open-source driver incurs a slight performance drop/few missing graphic operations but OpenGL/Direct3D in WINE retains the acceleration I'm unable to get on a fresh, pre-fglrx'ing install.