r300/500 rant

[rohcQaH]

1. both
2. not yet IIRC two things are missing
2.1: a solid formula to determine the minimum clocks required for the current screen resolution. Only when the hard limit is known, aggressive downclocking is safe.
2.2: a good logic to determine when to change clock speeds. There have been some experimental patches for dynamic clock-switching, but I'm not sure if anything went mainline yet.


KMS-based power management has some pretty sweet options, like synchronizing CPU + GPU sleep to save even more power. In theory, with enough work, it could beat powerplay. But it's not there yet.


/edit: oh, ninja'd.

[DirtyHairy]

The man page usually matches the driver pretty well, so it's likely that your driver code is too old to have the newer power saving features.

Ah, but see, this is _exactly_ why I was using the word "stable". I am on xf86-video-ati 6.12.4 which is the most recent released version. Same goes for comments along the line "KMS will eventually improve the situation" - the corresponding code is only partially written yet, lest alone well-tested and at least in beta state. I use this machine for work and, while I have done so in the past on several occasions, can't afford the time and risk of running it on highly unstable software anymore.

[DF5JT]

By the time they get their 'crap in order' your laptop will long be obsolete. Need a proper running performance driver in the mean time? All you will get is a 'we hear you, we understand but sucks to be you' reply back.

Nonsense. My T60p has a FireGL V5200 and I have inhouse service for this piece until March 2012.

I can safely assume that my laptop won't be obsolete until then, unless you'll find me a 4:3 with a non-glossy UXGA display and a keyboard of similiar quality.

You have probably never had a decent laptop to begin with.

The development of the drivers for this card has made enormous progress on all fronts and I trust the developers to continue their great work. It's not their problem that ATI has taken the support from r3oo cards.

These days a modern distro with up to date drm, xf86-drivers and mesa does all the things I expect it to do, except for power management that is, but that is coming along nicely thanks to some recent patches.

[pingufunkybeat]

I feel that AMD/ATi is making phenomenal progress on the free drivers.

Unfortunately, they are paying for many years of neglect, which left the Linux support for ATi cards in a miserable state. They have a lot of catching up to do. And due to the big changes going on in the X/gfx stack, the transition is not very smooth at the moment.

Still, they are doing the RIGHT thing -- working on free drivers which will not only benefit the users of ATi cards who want to run a free operating system, but also benefit all Linux users through the improved underlying infrastructure.

And this is why I bought an AMD card instead of an nVidia one. If I wanted to replace half of the Linux ecosystem (much of the kernel, half of X, everything relating to graphics) with a binary blob developed for Windows, I'd stick to a closed source operating system. I can run bash and Vim on top of Windows anyway.

[DirtyHairy]

I feel that AMD/ATi is making phenomenal progress on the free drivers.

Well, I guess next time my battery runs out on the train without a socket nearby, I will try to bring myself in the mood of feeling the good KMS / powermanagment vibes coming nearer every month I'm waiting.

And this is why I bought an AMD card instead of an nVidia one. If I wanted to replace half of the Linux ecosystem (much of the kernel, half of X, everything relating to graphics) with a binary blob developed for Windows, I'd stick to a closed source operating system. I can run bash and Vim on top of Windows anyway.

You're not serious about a binary driver replacing "much of the kernel"? That's utter crap, most of the duties of the kernel are not related to graphics in any way. In addition, the "linux ecosystem" as such doesn't exist, there's the kernel, then there's GNU, and in addition, we have X (and mark, all of these are independent components which can well live without each other). Therefore, a closed-source driver is not "replacing half of the linux ecosystem" but just supplying the necessary functionality to drive the hardware in question to a well defined, small subset of the system you are using.

Don't get me wrong, I am very much an advocat of open source, but the mere existence closed source software doesn't kill puppies, and I prefer a closed source driver which operates my hardware as intended over an open source one which may have the necessary functionality at some point in the future. If they would have supported "legacy" hardware in fglrx long enough for essential functionality to be replaced by the open source driver, I would have been the first to applaud ATI for its move, but as things stand, customers have been dropped and the task of reimplementing the lost functionality has been left to "the community" (although, as ATI is backing this development with manpower, this statement is a little bit too harsh).

[rohcQaH]

I don't see where you're going here. You don't want experimental software on your computer, but on the other hand refuse to use the tried and working (and supported) packages where fglrx still works. Uhh, what?

Which of the kernel features from 2.6.29+ do you absolutely need? Which of the features from Xorg 7.5?

guess what both my gentoo (stable) systems say?

~> uname -a
Linux tubbook 2.6.28-gentoo-r5 #1 SMP Sun Apr 26 20:43:05 CEST 2009 x86_64 Intel
~> emerge -pv x11-base/xorg-x11
[ebuild R ] x11-base/xorg-x11-7.4-r1 0 kB

so it would still run 9.3, even though one's got an nvidia GPU and one a HD5770!

Maybe I was a bit lazy with the kernel upgrades, but it really shouldn't be a problem to echo '>=gentoo-sources-2.6.29' >>/etc/portage/package.mask and keep using fglrx for a while longer.



The decision to drop fglrx-support was made by the windows-guys, so the linux guys had two options:

a) invest additional work to maintain the older fglrx-versions a bit longer (without help from the windows driver team)
b) invest additional work to make the OS drivers appear faster.

They choose b), and I'm very happy that they did.

[pingufunkybeat]

Don't get me wrong, I am very much an advocat of open source, but the mere existence closed source software doesn't kill puppies

I completely agree with you, and I feel the same way when it comes to running a specialised closed-source application on a Free OS.

But when the closed source software allocates memory, manages hardware, reprograms the clocks and directly affects the ability of each and every Linux program to draw a pixel, then it's very close to killing puppies

And for this reason I appreciate the open source efforts. True, dynamic powersaving (the static one has been available forever!) is taking its time, but this is also due to having to wait for the KMS infrastructure to mature. It's unfortunate, but if you're willing to get your hands dirty a bit and build from source, you'll be good to go.

[pingufunkybeat]

Therefore, a closed-source driver is not "replacing half of the linux ecosystem" but just supplying the necessary functionality to drive the hardware in question to a well defined, small subset of the system you are using.

It's replacing the bottom half of X, all of Mesa and the parts of the kernel which deal with memory management and the GPU (TTM/GEM, KMS, etc.)

So instead of running an open-source X server, and open-source OpenGL implementation and an open-source in-kernel memory manager, you are running a big fat blob.

Surely you see why many of us feel that this is very different from running Doom3 with user privileges on a Free OS?

[pedepy]

the whole point of my rant was that i don't understand that *any* development time is spent on bringing open source support to anything that currently is still undet the scope of fglrx; either you guys drop fglrx altogheter or you quit duplicating efforts and get all of your open source developers to work on hardware which has been "deprecated".

[DanL]

You're only looking at it from your point of view and not ATI's. They're not duplicating effort. A lot of fglrx is code from the Windows Catalyst driver and some of ATI's customers need that driver.