AMD Catalyst 9.7 For Linux Released

[skywarp04]

I just want support for my $200 video card

Posted by bridgman
The fglrx driver is primarily written for the professional workstation market (CAD etc..), which is the main market for Linux graphics right now. It is unlikely that the open source drivers will ever be able to replace fglrx in that segment.

I'm so tired of hearing this. Does that change the fact that I paid $200 for a video card that is not fully supported by the company that made it? You don't hear this kind of lame statement from your competition. I have products from AMD and Nvidia in my home and I can testify to the support of both companies. AMD and Nvidia are on even ground when it comes to Windows in my opinion even though there are still tons of people who claim otherwise. But when it comes to Linux Nvidia mops the floor with AMD. I don't hear Nvidia claiming that they only support their workstation clients. There Linux driver isn't called quadro. So what you are telling me is that I should just use Windows. Oh, there's the open source drivers you say. Have you tried them lately? Apparently not because if you had you would know that they are still not that usable for my hardware. Unless the latest version is being used with the 2.6.30 kernel the Radeon 3850 doesn't have any 3D acceleration. That's not happening until Ubuntu 9.10 is released.

Posted by bridgman
Jaunty also shipped with the open source drivers by default. If they do everything you want then I think everyone would agree that staying with the default drivers is your best bet.

Obviously they don't do what I want, otherwise I wouldn't be trying to use fglrx. The version of the radeon driver that ships with Jaunty only provides 2D acceleration for my card. Why shouldn't I have be able to access my $200 video cards 3D capabilities? I constantly can't do compositing or play simple games because fglrx is too damn flaky. I've seen you mention using XRender before for doing compositing because compositing doesn't require full OpenGL acceleration. Have you ever tried using XRender? It's slow as molasses and doesn't always work right.

Posted by bridgman
The purpose of the ati.cchtml.com tracker is to collect and organize enough information that a developer will be able to reproduce the problem in house. For the problem you are describing we would need more information than just "3850" and "Kubuntu", of course, eg which applications we should be running to make the problem appear.

Yes, it does seem to be a memory leak. As far as what applications I use, it doesn't matter. It happens with different programs all the time. Firefox, Konsole, VirtualBox, Amarok, whatever. I'll give you information like my motherboard model, CPU, Ram, whatever all day long if I know someone is actually going to do something with it. But like you said fglrx is for CAD workstations, average consumers don't matter to AMD. Unless you can tell me someone is actually going to investigate something I'm not going to waste my time putting together my specs for a company who doesn't care unless I'm using CAD.

I hate that you have to take the brunt of everyone's anger here in this forum and I commend you for keeping your cool as well as you do with some of the posts I've seen directed towards you. I wish you would drop the "we only support CAD users" defense though. I hope AMD pays you well for monitoring this forum.

[bridgman]

Because you're ignoring a distribution with many ati users? Fedora / RH will always push the opensource (and patent unencumbered) solution -- but it doesn't mean that magically it suits all users (that's why there's RPM fusion etc.). To put it in perspective, there are an estimate ~16 million fedora machines (albeit the vast majority are too smart to have an ATI card).

With respect, if you add the installed base estimates from all the major OSes, then compare the total with the estimated number of PCs in the world, something seems seriously out of whack. Most estimates indicate that roughly 1% of the billion PCs in the world are running Linux, or 10 million total. You can find higher estimates but if you dig a bit you find the numbers are things like "PCs sold without an OS, and we assume the user will run Linux rather than pirating Windows".

I don't think anyone knows what the real numbers are, but please understand that there is a huge range of estimates out there and the numbers you are citing sit out near the edge of the range. Also note that Fedora is probably the most "different" from upstream kernel & X server of any distro out there, so the idea that supporting Fedora would magically bring support for other distros is a bit of a myth. I do think you will see faster support for other distros over time, but we do want to make sure that we have a solid solution in place for the current range of distros first.

The amount of bad will this is generating is astounding, all for something you're going to add in a couple month anyway. (and is seriously not that big of a deal to add)

Two years ago everyone told us that writing drivers was easy and that if only we would open up the hardware specs then the community would write better drivers in no time. A year later that shifted to loud complaints that *we* weren't spending enough money writing the open source drivers ourselves, and what I'm hearing today is "we demand proprietary drivers, writing open source drivers is too hard and the open source drivers don't do enough". Despite all that, you are still telling us that adding features and support to the proprietary drivers is "easy" and that we're obviously stupid for not doing it.

The choice is pretty simple -- we can spread our efforts out and attemt to support all distros equally but make slower progress overall, or focus on a representative subset of distros and make progress more quickly, *then* focus on improving / speeding up support for the rest. The first approach means our overall quality of Linux support will be lower but we won't piss anyone off; the second approach means we can offer a good Linux user experience more quickly but will some distro users will feel insulted by our choices.

Considering the quality of all the previous fglrx releases, I don't see why fglrx is reduced to bare-maintenance releases and all effort put into a solution that doesn't suck (radeon drivers) including adding all the CAD stuff that you say opensource will never support ;D.

Simple. The workstation market is very fast moving and competitive, but it also represents the largest share of the Linux graphics market, so stopping is not an option. The whole point of fglrx is to give Linux users access to the proprietary code we share across all OSes, which we can't put in the open source drivers for a number of reasons.

[bridgman]

I'm so tired of hearing this. Does that change the fact that I paid $200 for a video card that is not fully supported by the company that made it?

You're taking my words out of context. I was responding to your comment that "Fglrx is a mess. It needs to be thrown to the bottom a scrap heap as a failure." - I said that we need fglrx for the workstation market, but you're responding as if I said "we don't care about you, only about the workstation market".

You don't hear this kind of lame statement from your competition.

Tell them that they should scrap their driver and see what they say

I have products from AMD and Nvidia in my home and I can testify to the support of both companies. AMD and Nvidia are on even ground when it comes to Windows in my opinion even though there are still tons of people who claim otherwise. But when it comes to Linux Nvidia mops the floor with AMD.

Yep, and what we've been doing is bringing that same core code to Linux users and then knocking off the Linux-specific issues that result. I know you would like all the work to be finished in a month, but it doesn't work that way.

I don't hear Nvidia claiming that they only support their workstation clients.

You don't hear us claiming it either

There Linux driver isn't called quadro. So what you are telling me is that I should just use Windows. Oh, there's the open source drivers you say. Have you tried them lately? Apparently not because if you had you would know that they are still not that usable for my hardware. Unless the latest version is being used with the 2.6.30 kernel the Radeon 3850 doesn't have any 3D acceleration. That's not happening until Ubuntu 9.10 is released.

I actually use the open source drivers all the time on rv570, rv620 and rv770 so I think I have a pretty good idea what they do

2.6.30 is for 2D acceleration, but that code was backported into the Ubuntu kernel for 9.04. Kernel support for 3D will probably go into 2.6.32.

The main thing gating 3D driver availability is writing the %^$&%@! driver. We released programming info and sample code at the end of 2008 but it turned out that all of the community developers were still working on other important projects, so it's really just been our guys writing the 3D driver. It's making pretty good progress though; there's still a "mystery problem" with textures and some intermittent corruption that looks like something we don't understand about the new radeon-rewrite code, but other than that it seems to be coming together pretty well.

Obviously they don't do what I want, otherwise I wouldn't be trying to use fglrx. The version of the radeon driver that ships with Jaunty only provides 2D acceleration for my card. Why shouldn't I have be able to access my $200 video cards 3D capabilities? I constantly can't do compositing or play simple games because fglrx is too damn flaky. I've seen you mention using XRender before for doing compositing because compositing doesn't require full OpenGL acceleration. Have you ever tried using XRender? It's slow as molasses and doesn't always work right.

It worked pretty well for me, although I was mostly using Metacity. I can try it again on the weekend if you like.

Yes, it does seem to be a memory leak. As far as what applications I use, it doesn't matter. It happens with different programs all the time. Firefox, Konsole, VirtualBox, Amarok, whatever. I'll give you information like my motherboard model, CPU, Ram, whatever all day long if I know someone is actually going to do something with it. But like you said fglrx is for CAD workstations, average consumers don't matter to AMD.

No, that's what *you* said. What I said was that we could not "throw it to the bottom of the scrap heap" because we needed it for the workstation market.

Unless you can tell me someone is actually going to investigate something I'm not going to waste my time putting together my specs for a company who doesn't care unless I'm using CAD.

Again, the CAD thing is your statement not mine. I"m only suggesting that you fill out a bug ticket so our devs can reproduce the problem and have a chance of fixing it. The only promise I can make for sure is that if we *can't* reproduce it then it's only likely to get fixed by accident since nobody will be able to work on it.

I hate that you have to take the brunt of everyone's anger here in this forum and I commend you for keeping your cool as well as you do with some of the posts I've seen directed towards you. I wish you would drop the "we only support CAD users" defense though.

I'm not saying that. I am saying that workstation users represent the largest part of the market and that we need to allocate resources in a way that is at least approximately based on market size (be it current or future). Any company who says they don't do that is probably lying to you.

I hope AMD pays you well for monitoring this forum.

My "job" here is as the open source guy, but I try to help out with fglrx questions where I can. Everyone should have a hobby

[Erikina]

With respect, if you add the installed base estimates from all the major OSes, then compare the total with the estimated number of PCs in the world, something seems seriously out of whack.

Agreed, but Fedora figures are backed up by statistics and methdology (something I don't think any other distro has provided.) Also, keep in mind all those cheap One-Laptop-Per-Child (>1 mill) are running Fedora, as a fair few servers, and it's a favourite at universities.

Most estimates indicate that roughly 1% of the billion PCs in the world are running Linux, or 10 million total. You can find higher estimates but if you dig a bit you find the numbers are things like "PCs sold without an OS, and we assume the user will run Linux rather than pirating Windows".

The 1-2% estimates are generally derived from looking at statistics from web browser traffic. Something that will under-represent Linux as a whole .. although probably a decent metric for looking at desktop usage (as surfing the net net is a pretty desktop oriented activity).

Two years ago everyone told us that writing drivers was easy and that if only we would open up the hardware specs then the community would write better drivers in no time.

For what it's worth, opening the specs is the reason I have a ATI card now.

A year later that shifted to loud complaints that *we* weren't spending enough money writing the open source drivers ourselves, and what I'm hearing today is "we demand proprietary drivers, writing open source drivers is too hard and the open source drivers don't do enough".

What you're hearing is the complaints of people who want their hardware to work. I wouldn't exactly take it as a technical guide to writing drivers.

Despite all that, you are still telling us that adding features and support to the proprietary drivers is "easy" and that we're obviously stupid for not doing it.

Oh please. I've seen patches sitting in your bug tracker for months without getting applied. And I'm not even asking for a million new features, I just want it to work in a _4 month_ old kernel that you claim to support. The patch to get 9.6/9.7 to compile with later kernels is relatively minor (although, it looks like there are a couple other changes needed in your binary).

The website and release notes don't even mention the fact, that your driver will not work. The installer is all too happy to break the system. If this the best you can do, then maybe there is something wrong with the dev team..

The choice is pretty simple -- we can spread our efforts out and attemt to support all distros equally but make slower progress overall, or focus on a representative subset of distros and make progress more quickly, *then* focus on improving / speeding up support for the rest. The first approach means our overall quality of Linux support will be lower but we won't piss anyone off; the second approach means we can offer a good Linux user experience more quickly but will some distro users will feel insulted by our choices.

Again, I don't buy it. The way any decent team would do it, is make sure it works with the latest stable software (kernel, xorg). Then focus on the supported distros and features. Unsupported distros / build-it-yourself-stuff would just need to figure out how to package it themselves. That way, you're not increasing the amount of work (supporting later software is inevitable) while not completely shafting users.

[skywarp04]

Ok, maybe I took it a little out of context, but just to clarify these were your words.

Posted by bridgman
The fglrx driver is primarily written for the professional workstation market (CAD etc..), which is the main market for Linux graphics right now. It is unlikely that the open source drivers will ever be able to replace fglrx in that segment.

Go to page 6. Instead of saying CAD, maybe I should workstation, either way your not referring to the average consumer.

I am waiting for the open source driver to mature more. It just sucks not being able to fully utilize what I paid for. Surely anyway can understand that. The main reason I purchased an AMD card was because they were open sourcing their driver. I think a lot of progress has been made on it and I don't claim to think it will happen in a short amount of time. I'm just tired of being forced to use Windows to fully utilize my hardware just because hardware manufacturers choose not to support Linux. I'm know they are things like market share and other such business crap that drive where support is given, but I'm a computer science major. I hate business. I just like my crap to work the way it's supposed to.

[doubledr]

fairly speaking, 9.7 solves the slow xv problem and I am very happy.

[bridgman]

Ok, maybe I took it a little out of context, but just to clarify these were your words. Go to page 6. Instead of saying CAD, maybe I should workstation, either way your not referring to the average consumer.

I had to go back to page 6 the first time to remember the context

The point though is that I was responding to your suggestion that we scrap fglrx, and trying to make the point that even if it wasn't making consumer users totally happy yet scrapping it didn't make sense because fglrx had another market to serve.

I am waiting for the open source driver to mature more. It just sucks not being able to fully utilize what I paid for. Surely anyway can understand that.

You can fully utilize what you paid for today. You bought a card which advertised specific features and functions, and which said you required Windows in order to use those features and functions.

We are *also* bringing most of those features to Linux.

The main reason I purchased an AMD card was because they were open sourcing their driver. I think a lot of progress has been made on it and I don't claim to think it will happen in a short amount of time. I'm just tired of being forced to use Windows to fully utilize my hardware just because hardware manufacturers choose not to support Linux.

If you bought into the myth that we were just a bunch of idiots and that the open source community would be able to write a faster and more feature-rich driver in no time I'm sorry; I made a point of setting expectations low but there were a lot of wild statements floating around at the time. The reality is that modern graphics drivers are large and complicated, and that they take time to mature. We started working on consumer support about 18 months ago and have made a lot of progress, but it doesn't happen overnight and insulting the developers and their management doesn't actually seem to help, strangely enough

I'm know they are things like market share and other such business crap that drive where support is given, but I'm a computer science major. I hate business. I just like my crap to work the way it's supposed to.

Welcome to the club. There's a reason I put up with the abuse

[Joe Sixpack]

Let's All Grow Up...

I'm going to say something controversial, but I've been feeling like this for a long time now...

Most ATI users on this website need to have a tall glass of STFU! Yes, the driver has issues. Yes, you have to right to feel however you choose after you spend your money on hardware. But what annoys me is the lack of maturity amongst ATI users on this website. Even though fglrx has issues, the driver has had issues for years now (way before AMD bought ATI). So to expect a flawless driver experience is completely unrealistic. That's not making excuses for them - that's stating the obvious. When I bought my HD 3870 I intended on playing games in Windows, and I expected Linux support to be a work in progress. Which brings me to my second point:

Just because the feature you want isn't working doesn't make the new driver a complete waste. I'm sorry, but that line of reasoning is completely fucking childish. Catalyst 9.3 & 9.4 improved compositing support. Catalyst 9.4 - 9.6 each fixed bugs and significantly improved wine support (perfect example is HL 2 supports DirectX 9 now). So just like you can make the argument that theses drivers are worthless, I can make the argument that they're excellent because they fixed the bugs *I* was waiting on.

[Wintervenom]

Woo-hoo:

Looks like the linux driver was leaked for x86_64 people. It builds cleanly against 2.6.30 kernels and latest stable xorg. Performance is better than the recently released 9.7 driver.

Some more info:
OpenGL version string: 2.1.8975
2D driver version: 8.65.2
New extension: GLX_SGI_swap_control

http://www.simpleupload.net/download/448487/fglrx-8.650-ubuntu9.04-amd64only.7z.html

Source | OP Thread

[nanonyme]

Egh, so since not knowing what's inside closed drivers wasn't bad enough, now you also download unofficial releases of them from third-party websites.