Pete's Open Source Journal: Unsupported and bitter

[Kombatant]

My latest Open Source Journal entry is up on DriverHeaven, and it tackles a very "sensitive" subject: ATI drivers and Linux. You can read it here: http://www.driverheaven.net/~pete/article5.htm

I'd like your opinions on it :)

[Rys]

I think you're spot on for the most part. I get the very distinct feeling that they simply don't have the people in the company to devote to it, either. Without those people, it won't move forward.

I'm a moderate Linux user like yourself, using SuSE 10 with an ATI GPU (X600M) on my laptop, and one of the reasons I don't run X1K on my desktop is because I quad boot a bunch of operating systems (including SuSE 10) and an NVIDIA card is what works in all four, using the displays I have.

[Hector]

Ati linux support sucks badly. Very badly. Bad drivers, updates with almost no changes, slow in supporting extensions...

My GF6200 TC does better than my 9800pro, and it let me use XGL and all those pretty things.

[Ragemare]

I use Linux for everything, except for windows only games, so this really puts me off buying another ATi card. My 9800 works ok in Linux, I get a good frame rate in UT2004 and can use XGL, but NWN's water shader doesn't work and it was only receantly that 3D accelleration started working properly for me.

I completely agree with the article.

[Rolf N]

The problem lies mostly with ATI’s priorities really. From various talks I’ve had with ATI officials the past years, I always felt that ATI’s support for Linux depends on the operating system’s market share; so if Linux has a 5% usage among users, ATI will allocate 5 programmers out of 100 totally working on drivers (that’s of course a rough estimate, but it will give you an idea of the way ATI officials think about it).Heh. Think again. What exactly is Vista's market share?
What's this (http://support.ati.com/ics/support/KBAnswer.asp?questionID=19933)? And, for crying out lout, what's the first entry in the "Drivers & Software" dropdown on ati.com?

[Rys]

Heh. Think again. What exactly is Vista's market share?
What's this (http://support.ati.com/ics/support/KBAnswer.asp?questionID=19933)? And, for crying out lout, what's the first entry in the "Drivers & Software" dropdown on ati.com?
The difference is Vista will have larger market share fairly shortly after it's release, moving on to massive dominance over the course of its product cycle, simply because of the existing dominance of XP (and 2K before that, and 98 before that, etc).

[Ragemare]

Vista will have a massive market share eventually though. Although ATi probably want to hype Vista to sell hardware, which isn't strictly related to current market share.

[Dave Baumann]

ATI are already making money on Vista as they have OEM design wins which are being touted as "Vista Ready", this is largely because of the support they have already put into the Vista Beta's.

[Geo]

If they want to be the leaders they claim to be, then they need to make the investment. That simple. Both here and opengl (the fabled oft-delayed rewrite). All it takes is commitment (read $$) and willingness to stand up for it.

Orton: Terry, I want quarterly Linux drivers for all current products, and Linux drivers within a month of release for new products.

Makedon: Dave, I need another 10 headcount to make that happen.

Orton: Okayfine.

Analysts: Opex is up! Waaaaahhhh!!!

Orton: The price of leadership and good will with the enthusiast community. Besides, it's a competitive issue as our competitor is beating our brains in there on mind share with a dedicated, knowledgeable, and loud subset of enthusiasts. Deal with it.

There are a great many thorny problems to solve in graphics --this ain't one of them. Just a little matter of will.

[Rolf N]

Vista will have a massive market share eventually though. Although ATi probably want to hype Vista to sell hardware, which isn't strictly related to current market share.Whatever.
Linux had a non-zero market share five months ago. Vista didn't. And I'd also wager that Linux currently has a larger market share than the Vista beta versions.

My point revolved around the "We allocate our efforts in accordance to market share" line of reasoning.

[Dave Baumann]

From a business perspective its far more important to have good Vista support even before the release because OEM's will be shifting millions of units right after the release of it - support of Vista now is already gaining OEM wins in anticipation of it. If a major OEM start shipping with Linux out of the box then the dynamics start changing.

[Geo]

I'm a moderate Linux user like yourself, using SuSE 10 with an ATI GPU (X600M) on my laptop, and one of the reasons I don't run X1K on my desktop is because I quad boot a bunch of operating systems (including SuSE 10) and an NVIDIA card is what works in all four, using the displays I have.

I suspect that's the part they don't get. They say to themselves, "well, no one is gaming on Linux and no one buys a new top-end card except to game" and let it go at that.

[Kombatant]

From a business perspective its far more important to have good Vista support even before the release because OEM's will be shifting millions of units right after the release of it - support of Vista now is already gaining OEM wins in anticipation of it. If a major OEM start shipping with Linux out of the box then the dynamics start changing. One thing does not exclude the other. In order to make a better Linux driver, other driver versions do not need to suffer. You just need to accelerate development and shift priorities. And besides, nVidia has a better Linux driver; that doesn't seem to affect its Vista/Windows driver development afaik.

[Dave Baumann]

What on earth does NVIDIA's resources and scheduling have to do with ATI's? Nothing, they are different companies with different priorities.

If, as you say, it doesn't have to affect any current software development then the logical conculsion to that is that they need more resource - when a company is getting beaten up financially, in terms of revenues, margins and by the financial community (and having serious issues with an entire product line, delaying the software development for their primary platform) how prudent is it to add more resource to an area that, if it carries on as it is at the moment, is almost certainly going to sap more finances than it is going to grow them?

Different companies, dependant on the priorities may have, will have different outlooks to that question. But for the time being it looks like ATI won't dedicate much resource to it and put up with these kind of occasional outbursts from the Linux community every so often.

[Geo]

I'd completely forgotten that they'd moved to monthly Linux driver releases. That makes it even odder in my mind.

[Kombatant]

What on earth does NVIDIA's resources and scheduling have to do with ATI's? Nothing, they are different companies with different priorities.

If, as you say, it doesn't have to affect any current software development then the logical conculsion to that is that they need more resource - when a company is getting beaten up financially, in terms of revenues, margins and by the financial community (and having serious issues with an entire product line, delaying the software development for their primary platform) how prudent is it to add more resource to an area that, if it carries on as it is at the moment, is almost certainly going to sap more finances than it is going to grow them?

Different companies, dependant on the priorities may have, will have different outlooks to that question. But for the time being it looks like ATI won't dedicate much resource to it and put up with these kind of occasional outbursts from the Linux community every so often. You mention that ATI is beaten up financially etc etc etc - this is merely temporary, ATI's stance towards Linux drivers has nothing to do with their current financial abilities, revenues etc - because if it were so, it would have made excellent drivers the previous years when it didn't face such problems. So your argument is kind of moot.

All in all, you are just re-stating something we already know Dave; that ATI has made certain business decisions which state that Linux cannot get enough support because there are other priorities that come first. We already know that. The reason behind the article was that, in my opinion, this needs to change. There are ways to do that without it costing an arm and a leg in human resources and $$$ - and they have been made aware of it. All it takes is proper planning.

[Dave Baumann]

All it takes is proper planning.
Seriously, I love these statements! Have you put that to them? ;)

Question - do you think the X1000 software development is anywhere near where it should have been all over? IMO the entire thing was behind where it should have been because of the delays to the chips - the big indicator to that was the release of the Avivo updates over Christmas; this is a big feature all over, but especially in the notebook segment which will a very important selling feature and I don't think there is any way they would not have wante that ready in the OEM selling period for design wins. How would you plan for situations like these?

[Kombatant]

Seriously, I love these statements! Have you put that to them? ;)

Question - do you think the X1000 software development is anywhere near where it should have been all over? IMO the entire thing was behind where it should have been because of the delays to the chips - the big indicator to that was the release of the Avivo updates over Christmas; this is a big feature all over, but especially in the notebook segment which will a very important selling feature and I don't think there is any way they would not have wante that ready in the OEM selling period for design wins. How would you plan for situations like these?
Before answering your question, let me ask you something: do you think that ATI has ONE team, for both Windows and Linux/MacOS/other OSes, and they merely shift programmers from one task to the other? Because what you say would make sense only in that case.

[Dave Baumann]

Its actually applies in either case.

The reason why software development was behind on Windows was due to the nature of the issue they had with R520 prevented much in the way of any software development (most of the Windows developers were still optimising for X800 series very close up to the release of X1800, which is why you actually see a performance pop on X800's on the drivers released at that point). That same issue would have prevented development on Linux, so if they have a small dedicated resource then its just going to take time to get it all through, or if they have a more fluid resource allocation then other OS's are getting resource priority. (I would assume that there can be a fair bit of sharing between Mac and Linux these days, at least, however given the opportunity they have with Apple at the moment its probably not much surprise that would get the lionshare of resource recently)

[Kombatant]

Its actually applies in either case.

The reason why software development was behind on Windows was due to the nature of the issue they had with R520 prevented much in the way of any software development (most of the Windows developers were still optimising for X800 series very close up to the release of X1800, which is why you actually see a performance pop on X800's on the drivers released at that point). That same issue would have prevented development on Linux, so if they have a small dedicated resource then its just going to take time to get it all through, or if they have a more fluid resource allocation then other OS's are getting resource priority. (I would assume that there can be a fair bit of sharing between Mac and Linux these days, at least, however given the opportunity they have with Apple at the moment its probably not much surprise that would get the lionshare of resource recently) It actually doesn't apply in either case. And allow me to explain why.

First of all, Linux and Windows driver development teams are discrete as far as I know - they even moved development from Germany to Canada in order to have a better control of the situation. In the past year, there has been an effort on ATI's part to re-use some of the code that exists in the Windows driver to its Linux counterpart. In order to do that, the fglrx driver was put under the Catalyst umbrella and they started updating it more often and predictably than before (I believe you can look up how often the driver used to get updated before that). As I said, nowadays some parts of the code are common for the two drivers, so that there is no redundancy (and their aim is to have more and more common pieces of code).

What you say would make sense in the situation that there was one team doing drivers. In that case, ATI would put all of its team to finish and polish the Windows driver, with other versions falling behind. That is not the case though. The Linux team has different milestones, different bugs to fix, different development cycle. You can probably understand how that differs from the situation you described in your post. And after all (and I believe I mentioned it in my article) nobody actually asked for fancy 3D/gaming support - just 2D! They could have said "Here people, for such and such reason we didn't have time to do 3D development, but here's a driver in order for you to be able to at least use a GUI". I'd advise you to look at previous launches as well, to see how ATI offered support (usually with the next Catalyst release) in the driver.

Finally, allow me to quote ATI's Knowledge Base, in ATI Customer Care > Drivers and Software > Linux > General LINUX and XFree86 Information - For Open Source and Proprietary Drivers:

2D Graphics Acceleration Full-featured and optimized XFree86 drivers are available on the ATI website for the following product families ONLY:
RADEON 8500 and later retail products
MOBILITY RADEON 9000 and later mobile products
FireGL 8800/8700 and later workstation productsXFree86 currently provides hardware 2D acceleration for most ATI graphics adapters. For new products it normally takes at least 12 weeks from retail product shipment for driver support to be become publicly available.
Chances are your Linux distribution (https://support.ati.com/ics/support/KBAnswer.asp?questionID=23#vendors) includes a version of XFree86, which already includes support for your ATI product. The most recent release of XFree86 is available from www.xfree86.org (http://www.xfree86.org/).
Xi Graphics, Inc (http://www.xig.com/). specializes in creating high-performance, premium quality graphics software to support graphics hardware on computer platforms running Linux/UNIX®. They offer their own hardware accelerated X-Windows products as a commercial alternative to the freeware XFree86.See why I made this article? Few journalists use Linux and even fewer write about it. I felt it was a problem that needed to be known. I just call them as I see them, as always. Some people at ATI are probably pissed at me right now, and I can understand why. But criticism makes a better product - and their Windows driver is an EXCELLENT testament to that. If we start finding excuses for something that's not acceptable, we don't help fixing it. After all, this article came after more than 5 months of waiting for ATI to release a driver - not a couple of weeks.

[Dave Baumann]

12 weeks post the launch of the part probably involves about at least as much time on test hardware prior to the launch. If they weren't able to work on software prior to the release that because of the issues with the chips then its going to put the development behind, yes?

[Kombatant]

12 weeks post the launch of the part probably involves about at least as much time on test hardware prior to the launch. If they weren't able to work on software prior to the release that because of the issues with the chips then its going to put the development behind, yes? No; we are not talking about a full-fledged driver here, merely a 2D driver. Plus (and unless you have info that says otherwise, and contrary to what ATI has been saying all along), the problems they had with the R520 chip were simply that they couldn't reach the target clocks and nothing more. Thus, logic dictates that feature-wise, programmers didn't have a problem with the driver.

Dave, don't forget that all of the X1000 line of cards is unsupported, not just the X1800 cards. Plus, the 2D portion of the chip, which is probably shared in all X1000 line, was unaffected by the said problems (unless again you know otherwise)

[Dave Baumann]

No; we are not talking about a full-fledged driver here, merely a 2D driver. Plus (and unless you have info that says otherwise, and contrary to what ATI has been saying all along), the problems they had with the R520 chip were simply that they couldn't reach the target clocks and nothing more. Thus, logic dictates that feature-wise, programmers didn't have a problem with the driver.

A little before the Ibiza vent I was at the Develop awards in London and had Richard Huddy lamenting that we (press) would get boards before any of his developers would - "Hold on", I said, "Remedy had one at E3". "Yes, it was taken back" was his reply. At the Ibiza event during a conversation with Eric and Raja I pointed out that the drivers must be pretty polished given that they have had about 9 months of test boards to work on them - "No", was their reply, software couldn't be worked on because of the nature of the issue, that’s why (as I said) driver devs were still optimising for the X800 series and you see a performance pop. Eric explained that the issue they had with R520 was effectively random - every time they thought they had chased it down and written a workaround it would pop-up elsewhere. For this reason R520 didn't really go out of testing and validation until very late and that’s why you saw things like the CCC controls for the R5xx features being very rough in their initial drivers, many features not actually coming in until 3 months after its announcement and no developers having anything until after the release! For these reasons it would come as no surprise to me that the Linux side of the business may not have even seen boards until very close, or even after they were shipping.

With a normal release of a series of boards it’s usual to get some time before they are released with the "root" development chip beforehand, and that code is rolled out to the derivatives – that fact that all the X1000 boards are unsupported doesn’t have any bearing on what being discussed here as all the boards ended up about the same time (well, RV515 came at about the same time as early R520 XL’s); that still means they had the same development time with any of the X1000 boards and if the issue with R520 did prevent the boards going out of ATI’s testing and validation group do any degree then significant work for anyone of them prior to the release would have been delayed.

[Geo]

(and unless you have info that says otherwise, and contrary to what ATI has been saying all along), the problems they had with the R520 chip were simply that they couldn't reach the target clocks and nothing more.

Uh, no. Remember Sherman at R3D and his RENDER ERRORS? Turns out that was at least in the ballpark. And, yeah, it's been talked about here in several places since launch. And, I think, R3D's review.

But the "2d-only" point is an interesting one --did you suggest that to them? What did they say about the practicality of doing that?

[Kombatant]

A little before the Ibiza vent I was at the Develop awards in London and had Richard Huddy lamenting that we (press) would get boards before any of his developers would - "Hold on", I said, "Remedy had one at E3". "Yes, it was taken back" was his reply. At the Ibiza event during a conversation with Eric and Raja I pointed out that the drivers must be pretty polished given that they have had about 9 months of test boards to work on them - "No", was their reply, software couldn't be worked on because of the nature of the issue, that’s why (as I said) driver devs were still optimising for the X800 series and you see a performance pop. Eric explained that the issue they had with R520 was effectively random - every time they thought they had chased it down and written a workaround it would pop-up elsewhere. For this reason R520 didn't really go out of testing and validation until very late and that’s why you saw things like the CCC controls for the R5xx features being very rough in their initial drivers, many features not actually coming in until 3 months after its announcement and no developers having anything until after the release! For these reasons it would come as no surprise to me that the Linux side of the business may not have even seen boards until very close, or even after they were shipping.

With a normal release of a series of boards it’s usual to get some time before they are released with the "root" development chip beforehand, and that code is rolled out to the derivatives – that fact that all the X1000 boards are unsupported doesn’t have any bearing on what being discussed here as all the boards ended up about the same time (well, RV515 came at about the same time as early R520 XL’s); that still means they had the same development time with any of the X1000 boards and if the issue with R520 did prevent the boards going out of ATI’s testing and validation group do any degree then significant work for anyone of them prior to the release would have been delayed.
In my opinion, even if what you are mentioning here are the truth, do NOT explain the inability to provide even basic 2D support - don't forget that the driver gets updated very often for Linux too; it's not like it gets updated once every X months. So 3D-related problems shouldn't be an excuse (and I believe I don't have to remind you the percentage of the chip that does 2D nowadays).

[Dave Baumann]

According to your post above it takes at 12 weeks from the time of release to get a 2D driver, which would normally include whatever time they have had prior to that beforehand...

[Kombatant]

According to your post above it takes at 12 weeks from the time of release to get a 2D driver, which would normally include whatever time they have had prior to that beforehand... Dave, have you ever used the fglrx driver? Or even read the page from where I am quoting only a part? ATI does not release drivers with only 2D support; they talk about the driver in general when they mention the release.

[Geo]

http://www.hexus.net/content/item.php?item=5296

See, now all that, and a mere 6 weeks later. . . :wink:

Err, what's an "extra Avivo feature"? Like as in "Avivo? What's Avivo? We don't have no Avivo here. . ."?

[Nite_Hawk]

The only thing that I disliked about my 9700pro was the poor driver support under linux. It actually played a large part in why I purchased a 7800gtx to replace my 9700pro.

Nite_Hawk

[Geo]

I'm still waiting for someone with X1K and linux to come give us a report on the new drivers. . .