AMD Catalyst 9.12 For Linux Released

[rbmorse]

the Cloused Source driver is Evil because of Burning Money!

Hey John: rather than burning money, tell the PTB to declare a dividend instead. Stockholders would appreciate the gesture.

[bridgman]

If we kill off the fglrx driver, what do you propose that we do about the workstation market ? Should we just walk away from that business ?

AFAIK we're using the cash we have to buy back debt, which should make shareholders happy as well. You might have noticed a little bump in the stock price recently.

[smitty3268]

HAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHA.

You know, i actually started responding point by point to q - and then i just cracked up at the absurdity of it all.

[smitty3268]

You might have noticed a little bump in the stock price recently.

It's pretty amazing what you can do with a billion dollar cash infusion.

[bridgman]

I understand Qaridarium's argument, but I don't think the numbers work out. The main issues are :

a) the development cost to achieve fglrx 3D features & performance via open source drivers would be even higher than using fglrx, since the work would have to be Linux-only rather than being able to share the costs across all OSes

b) it's fun to rant about $2000 graphics cards but $300 cards are a lot more common :

http://www.newegg.com/Product/Product.aspx?Item=N82E16814195074

c) even if killing off the workstation products and spending the fglrx budget on open source drivers magically gave us 100% of the Linux consumer market, or even if the entire Linux workstation market magically started buying our consumer cards instead of competitors workstation cards we would still be worse off financially.

It's a nice idea but I haven't seen any explanation of where the "making lots of money" part comes from.

The trend from OpenGL to DX for new workstation apps has been going on for a long time, but Linux and OpenGL still make up a big percentage of the workstation market, over 50% in some segments. Even the segments where DX has made big inroads usually require that vendors provide both Linux and Windows workstation driver support, so dropping fglrx would also cost us a chunk of Windows workstation business as well.

Do you have a plan to get NVidia to drop their workstation products as well, or would it just be us ?

[V!NCENT]

a) the development cost to achieve fglrx 3D features & performance via open source drivers would be even higher than using fglrx, since the work would have to be Linux-only rather than being able to share the costs across all OSes

Whoah... What about Galliuym3D? I thought that was cross platformish... Or is it cross all platforms, except for Windows? Or am I wrong, once again?

c) even if killing off the workstation products and spending the fglrx budget on open source drivers magically gave us 100% of the Linux consumer market, or even if the entire Linux workstation market magically started buying our consumer cards instead of competitors workstation cards we would still be worse off financially.

Isn't there at least some chunk of code that could be recycled? Both ways? Or are the drivers just totally different?

[bridgman]

Gallium3D is cross platform, but the point here is not the lack of cross-platform internal APIs (we already have those, right ?) but the fact that we can not open up our own cross-platform source because support for DRM is embedded all through the stack. Remember that robust DRM is a hard requirement for 98-99% of our market, and is considered to be "pure evil" in the other 1-2%, which makes for an interesting architectural constraint.

Going to Gallium3D doesn't change that.

At the start of the project we were hoping to leverage some code from the proprietary drivers, particularly the "hardware layer" which corresponds roughly to the Gallium3D API, but we found during the first year that this wasn't really practical. The proprietary drivers are much larger than the open soruce drivers, partly due to the additional features and partly as a result of performance optimization, so even the low level "hardware layer" was over a quarter-million lines, bigger than the entire open source driver stack.

The source code architecture of the proprietary drivers was intended for effective maintenance of "large code by a large team", so the degree of modularity was much higher than the open source drivers, which are designed more for maintaining "small code by a small team". The right amount of modularity is always good, of course, but the amount that is right for hundreds of developers and millions of lines of code is not the same as what you want for handful of developers and tens of thousands of lines...

The other big point is that the lower levels of the open source stac (memory management etc..) are not designed for portability across all the OSes we support, while our internal APIs are. Again, we can't just open up our internal APIs without putting our DRM implementations at risk, so even if we did use portions of our code they would have to be heavily modified to work with a different (probably the existing GEM/TTM) lower level stack, which would be less work than a complete rewrite but not much less.

[Panix]

I am just curious. This question is for bridgman. Has there been developers and/or a group who have suggested various plans to raise more money or resources on the fglrx driver? What about an idea to freeze development on the OSS driver until the fglrx drive is up to par on the Nvidia one? Or trying to find a method to raise more resources?

Like I have suggested before: seems like not enough resources and trying to please everybody has both groups complaining about something.

It's obvious why one (both the consumer and AMD/ATI) would want the fglrx driver working "100%" and the OSS driver is desired to work at an optimal level but wanting that support/performance with the proprietary drive is even particularly desired. You want all features an optimized performance and Windows is the OS where the money is. But, if you had a Linux-based proprietary ATI driver at par with Nvidia, I think there would be some signficiant financial gains. I think a lot of people dual boot and even if they are primary Windows users, a significantly improved ATI driver means more sales. ATI cards, down the line, are cheaper than Nvidia cards. I am talking about the consumer market but the gamers market is significant as well when totalling the numbers.

If I am dual booting Windows and Linu and I know an ATI card in Linux is on par with the Windows OS, I can save $50+ on any ATI card I get for the same or even better performance. When you look at 3D and not gaming, it is still important to some to have that option, to have good 3D. If you are an ATI owner and you are having issues no matter whether you use OSS drivers or fglrx, you are probably extremely frustrated and that is an understatement if you own a laptop and can't change cards easily.

Just my two cents. I'll shut up now...

[cutterjohn]

Hey... whoa... wait a second...

AMD releases specs and is helping FLOSS drivers. The 9800 pro (r300) I am using here at a friend of mines can play Half-Life2 under wine with the FLOSS driver at almost the same framerates as it did under XP when I bought that card.

I am happy with that. r600 is on its way and I can run my r600 card with the Catalyst driver.

Also do not forget about the, was it Steam under Wine?, bug that ATI has fixed in the Catalyst driver. So they do care.

Try to relax a little, because AMD's new ATI policy is still fresh and they are putting in efford to Linux. Be at least a little easy on them...

Linux has always had its 3D GPU issues... The situation is pretty well known: nVidia currently own with the proprietary driver, older ATI cards are doing fine under FLOSS. You should have known that. Always Google around first before just buying things. That counts for everything and not just computers.

I do agree that Catalyst for Linux is a bit of consumer hell as it is today...

Didn't have any choice in the matter. The notebook with the specs that I wanted only came with the mobility 4850 and I knew very well the crappy ATI linux driver situation which is why I avoided their products like the plague until this notebook. nVidia was lagging with their refreshed nb GPUs at the time as well, so it was go with an older G9x based nVidia product or a newer R7XX based ATI product, so I decided to experiment a little(see below).

Also at the time of purchase I figured that I was safe for a while with a desktop w/nVidia card, however the desktop crapped out (suspect x2 4800+ was damaged in moving it out of the way, unfortunately no spare socket 939 to test with and they simply aren't worth the price to replace).

Hey! No one answered my question about the 9.12 on the driver dl page being the one with the hotfix already included or not? i.e. will a 9.12 driver downloaded after the hotfix release have already been updated? Anyway to tell from version #s?

[bridgman]

My understanding was that the hotfix drivers were off a newer code base than the regular 9.12 drivers, ie they weren't just "9.12 plus a fix".

I'll be in the office this afternoon, will see if I can get more info.

EDIT - OK, looks like the regular Cat 9.12 driver is the "Final" build from the 8.681 release. The hotfix driver is a "Release Candidate" from the 8.682.2 release and has only been QA'ed on HD4xxx and HD5xxx -- presumably those are the cards where the changes are helpful.

So... the hotfix does not replace the regular Cat 9.12, it just offers specific fixes and should be picked up only if you need those fixes. If you have a 6xx product, stay with 9.12.