A lot of information comes from phoronix - but very few actual links to sources of information. Sometimes it's an opinion of phoronix, not actual current information. So until I read a direct quote from unigine, I will take any phoronix story with a grain of salt (just like semiaccurate articles).
Unfortunately, the only choice available to Linux user is Nvidia if this user wants fully functional GPU.
You want ALL features working ? buy Nvidia
You want video acceleration today ? buy Nvidia
You want stable 3D performance for gaming ? buy Nvidia
I don't like this situation as I own ATI card. Just looking to replace it. Under Linux this HD4850 is still useless as desktop card.
It's not important what I can do as far as there much more then one thing that I can't.
And don't say open ... The open source drivers are even less useful.
I like the idea, and wish them luck, but as far as hardware acceleration for video will not be there, they are useless on AVERAGE DESKTOP.
I hope that somehow Google's VP8 release as open source will allow AMD to release HW acceleration documentation for this format.
Also, AMD's drivers are typically best used when you stick to the opengl spec as strictly as possible, which many people don't do (even I've been caught out by it sometimes). AMD isn't always to blame; sometimes the developers need to fix their code instead.
I'd actually say that it's about 60% or so of the time it's the developer that oopsed something in their code- like you said, many don't apply a strict interpretation of the meanings in the spec docs. It's a source of recurring gotchas within many titles. They didn't get a shader coded right and missed a key value which NVidia supplies, but AMD doesn't. They didn't pay attention to the "may" in the spec doc and presumed it was a possibility- but it wasn't going to happen to them. That sort of thing.
This seems like a good time to remind that the OpenGL/OpenCL stack in the Linux Catalyst driver (aka fglrx) is the *same* as the one in the Windows Catalyst driver. The OS-specific glue is different, of course, but the core code (>95%) is common - that's the main reason binary drivers exist in the first place.
It's my understanding that much of the issues are in the OS-specific code in the first place, John. A good portion of the issues I know about are more due to that than the main codebase. The bulk of things wouldn't work very well in Windows if it was in the main code base- and we wouldn't be having this discussion because you'd be out of a yob because AMD would be out of the graphics business at that point.
I think the people that're complaining are asking for AMD to put a little more love into getting the 5-10% that's the cause for much of the consternation here straightened out- or at least moreso than it is now.