X1300 for a high resolution 1920x1200 LCD display

[nyc_paramedic]

Fastest card with open source 3D support today would be something like a 1950XT. The 1950 Pro is maybe 3/4 as fast but quite a bit less expensive if you can find one.

I forgot to ask. Is there a 3D performance difference between the open and closed drivers? If so, will that always be the case?

[Melcar]

I forgot to ask. Is there a 3D performance difference between the open and closed drivers? If so, will that always be the case?

On my 9800pro, x800, and 200M, I have seen on average a 30-70% drop compared to fglrx when it comes to pure speed.

[nyc_paramedic]

On my 9800pro, x800, and 200M, I have seen on average a 30-70% drop compared to fglrx when it comes to pure speed.

Ouch.

So even on really old legacy cards, like the 9800, the specs will never be truly open.

[Melcar]

Ouch.

So even on really old legacy cards, like the 9800, the specs will never be truly open.

It's not that the necessary docs. are not there, but rather that the driver is not optimized at the same level as fglrx.

[bridgman]

As Melcar said, it's not a documentation thing, more to do with a few missing things in the current driver architecture. The main "missing link" right now is the lack of a good memory manager in the 3d stack. That is being worked on now (the whole TTM/GEM thing you read about from time to time). After memory management is in place, the next obstacle will probably be a decent shader compiler for GLSL, and there is hope that the work around Gallium will help there.

There's no reason in principle why the open source driver couldn't match the performance of fglrx other than the huge amount of work it would take to get there. I expect that 10-20% of that work will give 60-80% of fglrx performance (assuming Gallium and the LLVM infrastructure around it works through as expected) and that is where the developers are likely to say "good enough" and go work on a different problem instead.

[nyc_paramedic]

As Melcar said, it's not a documentation thing, more to do with a few missing things in the current driver architecture. The main "missing link" right now is the lack of a good memory manager in the 3d stack. That is being worked on now (the whole TTM/GEM thing you read about from time to time). After memory management is in place, the next obstacle will probably be a decent shader compiler for GLSL, and there is hope that the work around Gallium will help there.

There's no reason in principle why the open source driver couldn't match the performance of fglrx other than the huge amount of work it would take to get there. I expect that 10-20% of that work will give 60-80% of fglrx performance (assuming Gallium and the LLVM infrastructure around it works through as expected) and that is where the developers are likely to say "good enough" and go work on a different problem instead.

Thanks for the detailed info. This makes a lot of sense.