Should I get an ati or nvidia card?

[Aradreth]

nvidias driver is a piece of shit, unstable, and 2d is completely broken on 8xxx.

hell, my core2 can with 20% of 1 core do software 2d FASTER than an 8800gt can do.. NVIDIA BLOB GREATNESS, yeah.... piece of shit.

Yeah nvidia has really dropped the ball driver wise over the last year or so and it's not just with linux. For what I've been told it their drivers are the cause of a lot of crashes if you use Vista.

[Redeeman]

not to mention xp 64bit - i know people that get crashes from it 3 times a day....

[Melcar]

It's true that nvidia's drivers suck; Vista drivers are incredibly flaky and Linux support for 8xxx and 9xxx cards is, well, lacking. However, 6xxx and 7xxx cards work wonderfully on Linux.
Still, the current fglrx incarnation is more than usable and gives out very good 3D performance on most of their supported cards.

[Redeeman]

It's true that nvidia's drivers suck; Vista drivers are incredibly flaky and Linux support for 8xxx and 9xxx cards is, well, lacking. However, 6xxx and 7xxx cards work wonderfully on Linux.
Still, the current fglrx incarnation is more than usable and gives out very good 3D performance on most of their supported cards.

3d performance on 8xxx seems fine on linux, but 2d is not there..

And you say 6xxx is working wonderfully? i wouldnt call regular hardlocks "wonderfully". Besides, its currently impossible to buy 7xxx cards now

[NeoBrain]

I really wonder why it's so hard for the binary drivers to get at least an acceptable 2D acceleration...
In my opinion the first thing a user notices when using a specific driver is the 2d accel, but neither ati nor nvidia seem to care about them (not including Catalyst 8.5, which was _supposed_ to improve 2d)?

[bridgman]

A cynic would say "because there are no 2D benchmarks"

I think the real issue is that we are all moving away from dedicated 2D hardware and moving towards using the 3D engine for 2D operations. The old 2D hardware and driver code had literally years of tweaking and optimizing; we have to do some of that work again on the 3D engine.

If you look at the progress of EXA on the radeon driver that will give you a pretty good idea of what is involved. It's not exactly the same situation but close enough.

- everyone accepted that EXA sucked and was too slow
- part of the problem was that all of the API functions were not supported
- once full support was added things got faster but much less than expected
- turned out that fixing the driver bottleneck exposed a server bottleneck
- otaylor made some changes to xserver, EXA text sped up dramatically

3D engines are generally faster than 2D engines these days, but sometimes are not faster at *exactly* the same operations. I suspect the whole stack will end up changing a bit since *different* things will be fast going forward. The good news is that so far it seems like the things we want to use (EXA, Render) will be faster, and "the only things that worked well before" (XAA etc..) will suffer a bit. I'm basing this on analogies from other OSes -- it's possible that the XAA functions are in fact identical to their EXA counterparts but I doubt it and will check today.

For R6xx acceleration we are probably going to do EXA first, for example.

[d2kx]

Well, my R580 has a 2D engine, and 2D performance in comparison with Windows is a joke. It is fast enough for most users I guess, but in comparison, it is still a bad joke. I can't tell for RV670 or newer...

[Redeeman]

hd 3450 on catalyst 8.5 is alot faster than nvidia 8800gt{s,x}

[deanjo]

hd 3450 on catalyst 8.5 is alot faster than nvidia 8800gt{s,x}

Would you care to post your gtkperf -c1000?

[bridgman]

dtkx: I was talking more about "new chips vs old chips" (driven by industry trends and techno realities) more than "linux vs windows" which we expect will continue to improve.

To the extent that Linux is across-the-board slower than Windows, that will have to be addressed more in the X framework than in the drivers themselves. That said, the framework is continuing to improve, and the framework changes made in conjunction with radeon driver optimization work should help all vendors and chips.