Originally Posted by Cambo
As long as a competitor can gain some advantage from deeper understanding of the workings of the GPU, the drivers will remain close-source.
See, here's the problem with that argument: A competitor can't
gain an advantage from a "deeper understanding of the workings of the GPU" based on the driver.
They can gain knowledge about how the GPU works (which is knowledge they did not have before), yes. But this does not automatically translate into an advantage. They'd still have to tool up an entire silicon production line (ATI already has a line, but they'd have to change their GPU's layout by quite a bit), and in the time they're spending doing this, the original GPU vendor can come out with its entire next generation of chips.
In a market that's advancing as fast as GPUs, you want
your competitors to copy you -- that means they're wasting time, and not making their own product better than the stuff you have in your pipeline already. At best (for them), they're making their product the same as your currently-released stuff, because that's what your drivers target. The next generation of hardware that you have in line is going to be better than that, almost by definition. So if you plan on releasing a new generation of your hardware every year or two, it's not going to hurt. (As long as your time between generations is shorter than the time it takes your competitor to tool up an entire silicon line. Even after they do that, their hardware is still no better than yours; at best (for them) they're at the same level.)
Last I heard, opening the drivers could not be done because of NDAs that nVidia has signed with at least S3, not any kind of competitive advantage argument. (OTOH, the S3TC code maybe could be put into a ROM on the GPU, and called from there. Depends on the terms of the NDA.)