Personally id love a driver option that allowed you to choose the precision used or let it be application controlled. NVIDIA's idea (or was that a rumour also) of having it use FP16 unless
it specifically requests FP32 seems a fair one. After all if a scene looks the same with FP16/32 why not use FP16 and have it run faster?
I imagine this one will be heard about for a long time, in the same way as the ATi 'Quack' issue is still dragged up even now. Mud tends to stick, and once a company gains a reputation for doing this kind of thing it takes a long time to shake it.
Well i think what they did with the drivers was, while not the best thing in the world, not the worst either. They must have realised that while the results of 3dmark03 were not a ‘fair’ representation of the cards performance that people are still going to take notice of it. Because after all they are coming from the position of having a card give a lower score in it so of course they are going to say its bad. The driver was a nice wakeup call to prove their point. They were open about it too – although they have not said what they have done. I’m sure they will admit whatever there is to admit when they have some WHQL sorted.