Aquamark3: Accurate Benchmarking for Old and New (DirectX9) Apps?

Quality Conclusion

It is astonishing how the bugs made their way into the NVIDIA drivers. As mentioned before, beta versions of AquaMark 3 have been available to the IHV for months, and they are so obvious that it's very hard to overlook them. Aquanox 2, using the same engine, has been available since the end of last year-it was already used as a DX9 benchmark in our lab so it couldn't have been overlooked by NVIDIA.

After discussions with 3DMark 2003, the filtering quality in UT2003 and the allegations made by Valve last week, NVIDIA is not keen on initiating a discussion on driver optimizations - which count for a lot in terms of image quality.

The tests in the game Aquanox 2 show that the missing lighting in the drivers is indeed a bug and that NVIDIA never intended to increase the performance by failing to render the lights.

That's also NVIDIA's statement on that. They say that they found a couple of broken things in the drivers, and they will have a new Detonator 50 build available that will fix the issues.

But how did the bugs get into the driver in the first place? The old version v44.03 renders absolutely perfectly. Do optimizations - which NVIDIA needs to put into its drivers to get the best performance out of its graphics processors - increase the risk for new bugs?

Finally, we can't estimate how much the missing lighting and shadows affect the score with AquaMark 3. But not to render objects normally does not result in lower scores. So all results with NVIDIA's v45.23 and v51.75 are invalid and should be taken as preliminary.

We can also say that the low anisotropic filtering quality of v51.75 is a bug, but even if we force the filtering in the control panel, we get a lower quality than in previous drivers. Our speculation is that NVIDIA reduced the filtering quality to achieve a better performance.