Fixing up Quality mode, IQ improvements for the future & HEXUS.afterburner
It's worth stating at this point that ATI's default filtering optimisations that they use in their drivers, "out of the box", do not exhibit the same extreme MIP boundary shimmering issues as NVIDIA's most recent public drivers do. ATI also use a variety of texture filtering optimisations activated by their driver's A.I. feature. It's the out of the box image quality that we've recently asked NVIDIA to investigate and definitely improve, even if it means a performance loss. ATI shouldn't sit still on that either. There's always room for a bit more image fidelity.Arguably, graphics hardware is powerful enough today that image quality should be the primary focus of the IHVs and their driver teams. Performance has increased over the last few years to the point where texture sampling capability has topped 10 gigatexels per second from chips clocked at just 430MHz. That's a sampling rate that should preclude maximum quality being enabled out of the box on products with that kind of performance.
NVIDIA have outwardly stated to HEXUS that they realise image quality at driver defaults is something they should start to focus on more readily. Performance, rather than image quality, has been the driving force for consumer 3D since its inception, but that should steadily change. Given various cinematic computing marketing pushes in recent years, you'd hope that they'd have not let this issue slip through, especially since the 7800 GT and GTX are so bloody fast to begin with.
I think everyone can agree that making games look better is something that consumers would be happy with at the expense of a small drop in performance, given the raw ability of the hardware available these days.
Looking forward into the time before WGF 2.0, here's hoping the IHVs work to improve image quality for the consumer, even if that assumes a performance hit. The driver knows full well what hardware's pushing the pixels, so even selectively enabling some of the more performance-sapping enhancements can be done on select SKUs, and that should be a fair task for them to set themselves.
With Windows Graphics Foundation 2.0 pushing for identical filtering quality regardless of the supporting hardware, there should be less wiggle room for optimisations such as the MIP boundary sampling investigated by HEXUS and others in recent weeks. The focus on frames per second might, if we're lucky, slowly disappear. Image fidelity is king, surely?
IQ improvements for the future in hardware might include the readoption of less angle-adaptive methods of anisotropic filtering and the use of massive texel sampling rates to simply take more samples, more often. In software, the IHV's devrel teams educating games and application developers on how to limit shader aliasing, something the hardware can't do too much about, will help remove that image quality issue as much as possible.
Ah, but we preach to the choir! We were remiss in not investigating it sooner, especially since the issue's been around in some form or another since 2004. If we ask IHVs and driver teams to focus on image quality, we must do so ourselves. Rest assured we'll try to do so as much as we're able to.
NVIDIA should post 78.03 on nZone shortly, for public download, before a WHQL candidate is validated and made available directly on NVIDIA.com.
Paul's afterburner, originally found here, was based on the original set of benchmark results that showed a much larger drop in performance than actually occurs. His comments about performance therefore aren't correct, so the afterburner was removed. I've got more commentary on the whole thing from my own perspective here in my post on HEXUS.blogz.
