So did AMD really build a better filter?

Arguably they didn't, in terms of overall image quality. The linear nature of the tent down-filter and the fact it takes into account contribution from neighbouring pixels means that image blurring is a natural side effect. There are simply better filter kernels to be applied, especially when you're looking outside the pixel, which is the key property of what AMD did. However balancing those filters against computational cost and hardware limitations is likely a key reason why tent was chosen.

That said, the key advantage comes not in terms of being able to apply these filters in the first instance; NVIDIA or anyone else is quite free to implement the same filters to give equivalent image quality, if that's what the customer is looking for.

No, the advantage comes because they built their hardware with a fast decompression path into their unified shader core. Samples and sample locations can be made known to the compute-heavy core of their latest chips in an efficient manner, in order for the filter to be computed and final pixel colour written. Current competing hardware doesn't have that luxury.

As for the filters they might want to investigate in the future, we (and we presume AMD, NVIDIA and everyone else doing post-filter AA for that matter!) are still working on it.

Down-filter Conclusion

We think it's likely that we'll see AMD experiment with different filters over the lifetime of their current product generation, before taking the concept to their next GPU architecture generation. They can't really abandon it, even if you argue it has only come around as a result of a disabled ROP resolve path, as we do.

That it might be a result of a hardware limitation elsewhere in the chip is somewhat serendipitous for all concerned, though. Combine non-box down-filtering with better subpixel coverage determination and you get AA image quality that increases another step over what we've got already, so it's likely that's going to happen across the board. The programmable AA down-filter will become a fast path in all consumer hardware in due course.

How to build that in hardware? It seems like a reasonable evolution of current approaches to us, but we'll certainly leave it to the experts! Hardware engineers will thus read this and shake their fists at the screen, but when we think about increasing post-filter AA quality in a modern graphics processor, better subpixel coverage determination and better-than-box filtering is what comes to mind.