WOW

http://www.iryoku.com/separable-sss-released

It's quite similar to the NVIDIA technique but the benefit of this technique seems that you can calculate it in screen space.

There's one thing that bothers me though. Usually in games you never see the faces of people so close and even if you do you rarely see models detailed enough to appreciate SSS. From a distance it looks quite ordinary and there's little use for this technology.

Still a great job though. It looks really nice most of the time. The specular lighting  is still missing something. But it's miles ahead of what we usually see in games.

Yes. Only practical for reltime in game cinematic.

If I'm not mistaken I already saw a screen space SSS approach before. I think it was in the latest GPU Pro book. Might be the same author of course.

Calculating the SSS in screen space instead of texture space does really make things easier when you have many people in your scene.

If I'm not mistaken I already saw a screen space SSS approach before. I think it was in the latest GPU Pro book. Might be the same author of course.

Calculating the SSS in screen space instead of texture space does really make things easier when you have many people in your scene.

Indeed, specially with deferred pipeline. But i think problem with full deferred and ssss is a limit number of buffers, unless going for dx10,11. I would not go for prepass method in dx9, cause it will drop performance alot.