Am I the only one that is happy to hear there is FS11? With diminishing list of features that can be added to each new version of FS, I didn't assume that there would be a FS11 for sure.

>Am I the only one that is happy to hear there is FS11? With>diminishing list of features that can be added to each new>version of FS, I didn't assume that there would be a FS11 for>sure. FS is a big money maker for MS, why wouldn't there be? :

money maker, yes. big, no, not in a company that has Windows and Office.

>DX10 is already mostly what it is cracked up to be; the parts>that arent can be attributed to 1st generation hw ( eg>Geometry Shader performance ).You've oft mentioned that geometry shader performance isn't what it was expected to be (and you are certainly not the only one I've heard state that). But what indication do we have that this is really a 1st gen hardware "problem" rather than the DX10 geometry calls instead simply requiring more processing resources than estimated? After all, even though the word on the street for a while was that the Radeon line would kill the NVidia line when it came to geometry shader performance, from what I've seen while the Radeon doesn't do *that* much better (despite having an "accelerated" geometry shader per AMD's marketing dept):http://www.digit-life.com/articles2/video/...ark2-page4.htmlAnd what about real-world payoff? Call of Juarez is proported to make heavy use of geometry shaders:http://www.anandtech.com/printarticle.aspx?i=3029So we've had 2 completely different architectures built to implement the same functions, and both perform poorly in this area compared to others. What reason do we have to expect greater improvement in this area compared to performance as a whole in the next generation? Might it simply be that the concept of geometry shaders isn't what it was cracked up to be?

hilarious :-lol

>If he wasn't using DX10 hardware then how could they have been>DX10 screenshots?Maybe that's what Phil meant when he said they were "not representational"

Phil thought he was breaking the NDA and hence posted DX10 screenshots. There will be little or no noticeable difference between DX10 and DX9.

>Phil thought he was breaking the NDA and hence posted DX10>screenshots. Huh? I'm quite certain the entire patch (which contains more than just DX10 support) would be under NDA, so the fact that he was posting screenshots of the beta would be breaking the rules whether or not he was using DX10 support.> There will be little or no noticeable difference between DX10 and DX9.You say that with confidence... are you a beta tester? (Don't answer that) Otherwise, we'll find out soon enough.

Don't worry wingnut is not a beta tester and has no idea what he is talking about.Jimhttp://www.hifisim.com/banners/hifi-community-sigbanner.jpghttp://www.hifisim.com/

Hey, Jim... is there something in there for you guys? (cough once for yes) ;)

1st gen DX8 hw wasnt a win for using shaders, as fixed function was just as fast ( and maybe a tad faster ). that changed quickly, as the hardware designers saw real-world usage and changed their pipelines. its an expectation, true, but I expect that the hardware designers will do the same with the new DX10 hw units.

Yeah, but previous generation cards were more of a rigid-path design, whereas the DX10 generation parts are explained to be much more abstract FPU engines (hence the GPGPU stuff). So it seems to me that maybe it just comes down to the raw FPU power required which will certainly go up in future generations, but which wouldn't really benefit the geometry shader any more than the other shaders.But obviously you would certainly have a much better idea than me... I guess we'll find out.

Ok, I am not a hw designer. But as I understand it, the main issue is the GS unit emits vertex data, which touches memory, which changes the latencies through the pipeline. By that I mean the more vertices you emit ( the more you amplify ) the more memory you are touching and the greater the latency that creates. And of course the unit is more useful when you can emit much more than 4 vertices per vertex to be amplified ( which is essentially what a point sprite is, and those already work in DX9 ).This is fundamentally different than the other hw units, which consume the vertex and pixel data without creating any new data, and have deep pipelines and celver techniques to jam as much thru them at once and thus reduce latency. Parallelism via SIMD is what 3D hw is all about, since RGB and XYZ are both 3-tuples that can be operated on by the same instruction so you get 3 data paths touched by one operator; true single instruction multiple data.So the GS is different, it is harder to parallelize since the emit count is essentially unknown until runtime, and will require some rethinking to get it past where it is today. Fixed function is rigid path; DX8 and shaders changed that. DX9 and SM 2.0 are certainly not rigid path. GPGPU was certainly done on DX9; its just better on DX10.

gotcha, makes sense. thanks!

>Don't worry wingnut is not a beta tester and has no idea what he is talking about.Jimbo, you have no idea if I am a beta tester or not. Let us see after Phils next blog if I what I said was right.