If you were to stick with the 'xS' terminology as it was initially intended no 'xS' modes should work with OpenGL - they have to stick within the confines of the multisampling definintion.
Has anyone seen the txture sample positions fro 8X or 8xS?
GFFX Anti-Aliasing with FSAA Tester
- Thread starter Brent
- Start date
If you were to stick with the 'xS' terminology as it was initially intended no 'xS' modes should work with OpenGL - they have to stick within the confines of the multisampling definintion.
Has anyone seen the txture sample positions fro 8X or 8xS?
All of this just seems like a major hardware limitation. The programmers are trying to code around the fact that Nvidia can't do more than 2X RGMS in hardware, I guess? But then it doesn't make any sense, because why does 2X AA on the FX suck so much? Couldn't they do that as straight up 2X RGMS to make it equal to the R9700's? I mean, the R9700 looks a lot better if you just compare 2x to 2x, and I don't believe it's all gamma correction...
THe_KELRaTH
Regular
Advantages of R300 using a Programmable Jitter pattern perhaps!
Brent, give the Tenebrae mod a go. I'm curious about the GeFX's stencil fillrate as well but with additional fillrate requirement as well as per-pixel operations thrown in. Let me know ( reverend@beyond3d.com) if you want to know the console variables wrt stencil-related cvars.
martrox said:
3dfx used accumulation buffers, in VSA-100 the T-buffer, in products past that the M-buffer with 2x/4x RGMS.
What I recall quickly from the top of my head being present on NV25 coming from the 3dfx "portofolio", was the "filter on scanout" trick for 2xRG/Quincunx, the VPE, some re-arrangements for the gaussian filtered modes, 4xS...I'm sure I'm forgetting a few others, but most of them if not all carried over to the FX too as it looks like.
What Gigapixel concerns if someone does have some old caching/memory optimisations patents from them stored they might become handy too.
I recall NV claiming in ancient powerpoint presentations that A-buffers are hard to implement in hardware; no idea why they never opted for one. It would have been better for MSAA IMO.
Perspective is correct. Yet they had plans for MSAA in the products after the VSA-100.
Why wouldn't it be optimum? Anything else and there are going to be some preference to horizontal edges or vertical edges in AA quality. The only difference in the sampling pattern between the Radeon 9700 Pro and GeForce FX is that the 9700 Pro's pattern is at 45 degrees, while the GeForce FX's is at 135 degrees (0 degrees points straight right). This means that these angles are the ones that they'll each look the worst.antlers4 said:
As for the 9700 Pro's pattern, it is a static pattern, at least in today's drivers. ATI has not yet implemented any sort of per-pixel pattern switching. I don't know if it is possible to do, and I doubt the effect would be positive on 2x AA.
OK, I don't know the theory behind this, but I do know that when the VSA-100 did 2x it did not do 45 (or 135) degrees. I assume they had a reason for not doing 45 degrees, since it wasn't any easier to do 25 or whatever it was they ended up doing.
It was my understanding (although I could be mistaken, again) that the 9700 doesn't do 45 degree either, but somewhere around 30.
I think the reason for not doing 45 degrees is so the samples don't line up in neat diagonal stripes, which I believe is more prone to artifacts than the herringbone pattern you get with other sample angles. I think this is supposed to more than make up for favoring the vertical or horizontal component of anti-aliasing.
It was my understanding (although I could be mistaken, again) that the 9700 doesn't do 45 degree either, but somewhere around 30.
I think the reason for not doing 45 degrees is so the samples don't line up in neat diagonal stripes, which I believe is more prone to artifacts than the herringbone pattern you get with other sample angles. I think this is supposed to more than make up for favoring the vertical or horizontal component of anti-aliasing.
You say you know what the V5 did, but you are wrong. The 2x FSAA in the V5 was at 45 degrees. The drivers say it is so.... if only basics program worked in fullscreen....
CeiserSöze
Newcomer
45 degrees is non-optimal because 40-50 degree line aliasing effects are the next most visible to the eye because of the single-pixel-stair-step effect (as soon as you divert from the 1-across 1-up pattern then you get a noticeable discontinuity).
It's all about maximising the effectiveness of the few samples you have. As the number of samples rises the pattern gets less important - but of course, anything that can be done to make the best of the critical H, V and 45 degree lines is well worth it.
It's all about maximising the effectiveness of the few samples you have. As the number of samples rises the pattern gets less important - but of course, anything that can be done to make the best of the critical H, V and 45 degree lines is well worth it.
