Luminescent
Veteran
Are G80's MSAA sample patterns programmable?
G80 MSAA sample patterns
- Thread starter Luminescent
- Start date
No. Or at least, I've seen no indication that they are.
Follow up question: what are the advantages of having a programmable MSAA pattern?
I don't quit see why it's bad. The multisample patterns of the G8x are quite good.
It's quite bad for SLIAA & friends though...
Not *that* bad, and even though it's not ideal for SLIAA, it's not like those modes are going to improve things much beyond the excellent modes already available.It's quite bad for SLIAA & friends though...
I thought line AA was a different thing anyway? That they use an analytical AA method?
As far as multisampling position values. An interesting comparison might be SLI 16x on the G80. Which consists 8x CSAA ((4x Color Samples *2 SS, and 8 CSAA)) The most interesting comparison is how it compares to 16xQ which is also 8 color samples per grid. ((it looks the same for the most part. Not calculating the additional SS for alpha/texture variants))
Though for those curious. SLI32xQ uses this sampling position for its color samples. ((Not including the CSAA which should distinctly share the same position as they do in 16xQ))
Though for those curious. SLI32xQ uses this sampling position for its color samples. ((Not including the CSAA which should distinctly share the same position as they do in 16xQ))
Last edited by a moderator:
Luminescent
Veteran
Those are fairly interesting AA modes, Chalnoth, and you may be right about questioning the value of programmable patterns, given the modes already provided by Nvidia. Being that Nvidia has chosen not to support programmable patterns for two major architecture generations, why did Ati choose to do so? I do not remember hearing much about this capability being taken advantage of by developers; maybe Nvidia forsaw this, while Ati implemented it too early in the development of DX9 to see it as a waste. Let's say it was/could have been useful, could the possibility of patent infringement have deterred Nvidia from trying something similar?
Programmability of ROP hardware maybe available at some point in the near future, but how near is it? I think it all comes down to the API and the fact that programmable MSAA sample patterns are clearly not a highlighted feature of DX10, although this is also the case for CSAA.
For some reason I felt programmable patterns were a good step in the evolution of the ROP from being a fixed function device to a programmable/more configurable one. This was the case with the register combiner, which became the shader processor. It maybe that full programmablility of of the rop will never come to be, since doing so would forfeit the speed advantages gained by using custom shaders.
Finally, have we considered whether the lack of sample pattern programmability is in fact a limitation of the hardware?
Programmability of ROP hardware maybe available at some point in the near future, but how near is it? I think it all comes down to the API and the fact that programmable MSAA sample patterns are clearly not a highlighted feature of DX10, although this is also the case for CSAA.
For some reason I felt programmable patterns were a good step in the evolution of the ROP from being a fixed function device to a programmable/more configurable one. This was the case with the register combiner, which became the shader processor. It maybe that full programmablility of of the rop will never come to be, since doing so would forfeit the speed advantages gained by using custom shaders.
Finally, have we considered whether the lack of sample pattern programmability is in fact a limitation of the hardware?
Last edited by a moderator:
I'm not talking about line-AA but about some specialized requirements. Take Flight Sims as another example. Most of the time horizontal lines should be of more importance than vertical ones. At least that'd be my guess.
Right, and I can see that, but I guess I'm wondering why there would be much difference from what is already offered.
There's no API for it so they can't. It's very useful for the driver writers though for things like for instance the CrossFire SuperAA modes.
I can see the benefits for CF/SLI.
But if it's true that programmable patterns will be introduced in DX10.1, I assume there also have to be benefits in non-CrossFile/SLI modes? Any ideas what they may be?
It was suggested above that some patterns may lend themselves better for, e.g., flight simulators where you have a lot of horizontal lines. I can see that there may be some benefit if the lines are truly horizontal, but in real life they never are 100%.
NocturnDragon
Regular
The position of the samples, let's make a simple example with 2 samples.Q: What makes a pattern more suited towards X angle
Code:
_______
| * /| In this case evey sample will catch a different color.
| / | (suppose the pixel is half white and half black the result will be 50% gray)
| / *|
|/_____ |
_______
|\* | In this case both samples with catch the same color.
| \ | So the result will be either be black or white
| \ *|
|______\|
Last edited by a moderator:
