How much Video Ram do your games use @ 2560x1600?

[bystander36]

RGSSAA is rotated grid, and OGSSAA is ordered grid. Two completely different things. NV is stuck on good old OGSSAA, and I dont think they will ever improve it or do anything at all, they literally consider full-scene SSAA to be an out-dated remedy.

So SGSSAA is a bit of a bastardized SSAA that trades performance for quality vs. RGSSAA, but offers better performance. But it can actually result in the worst quality between all three because it tends to blur the most.

But anyways, the future is not any of these. Besides high PPI displays which is the true solution, the best quality to performance trade-off is a hybrid of multi sampling AA, morphological AA, and temporal SSAA. And the only algorithm that addresses all of them so far is SMAA.

SMAA 4x produces image quality can actually be better than SGSSAA, and performance is way better.

I'll play with that tool. I'm good with compromise.

Btw, when I said RGSSAA, I was referring to Regular Grid SSAA. Apparently RGSSAA can stand for both.

 

[Pottuvoi]

Hey all, so i'm hoping to get a new 30" IPS @ 2560x1600 and wanna know how much VRAM games use @ that resolution?

Here is the way to calculate the memory consumption of framebuffer. (textures and other things take memory usage independently from screen resolution.
((RGBA buffer for framebuffer X*Y*bytes in buffer per pixel)+(Z-buffer 4bytes))*AA samples/conversion to megabytes.


So for 2560x1600 4x MSAA framebuffer using fp16 HDR RGBA buffer (where each component of RGBA is 2bytes) and normal z-buffer.
((2560*1600*8)+(2560*1600*4))*4/1048576=187.5MB
Front buffer is also 2560x1600*4 in size and is what is currently shown to monitor. (~16MB)

That is weird, because the articles I've found say SSAA, RGSSAA and OGSSAA are different names for the same thing, and it is SGSSAA that is the compromise.

SSAA just means that samples are super sampled. (all sub-samples get their own texture/shader calculations)
All these AA methods are variations of it.

Thus the difference is in the locations of those sub-samples and all these AA methods should be identical in their performance. (they do the same amount of work. (amount of samples within pixels))

OG = ordered grid a simple two dimensional array. (also a regular grid AA if I understand your posts correctly, first time I heard the term. (if you men another pattern please tell.))
RG = rotated grid, samples within pixel are rotated depending on how many samples. (2x 45 degrees, 4x and above 22.5 degrees.)
SG = sparse grid which is identical to MSAA sample pattern which is superior. (no weak angles.)

It's more compatible and runs faster.

For same sample count it's just as fast.
Thing is that 2x2 is as fast as 4x SGSSAA and has worse sample pattern.
3x3 has 9 samples so it is slower than 8xSGSSAA with worse sample pattern.
4x4 .. 16 samples, looks gorgeous not usable in many games.

 

[bystander36]

SSAA just means that samples are super sampled. (all sub-samples get their own texture/shader calculations)
All these AA methods are variations of it.

Thus the difference is in the locations of those sub-samples and all these AA methods should be identical in their performance. (they do the same amount of work. (amount of samples within pixels))

OG = ordered grid a simple two dimensional array. (also a regular grid AA if I understand your posts correctly, first time I heard the term. (if you men another pattern please tell.))

RG = rotated grid, samples within pixel are rotated depending on how many samples. (2x 45 degrees, 4x and above 22.5 degrees.)

SG = sparse grid which is identical to MSAA sample pattern which is superior. (no weak angles.)

For same sample count it's just as fast.
Thing is that 2x2 is as fast as SGSSAA and has worse sample pattern.
3x3 has 9 samples so it is slower than 8xSGSSAA with worse sample pattern.
4x4 .. 16 samples, looks gorgeous not usable in many games.

I've heard 3 different analysis of the same names.

Btw, RG can mean "regular grid" as well. It doesn't have to mean Rotated. And by all the searches I've looked at, rather than forums, SSAA is the highest form, and the others are either a different name, or compromised versions, but sometimes compromises can be better if you can apply higher levels of it for the same performance.

That said, you may be right about 8xSGSSAA looking better than 2x2 SSAA. I use 4x4 in Dragon Age Origins, unless I'm in 3D Vision, then I use 3x3.

 

[bystander36]

Sorry for continuing to go off topic, but I've finally got SGSSAA to work. That Geforce SSAA Tool doesn't appear to work, but I have been able to enable it by settings Transparencies to SGSSAA and MSAA at 8x, which was what was suggested in the other link I posted where I found it talked about. I have it set for Dragon Age: Origins, the one game I'm currently playing that really needs SSAA applied, because the armor textures look horrible without it.

I have gotten what appears to be equal IQ from 8x SGSSAA and 3x3 SSAA. I'm going to see if I can see a performance difference now.

I have tested it in game, and found up to 5 FPS difference in my limited testing. Of course I'm getting at worst, 88 FPS, so it wasn't much of a difference, but the 8x SGSSAA is maybe 5% faster, but there is 8x vs 9x sampling, which might make the difference. It's pretty cool either way.

 

[cmdrdredd]

I generally use SGSSA 4x and 4x msaa.

 

[BFG10K]

Thus the difference is in the locations of those sub-samples and all these AA methods should be identical in their performance. (they do the same amount of work. (amount of samples within pixels))

The workload is the same, but the performance isn't. OGSS is a "software" approach while RGSS/SGSS can leverage some of the optimizations made for MSAA at the hardware and driver level (e.g. generating and fetching the samples).

Also from Fermi onwards, nVidia has extra hardware acceleration for the jitter/blend operations needed for this scheme. I'd expect AMD's hardware from Juniper/Cypress onward has something similar.

So to put it simply, 4xRGSS is faster than 4xOGSS (for example), and I've personally confirmed it in numerous tests.

 

[BFG10K]

SGSSAA is only available in Nvidia Inspector under "Antialiasing - Transparency". It is not available under "Antialiasing - Setting". Under "Antialiasing - Setting" there is 2x1, 1x2, 2x2, 3x3, and 4x4 Supersampling. There is no SGSSAA option.

That's because it's the same setting that controls it. Normal TrAA is triggered when the pixel passes the alpha test. With the other setting the driver makes all of the pixels pass it.

There's no reason to use the X modes unless you come across some fringe IQ or compatibility case. These modes have an extremely poor EER and thus low a IQ/performance ratio.

They're actually based on ancient hacks from the original Radeon/GeForce days that were hastily added to compete with 3dfx.

 

[Pottuvoi]

That said, you may be right about 8xSGSSAA looking better than 2x2 SSAA. I use 4x4 in Dragon Age Origins, unless I'm in 3D Vision, then I use 3x3.

I meant 4xSGSSAA, somehow forgot the multiplier there.
Will fix in previous post.

The workload is the same, but the performance isn't. OGSS is a "software" approach while RGSS/SGSS can leverage some of the optimizations made for MSAA at the hardware and driver level (e.g. generating and fetching the samples).

Yes, I would suspect as much.