I don't think the grainy look had anything to do with QAA. If anything MLAA would make the image sharper and if there is any grainy look, it would be increased according to the increase in sharpness.
Killzone 3 and MLAA
- Thread starter HollovVpo1nt
- Start date
A triangle smaller than a pixel. the rendered pixel needs to be an average of this triangle and the triangle behind it, which needs sampling of both surfaces. The more samples, the more accurate the representation of the subpixel-triangle's coverage.
L. Scofield
Veteran
Indeed. Using MLAA here is a step backwards.
They should replace the railing drawing method...
L. Scofield
Veteran
While certainly MLAA helps in making a sharper image, Guerilla seems very fond of blurring the screen to hell every change they get (which is about every two seconds) so QAA actually would be more suited to the graphics style.
That was one of the reasons KZ2 graphics weren't so eye-catchy for me. I think MLAA on KZ3 makes sense if you can appreciate the little details better, and since it's a pre-alpha they have time to improve the potential flaws over time.
I see MLAA as some kind of filter, like saying a Super2xSaI, Super Eagle. AdvancedMAME Scale2x, etc, but more complex in its own right.
Over time console GPUs will get more powerful so maybe the CPU won't be necessary to apply these kind of filters which are very CPU intensive.
I wonder how useful can be a CPU in the future, perhaps to apply anisotropic filtering if need be??...
Because as native resolutions increase, the level of AA graphics need to look clean decreases, which looks like simple math. :smile:
Doesn't QAA have a higher impact on the RSX VRAM bandwidth than MLAA?
Or is it only 2X MSAA that costs?
There is no way around it, atleast a way that's practical with the current hardware...from it, I mean aliasing on sub pixel edges.
Yes, because QAA uses 2 samples per pixel same as 2xMSAA.
Right. So MLAA could be useful just for shifting load around on the PS3, even if the graphics style wouldn't necessarily seem to call for it.
It seems like the next HD gen might be really boring for graphics programmers, given the fixed resolution requirements of HDTV displays.
But that's a different thread.
Right. So MLAA could be useful just for shifting load around on the PS3, even if the graphics style wouldn't necessarily seem to call for it.
It seems like the next HD gen might be really boring for graphics programmers, given the fixed resolution requirements of HDTV displays.
But that's a different thread.
Don't worry about them, they'll have 3D to worry about.
Definitely. The GOW crew called it a gain in terms of trading GPU for CPU, as has Media Molecule.
I've got a question about MSAA, actually. I believe I understand that Quincunx AA samples in a quincunx pattern, against a higher-than-display-resolution buffer.
Is QAA/2X MSAA cheaper than 4X because that high res buffer doesn't have to be at as high a resolution, or is it just cheaper because there's less sampling to the background buffer?
Does using MLAA remove the need for the extra VRAM for the high res background buffer in addition to the bandwidth for the sampling?
Is QAA/2X MSAA cheaper than 4X because that high res buffer doesn't have to be at as high a resolution, or is it just cheaper because there's less sampling to the background buffer?
Does using MLAA remove the need for the extra VRAM for the high res background buffer in addition to the bandwidth for the sampling?
Quincunx does use 2xMSAA. RSX/G7x ROPs are single cycle, so yes. 4x is costly there. For devs, the extra memory cost of the buffers is likely an issue on console.
4x MSAA on PS3 is simply costly due to the extra cycle for the ROPs and the extra memory (<256MB).
MLAA shifts things somewhat as it's more edge-related. By that I mean... more edges to process = more cost.
Edit: and now it's time to flee.
4x MSAA on PS3 is simply costly due to the extra cycle for the ROPs and the extra memory (<256MB).
MLAA shifts things somewhat as it's more edge-related. By that I mean... more edges to process = more cost.
Edit: and now it's time to flee.
Some people complained they had headache when looking at blurred stereoscopic 3D images. When their eyes focus on certain area, the brain expects the image to become sharper. But if it remains blurred, I was told some gamers would experience headache.
The sharper (and higher quality) MLAA treatment may make sense in this context. Then in 2D KZ3, apply more post processing as required, like L.Scofield suggested. At least DoF needs to be back.
The sharper (and higher quality) MLAA treatment may make sense in this context. Then in 2D KZ3, apply more post processing as required, like L.Scofield suggested. At least DoF needs to be back.
Its strange but I don't actually notice dynamic DOF in KZ2 while playing. Whenever I use the iron sight I should get a depth of field effect with the out of focus areas being blurred subtly but that certainly wasn't the case in KZ2. The only place where I notice something like DOF is the blur on teh gun itself. I did saw DOF in cutscenes though.
Borderlands has thrown up issuse with DOF for me. You can be shooting at someone and the engine decides they are out of focus. Suddenly they and the damage you deal gets blurred into a mess. I think this has caused some eyestrain as I try to focus on an out-of-focus target. I don't recall any issues with U2 though, so it's probably just a poor implementation failing to set the right focal distance.
The DoF in KZ2 seems subtle:
http://www.guerrilla-games.com/publications/dr_kz2_rsx_dev07.pdf
http://www.guerrilla-games.com/publications/dr_kz2_rsx_dev07.pdf
Billy Idol
Legend
Thanks!!
