Cost of 3dfx RGSS AA on current hardware?
- Thread starter Reverend
- Start date
Have you ever used V5? :smile:
Yes, why?
That's an odd question. The theoretical cost of an algorithm doesn't change with the existence of an implementation.
In practice, depending on implementation details, the cost is usually between rendering n frames + downsampling and rendering at n times the resolution + downsampling.
Silent_Buddha
Legend
My memory might be sketchy on this, however...
I thought 3dfx's RGSS AA was basically rendering multiple versions of the same scene, rotating each one 45(?) degree's and then blending it back together into 1 scene. Rather than rendering to a higher resolution and then downsampling. Each VSA100 chip was responsible for rending each scene if there were multiple, however in the case of 4x RGSS, each chip of the V5 5500 has to render each scene twice.
This resulted in a fairly linear performance degradation but absolutely gorgeous visuals.
Considering that this method doesn't render to a higher resolution nor does it render a different frame (all frames are identical just rendered and rotated) would it be possible to just store 4 "copies" of each rendered frame in memory, rotate it, and then blend it?
If that is the case then with enough memory bandwidth (and memory) wouldn't it be possible to have a relatively small performance hit?
Regards,
SB
I thought 3dfx's RGSS AA was basically rendering multiple versions of the same scene, rotating each one 45(?) degree's and then blending it back together into 1 scene. Rather than rendering to a higher resolution and then downsampling. Each VSA100 chip was responsible for rending each scene if there were multiple, however in the case of 4x RGSS, each chip of the V5 5500 has to render each scene twice.
This resulted in a fairly linear performance degradation but absolutely gorgeous visuals.
Considering that this method doesn't render to a higher resolution nor does it render a different frame (all frames are identical just rendered and rotated) would it be possible to just store 4 "copies" of each rendered frame in memory, rotate it, and then blend it?
If that is the case then with enough memory bandwidth (and memory) wouldn't it be possible to have a relatively small performance hit?
Regards,
SB
Silent_Buddha: Yes, but only the grid looked rotated. In fact, the scene wasn't. Each scene was rendered multiple times and each vesion was sampled with slight offset. It was great, because the pattern was fully programmable, adjustable from frame to frame, frame-buffer resolution wasn't affected, so this method was 100% compatible (oversampling from ATi/nVidia wasn't). It didn't comsume additional fillrate - blending was done in RAMDAC... And the IQ was awesome - with adjusted mip-map LOD, textures were sharp, shimmer-free, alpha textures looked great, edges too. And in addition, it reduced dithering artifacts / increased output color-depth (when using 16bit rendering) to 22bit equivalent, which sometimes (especially when using glide) looked better than competitors 32bit output. (my favourite screenshot from WoT with 3Dfx RGSS 2x @16bit and w/o AF - VSA-100 didn't support it)
zsouthboy:
PowerSlide (RGSS 4x) (very old glide game
)
R7x00 4x (same as GF2)
3Dfx 4x
3Dfx 4x (adjusted)
zsouthboy:
PowerSlide (RGSS 4x) (very old glide game
)R7x00 4x (same as GF2)
3Dfx 4x
3Dfx 4x (adjusted)
NM about the link thing. I found this when I googled:
http://www.onversity.net/doc/fsaa.pdf
Written by Kristof and Dave themselves
EDIT: Thanks No-X!
http://www.onversity.net/doc/fsaa.pdf
Written by Kristof and Dave themselves
EDIT: Thanks No-X!
Last edited by a moderator:
I've heard a few people recently refer to the 3dfx VSA-100 anti-aliasing method as "super-sampling"; but would I be correct in thinking that it would be more accurate to describe it as "multi-sampling", given that there was no increase in texture resolution? True super-sampling (i.e. rendering at double the horizontal and vertical screen resolution and then down-scaling) would also have involved using a higher mip-map level, but the 3dfx method didn't.
The frames aren't identical. They are essentially rendered with different pixel centers. They may look almost entirely the same, but the tiny differences is what creates the AA. You can't just use one frame, copy that around with some shift and blend. That would just result in blur. That could also be trivially implemented in a shader on pretty much any hardware.
CouldntResist
Regular
Regarding cost, there are 2 nice properties to take into account:
1. The cost of postprocessing shaders is only payed once, since you only need to do it after the accumulation phase.
2. You get motion blur for free.
1. The cost of postprocessing shaders is only payed once, since you only need to do it after the accumulation phase.
2. You get motion blur for free.
Silent_Buddha
Legend
Thanks for clearing that up Humus, I knew I was missing something in my over simple explanation.
Of course this brings a question to my mind.
Would be it possible for any savings in rendering an image due to much of it being similar or will the whole scene always have to be rendered for each sample of a RGSS implementation?
Regards,
SB
Of course this brings a question to my mind.
Would be it possible for any savings in rendering an image due to much of it being similar or will the whole scene always have to be rendered for each sample of a RGSS implementation?
Regards,
SB
I am sure Rys is frustrated and exasperated that you had to Google that and found it elsewhere.NM about the link thing. I found this when I googled:
http://www.onversity.net/doc/fsaa.pdf
Written by Kristof and Dave themselves
EDIT: Thanks No-X!
Back on-topic. Ignore my mention of the word "per-sample" and let me re-phrase : How useable would 8xRGSS AA, as implemented by 3dfx in their VSA100 chips, be in a machine with high-end sub-systems using one of the latest single-card high-end video cards (powered by ATI and NVIDIA)?
Guys,
I think we need to get some terminology correct here. Rotated Grid Super Sampling (as opposed to sampling on a standard grid) is just a form of sparse sampling pattern. What people here seem to be really discussing is the T-Buffer which is an alternative way of implementing an accumulation buffer.
I think we need to get some terminology correct here. Rotated Grid Super Sampling (as opposed to sampling on a standard grid) is just a form of sparse sampling pattern. What people here seem to be really discussing is the T-Buffer which is an alternative way of implementing an accumulation buffer.
I don't really care what it's called (btw, is "T-Buffer" patented, Simon? [edit]Just trying to keep things light, Simon... no offence
... although I really don't know if "T-Buffer" is a patented algorithm or a copyrighted phrase...[/edit]) and I don't really want to talk about it all over again. If anyone would like to discuss the "T-Buffer" again, I'd appreciate it if it goes in another dedicated thread (which I'd be happy to participate in). I can then talk about other things the "T-Buffer" can be useful for which have nothing to do with AA.I just want to know :
Myself said:
If I added that I'd like to know how comparable 3dfx's 8xRGSS (Colourless...?) anti-aliasing quality-vis-a-vis-performance is compared to the best the current latest video hardware has to offer, I probably should start another thread, right?
Last edited: