New D3D FSAA Viewer

[KimB]

I still can't get the damn TemporalAAFrameThreshold setting to kick in.

The threshold appears to turn off Temporal AA when the framerate is below the value.

For example, if you run the FSAA viewer with a Temporal AA threshold of, well, anything, you'll get per-frame changing AA samples for each of the 500 or so frames per second. If you go to single-frame mode, the next 3 (for 3xT) samples will be changing, but after that it will revert to "normal" AA. Go back to realtime mode and the samples will quickly return to 3xT.

As a side note, this method of temporal AA really isn't the best in my opinion. I believe that each one of the sample patterns used should individually be a sparse-sampled pattern. This would, of course, limit 2x and 4x to 2xT, but I don't think that's really a bad thing, as lower numbers of samples will produce more problems with temporal AA.

The benefit of having each individual temporal AA pattern being a sparse-sampled pattern is basically that the overall edge quality will not change much for any edge when the patterns are switched, resulting in a reduction in "flickering" and much better quality for screens with lots of motion.

 

[DemoCoder]

I almost can't stand to look at CRTs anymore after using LCDs for so long. Even at high refresh rates, the best CRTs look dim and shimmery to me. (I used to have a Cornerstone 21" at 100+Hz 1600x1200)

And of course, the amount of space and weight that CRTs take up really annoy me. I banished them from my house.

I just bought what will be my last CRT last year. I wanted a LCD but I just couldn't justify twice the cost for a comparable size LCD.

If you can't justify the costs, don't buy exactly the same size. For some reason, 19" and 20" LCDs are alot cheaper than 21" LCDs. I bought an 18" Dell UltraSharp for my wife that has more displayable area than an 18" CRT and got it for only $280 using a deal at Dell.

 

[trinibwoy]

Will try this when I get home but from the previous posts it seems that this topic is mega-subjective.

However, I'm not clear on something. Based on the descriptions given here and not any first hand viewing I can't understand how IQ is better in a fast paced game like UT2004.

Isn't the free blending effect of TemporalAA dependent on the same frame being viewed with each temporal grid in the sequence. Let's assume we're using 4xT2. If I use the first 4x temporal grid for frame 1 and the other 4x grid for frame2 where all of the pixels have been shifted how am I getting anything more than 4x? There is no blending effect there to speak of. It's quite obvious that it's great for still objects but the guys touting great IQ in UT2004, what is going on onscreen during these tests? My doubts come as a result of:

1) If it's moving fast the free blending doesn't occur
2) If it's moving fast is the difference between 4x and 4xT2 really that discernable?

I have to go home and check this out for myself. Unless I end up playing Warcraft till 3am like last night

 

[Joe DeFuria]

1) If it's moving fast the free blending doesn't occur
2) If it's moving fast is the difference between 4x and 4xT2 really that discernable?

That's basically correct.

However, the most "noticable" artificats arise when you have "slow moving" edges, which makes that "crawling" (or popping/shimmering for small polys) effect. For example, a driving game or flight sim....you'll notice the horizon "aliasing" before you notice a "jaggy" on a vehicle edge blowing by you.

So in other words, yes....the faster "moving" an object is, the less of an impact temporal AA has. But at the same time, the faster an object is, the less you notice aliasing in general.

 

[Moose]

2) If it's moving fast is the difference between 4x and 4xT2 really that discernable?

Well, what I notice with regular AA while playing, is the crawling effect on edges as the angles change when you move around in game.

With temporal AA on, the crawling effect is almost totally gone but is replaced with a small bit of shimmering on the edges.

It looks much better to me.

 

[MuFu]

trinibwoy, you're right - if you read this thread and others like it, it was an area of speculation before T-AA was unveiled and has been commented on a lot since.

Plus the "baseline" FSAA quality you get is actually worse than the normal modes because the component patterns aren't sparse (as Chalnoth mentioned above).

Edit - doh, beaten to it.

 

[trinibwoy]

If you can't justify the costs, don't buy exactly the same size. For some reason, 19" and 20" LCDs are alot cheaper than 21" LCDs. I bought an 18" Dell UltraSharp for my wife that has more displayable area than an 18" CRT and got it for only $280 using a deal at Dell.

Just a guess but wouldn't a $280 18" LCD suck for gaming? I doubt any one who uses their machine primarily for gaming can justify the purchase of an LCD. I want to get one but the cost/benefit ratio is horrible at the moment. I use an LCD at work and I guess that's where I need the LCD benefits anyway.

On another note, does tearing only occur when your frame rate is higher than you refresh rate since another frame is in the buffer midway through a screen refresh. Or can it happen the other way around too?

 

[MuFu]

On another note, does tearing only occur when your frame rate is higher than you refresh rate since another frame is in the buffer midway through a screen refresh. Or can it happen the other way around too?

The back/front buffer swap is still totally out of sync with the vertical refresh cycle so it can happen the other way round, yes (i.e. the frame rate might be lower, but that doesn't mean a swop won't take place while the front buffer is being drawn).

 

[dizietsma]

It supports two patterns: ordered grid and rotated grid, which can be switched every other frame for "2T" in the parlance being used here.

But moreover, programmable sampling positions are not required. The effect can also be achieved via pixel center offset.

How do you know other IHVs already didn't implement this along time ago but decided against exposing it because it didn't really improve IQ in most cases, but made it worse?

They might be tempted to impliment it now ?

I think this AA is down to Dave here at Beyond3d, he asked Ati for something to make other reviewers make more mistakes and also allow him to write 100 page reviews.

 

[digitalwanderer]

I like the term "crawling" for the weirdness sometimes seen with the temporal AA rather than "shimmer", "shimmer" describes something entirely different...at least to me.

"Crawl", I like it.

I've only seen it when I'm really looking for it and then only under certain circumstances, my experiences with this AA so far has been nothing but good!

I'm STILL bugged by getting a constant 100fps in everything now, I don't get why I'm not getting a fraction of my refresh rate using AA with v-sync on....but I'm luuuuuuving it. 8)

 

[trinibwoy]

On another note, does tearing only occur when your frame rate is higher than you refresh rate since another frame is in the buffer midway through a screen refresh. Or can it happen the other way around too?

The back/front buffer swap is still totally out of sync with the vertical refresh cycle so it can happen the other way round, yes (i.e. the frame rate might be lower, but that doesn't mean a swop won't take place while the front buffer is being drawn).

That's what I thought. Saw a couple posts about high refresh rate monitors helping with tearing though. Myth?

 

[BRiT]

I had an 18" Dell UltraSharp LCD for about a year. This is amongst the top 18" LCDs out there. I picked it up for $110. [Sold off the Dell system on Ebay to drop the cost of the LCD.] It does not suck for gaming. No ghosting at all. My only complaint, if you can call it that, is the color contrast/range isn't as large as I was used to. It was extremely usable for gaming. My system 17 months ago was an AMD Desktop Barton @ 2.4Ghz, ATI 9700 Pro.

I'm now using a 19" Planar LCD for about 5 months now. This is amongst the top 19" LCDs out there. I picked it up for under $400 straight from Dell without buying anything else [I didn't have to sell off a system on Ebay]. It's improvement over the Dell 18" UltraSharp is a bit more color contrast/range. No ghosting at all. It's extremely nice to use for gaming. My system is now an AMD Mobile Barton @ 2.65Ghz, ATI 9800 Pro.

 

[MuFu]

On another note, does tearing only occur when your frame rate is higher than you refresh rate since another frame is in the buffer midway through a screen refresh. Or can it happen the other way around too?

The back/front buffer swap is still totally out of sync with the vertical refresh cycle so it can happen the other way round, yes (i.e. the frame rate might be lower, but that doesn't mean a swop won't take place while the front buffer is being drawn).

That's what I thought. Saw a couple posts about high refresh rate monitors helping with tearing though. Myth?

I think that's only because you see "torn" raster cycles for less time. Not an expert though - maybe one of the coneheads will chime in.

MuFu.

 

[dr3amz]

2) If it's moving fast is the difference between 4x and 4xT2 really that discernable?

Well, what I notice with regular AA while playing, is the crawling effect on edges as the angles change when you move around in game.

With temporal AA on, the crawling effect is almost totally gone but is replaced with a small bit of shimmering on the edges.

It looks much better to me.

what he said the effect at 2xT looks like 4-6x aliasing, in 2xAA you get the crawling effect - try it on 3dmark2k1 - you notice it immediately on the roof.

however with 2xTemp its practially gone.

and again - with low fps i get NO shimmering etc with 2xTemp AA - i do with 3xTemp - but not with 2xTemp. so i dont know what is different on my rig and that other guys, but its no BS 8)

i'll try and make a video clip (try)

so just to confirm - what you're saying is that if i play farcry and am only getting 36fps say - that i should see shimmering?

 

[MuFu]

so just to confirm - what you're saying is that if i play farcry and am only getting 36fps say - that i should see shimmering?

Yes. You can't get 36FPS with vsync enabled anyway - if you see values other than 100, 50, 33, 25, 20, 17 etc then they're just the FPS counter displaying a "lazy" average.

BTW - I had some success making a 60FPS video but the filesize was huge (even using BINK), and of course won't work for people with different refresh rates.

 

[KimB]

On another note, does tearing only occur when your frame rate is higher than you refresh rate since another frame is in the buffer midway through a screen refresh. Or can it happen the other way around too?

Actually, the worst tearing will occur when the framerate is very close to your refresh rate (or an integer multiple thereof).

That is, if the refresh rate and frame rate are far apart (from each other or a small integer multiple), you'll get "tearing" lines appearing anywhere on the screen, in seemingly random locations.

But if your refresh rate and frame rate are nearly in sync, you'll get one line that slowly moves up and down the screen, something that is much more noticeable. Of course, with most games, such a thing could never happen, since the framerate varies pretty wildly from frame to frame. But this effect does crop up when, for example, you use a video camera to take pictures of a monitor (the monitor will appear to flicker heavily on the video camera due to this effect: changing the refresh rate can remove this flickering).

 

[dr3amz]

just ran farcry at 1024*768@120hz with 2xtempaa and i have to say - WOW (again) raising my monitor to 120hz really helped fps, but the tempaa is simply stunning - and this was just at 2xT + 2xAA

its just so much fun to play with this shit

 

[Randell]

I'm running a 9700Pro with a 15" 1025x768 LCD monitor at 60hz

It works well for me.

Far Cry looks great at 2x/2T - especially outdoors where all foliage get cleaned up. No shimmer/crawl on hill edges. Some shimmer on boxes/crates. Indoors no shimmer I could see (cooler area)

Op Flash - Like Far Cry (except foliage is almost exclusively alpha textures). Hill edges. vehicle edges are great.

Gothic 2 - in outdoor areas again works well. In town, rooftops shimmer like crazy.

UT2003 - some small shimmer if you stop and stare. Don't notice if you actually play teh game!!

 

[dr3amz]

tried to take a pic - you can see how its aliasing if you zoom in (big pic)

but basically, show the image full screen and thats what its like to play with, obviously sharper - but not blocky! 8)

http://homepage.ntlworld.com/dr3amz/tempaacam.jpg

suggest you save picture then open in windows picture browser (or similar prog) so you can full screen it to replicate what i see in game)

but you can see that its doing far more than 2xAA (this is 2xTempAA with 2xAA in settings) check the rope bridge at the top for example

screenshot from game using same settings:

http://homepage.ntlworld.com/dr3amz/tempaa.jpg

basically because of the way it renders, this image shows a few more jaggies - i've found the rough edges show more when stationary and disappear when moving?

check out the awesome aliasing on the wire fencing here - basically when moving, even very slowly, you get that same quality across the whole image

 

[dr3amz]

http://homepage.ntlworld.com/dr3amz/tempaacam.jpg

if you zoom into the edges of the tree you will imediately notice the amount of aliasing happening - far more many depths than with standard 2xAA.

hope that makes some sense (this was taken with 3xTx2AA@1024*768@120hz - flickering was barely noticable, is unoticable at 2xT) but image Q is easily 6xAA with 3xTx2AA).

still hard to capture the method as good as it looks running real time in front of you (better than any pic) use FSAAViewer for the simplest version what happens to the lower left edges in this, are what you get in game basically.