HL2 AA issue a thing of the past?

andypski said:
Just FYI.

If anyone wants to see some good examples of the kinds of artifacts that can be generated by texture sampling falling outside the covered subsamples then the game Max Payne shows them fairly clearly if you turn on AA. (If I recall correctly)

OK, decided to try it and with MP 1.05, whether I enabled AA within the game, setting the Cat 3.6's to Application preference with 4x FSAA turned on, AF enabled; or whether I turn FSAA off in the game and force 4xFSAA/16x AF through the control panel--his hair looks fine. In fact, I restarted with the Cpanel set to AP and turned off FSAA in the game with AF enabled in game--and his hair looks just the same--only it's aliased, of course.

What should I be seeing in the way of artifacts to illustrate?
 
WaltC said:
What should I be seeing in the way of artifacts to illustrate?

Normally what you would see is some sparking around the edges of the model - as you turn the camera around and look at different angles.

Try with 2x AA. I believe that statistically the error gets less visible as the number of subsamples increases, since each subsample contributes less.
 
andypski said:
WaltC said:
What should I be seeing in the way of artifacts to illustrate?

Normally what you would see is some sparking around the edges of the model - as you turn the camera around and look at different angles.

Try with 2x AA. I believe that statistically the error gets less visible as the number of subsamples increases, since each subsample contributes less.

Tried with 2x FSAA and experiemented with AF on and off--still no artifacts of any kind that I can see. At this point I'd guess what you saw was probably a driver bug which has since been fixed...Next I'll try 6x and see what happens...

Thanks for reminding me about this game, though--I'm quite enjoying playing it through again. Replaying games is a lot like watching movies you haven't seen recently--you forget a lot of things in the interim....;)
 
WaltC said:
Tried with 2x FSAA and experiemented with AF on and off--still no artifacts of any kind that I can see. At this point I'd guess what you saw was probably a driver bug which has since been fixed...Next I'll try 6x and see what happens...

I'm fairly sure that I would know about that if it was the case... ;)

Try turning AF on as well as AA and see if you can see anything then.

It is a good game :)
 
andypski said:
I'm fairly sure that I would know about that if it was the case... ;)

Try turning AF on as well as AA and see if you can see anything then.

It is a good game :)

Already done...tried every conceivable combination of 2x-6x FSAA modes on/off & AF on/off, both from the control panel and from within the game. No artifacts that I can see...even tried different levels of AF from the control panel, plus setting AF in the control panel and in the game itself (where you get a choice between Trilinear and AF)--just can't make any artifacts visible. Now, I haven't tried 16-bit integer display, of course....but, not going to....;) Only apparent difference I can see between between FSAA on and off in the game, regardless of what kind/level filtering I'm using at the same time, is the aliasing.

Cat 3.6, R9800P, WinXP, 1152x864x32.
 
Right - just retried it myself - the artifacts are definitely there, although they show up better in some areas than others - might be to do with the game's lighting system.

Just stand still with the camera in one place and watch as he turns his head - you should see occasional flashes of light pixels appearing around his hair as his head moves. If you don't see them then move to another area and try again. Try and get the outline of his head against a dark background so that you can see the flashes when they happen.

The effect can be quite subtle, so you will need to look carefully.
 
andypski said:
Right - just retried it myself - the artifacts are definitely there, although they show up better in some areas than others - might be to do with the game's lighting system.

Just stand still with the camera in one place and watch as he turns his head - you should see occasional flashes of light pixels appearing around his hair as his head moves. If you don't see them then move to another area and try again. Try and get the outline of his head against a dark background so that you can see the flashes when they happen.

The effect can be quite subtle, so you will need to look carefully.

Mhhhmmm.....what's your 3D hardware? I've played through the game up until the scene in the bar in chapter four (where I keep getting blown away), and haven't noticed the effect you describe while playing. What I did while testing for the effect last night is to run the tutorial and move the camera quite a bit both close in and at a distance, in both dark and lighted areas, and also paused in the game in which case the camera does a 360 around him--couldn't see it there or in the game so far. Also what's you FSAA level and AF, etc.?

Heh...;) It's not that big a deal deal obviously, if I have to work so hard to try and see the effect--but you've got my curiosity aroused. I'll get back a final time and let you know whether I could ever get the artifact to reproduce on my system...

Edit: Well, I even tried turning off DX texture compression and offsetting the pixel center, and using 16-bit textures...still no go--can't see any artifacts. In the tutorial, after you open the van doors and get the shotgun and blow the padlock off of the door, you can proceed through the door into an alleyway and at the 90-degree junction the corner there is almost pitch black--had the camera over his head and watched him move it for couple of minutes--no luck. Then ran down the alley and jumped up on the crates and over the wall and ran under the street light where I repeated--still no luck. So I've done all I can to recreate this artifact--just isn't going to happen (which I'm not disappointed about.) Heh--if the MSAA artifacts in HL2 are no worse than this I'm not going to mind them at all....;) Be a lot worse without FSAA, certainly. But maybe the HL2 situation is different--guess we'll see when Valve releases its FSAA artifact movie one of these days.
 
OpenGL guy said:
WaltC said:
Mhhhmmm.....what's your 3D hardware?
I've seen this on GeForce 4 4200 and 4600, Radeon 9700 and Radeon 9800.

Don't know what to say...I've done handstands trying to create a condition for it to manifest...Can't say I'm unhappy about not seeing it, though. All I can say is that if the effect is so subtle I can't see it with my eyes two-inches from screen for 30 seconds at a time (I've really given this a decent try, you know), in both dark and lighted areas, from a variety of camera positions, and I haven't noticed it at all while playing the game--well, I'm not too worried about it...;)
 
WaltC said:
Mhhhmmm.....what's your 3D hardware? I've played through the game up until the scene in the bar in chapter four (where I keep getting blown away), and haven't noticed the effect you describe while playing. What I did while testing for the effect last night is to run the tutorial and move the camera quite a bit both close in and at a distance, in both dark and lighted areas, and also paused in the game in which case the camera does a 360 around him--couldn't see it there or in the game so far. Also what's you FSAA level and AF, etc.?

Heh...;) It's not that big a deal deal obviously, if I have to work so hard to try and see the effect--but you've got my curiosity aroused. I'll get back a final time and let you know whether I could ever get the artifact to reproduce on my system...

This is on a 9600 Pro with 4xAA and no AF.

As to whether it's a big deal or not, the problem is really that the potential error for any given edge pixel is actually unbounded (well, actually it's bounded by the maximum error in the texel sample value - which is the full range). ie. you can sample a white texel value when you should be sampling a black one. The fact that this doesn't happen too much in Max Payne is a function of the texture data - with different textures you could get much larger errors.
 
It's there all right. Though you may not see it, if the camera is not in the right place.

I had the camera pretty much directly behind Max. I was running at 640x480 with 2x FSAA and 16x Anisotropic. Any higher res and they might be really hard to spot.

Anyway, around the edges of Max's hair there would be the occasional odd white pixel appearing. Was it obvious? No not really, but it's there.
 
Colourless said:
It's there all right. Though you may not see it, if the camera is not in the right place.

I had the camera pretty much directly behind Max. I was running at 640x480 with 2x FSAA and 16x Anisotropic. Any higher res and they might be really hard to spot.

Anyway, around the edges of Max's hair there would be the occasional odd white pixel appearing. Was it obvious? No not really, but it's there.

I'm sure you guys are seeing something no doubt, but, hey, if this effect is so insignifigant you've got have gigantic pixels and precise camera angles & lighting to even see it occasionally, it just strikes me that the lack of FSAA at 640x480 would be far more noticeable and objectionable in terms of aliasing artifacts than this. It's difficult to imagine Newell is talking about something like this, as even in the res the current HL2 movies are recorded in the aliasing artifacts are easily seen--but I guess we'll see what he's talking about relative to HL2 when he releases an FSAA sample movie. While trying to see this effect myself I wondered if people were seeing flashes of the lighter-toned highlights in max's hair from certain angles as he turns his head (which bobs and weaves like a water bird's at times)--anyway, thanks for the info.
 
WaltC said:
I'm sure you guys are seeing something no doubt, but, hey, if this effect is so insignifigant you've got have gigantic pixels and precise camera angles & lighting to even see it occasionally, it just strikes me that the lack of FSAA at 640x480 would be far more noticeable and objectionable in terms of aliasing artifacts than this. It's difficult to imagine Newell is talking about something like this, as even in the res the current HL2 movies are recorded in the aliasing artifacts are easily seen--but I guess we'll see what he's talking about relative to HL2 when he releases an FSAA sample movie. While trying to see this effect myself I wondered if people were seeing flashes of the lighter-toned highlights in max's hair from certain angles as he turns his head (which bobs and weaves like a water bird's at times)--anyway, thanks for the info.

No - there are definite rendering artifacts, it's not to do with seeing highlights in his hair.

As I said above, the problem is that the magnitude of the error can be very large, and with different texture data it can be a lot more visible.

I've got a better example for you to try that shows a similar effect - the 3DMark03 pixel shader 2.0 test. You should see lots of white flashes in this test around the edges of the objects if you turn AA on.
 
andypski said:
No - there are definite rendering artifacts, it's not to do with seeing highlights in his hair.

As I said above, the problem is that the magnitude of the error can be very large, and with different texture data it can be a lot more visible.

I've got a better example for you to try that shows a similar effect - the 3DMark03 pixel shader 2.0 test. You should see lots of white flashes in this test around the edges of the objects if you turn AA on.

I don't want you to think that I'm disagreeing with the basic statement, because I'm not. I was just stating that the example provided of Max Payne wasn't something in which I could see such artifacts. I have no reason to doubt Gabe or you...;)

I'm still not clear on centroid texturing as defined by the API. I've read various reports that it isn't supported under DX9, and others that it is. Of course, if it isn't supported under DX9, my next question is obvious as to its inclusion in certain DX8-9 "compliant" software. What's the story? (I looked back at my posts and saw they might be considered as an argument--but honestly I didn't intend that--I'm not as clear as I'd like about the centroid situation.) Thanks, if you time to enlighten me a bit about it...;)
 
andypski said:
I've got a better example for you to try that shows a similar effect - the 3DMark03 pixel shader 2.0 test. You should see lots of white flashes in this test around the edges of the objects if you turn AA on.
It sounds like a z-compression problem, like the edges aren't being properly-detected and taken care of. If this is the case, it might be a flaw in the hardware, or it might be a shortcut taken by the drivers, or it may just be a bug in the drivers (depends on how much control over the hardware the drivers actually have...).
 
Chalnoth said:
andypski said:
I've got a better example for you to try that shows a similar effect - the 3DMark03 pixel shader 2.0 test. You should see lots of white flashes in this test around the edges of the objects if you turn AA on.
It sounds like a z-compression problem, like the edges aren't being properly-detected and taken care of. If this is the case, it might be a flaw in the hardware, or it might be a shortcut taken by the drivers, or it may just be a bug in the drivers (depends on how much control over the hardware the drivers actually have...).
Sure, the exact same issue on GeForce FXs and Radeons? No. This is the same issue as has been discussed with Max Payne.
 
Back
Top