Halo 3 engine upgrade analysis (for ODST)

[swaaye]

With AAA titles using occasionally using bilinear and rarely using AA, I think some guesses can definitely be made about what Xenos can and can't do well regardless of what was marketed in tech write ups early on.

It is a curious thing that it would have such limitations considering ATI's older desktop GPUs such as R300 can do low AA and AF rather easily, let alone the desktop GPUs that were prevalent when Xenos launched. Hell, you can go back to R100, with its goofy 16X AF on only bilinear filtering that had little speed impact and see better texture quality than various Xbox 360 games.

 

[Rangers]

With AAA titles using occasionally using bilinear and rarely using AA, I think some guesses can definitely be made about what Xenos can and can't do well regardless of what was marketed in tech write ups early on.

It is a curious thing that it would have such limitations considering ATI's older desktop GPUs such as R300 can do low AA and AF rather easily, let alone the desktop GPUs that were prevalent when Xenos launched. Hell, you can go back to R100, with its goofy 16X AF on only bilinear filtering that had little speed impact and see better texture quality than various Xbox 360 games.

Isn't this stretching it a little bit? Isn't AA more common in 360 multiplatform games than the competition? And AF, I'm not sure but it seems to be commonly deficient on both consoles, not just 360. And I dont imagine 7800GTX has any problems with it on PC, so it's other factors at play.

 

[swaaye]

Actually G70 does have performance problems with AF relative to ATI's hardware. Don't you remember all the cheating they did, causing texture swimming and aliasing unless you set the driver to HQ mode (which is of course measurably slower)? I have a 7800Go GTX in my 'ol Dell Inspiron and have seen it first hand in various games. These are image quality problems not seen with R3x0 or later ATI parts on default driver settings. Sure they do their own filtering optimizations (Cat AI), but I've never noticed it.

When I think of 360 image quality, I think of all the claims made about nearly-free AA due to tiling. I also look at the bullshots. And then I go out and buy a hyped AAA game like Forza 2 and see bilinear filtering with ugly mipmap boundaries right on the road, and see the framerate take a major dive with minimal AA. DIRT runs AA full-time but its framerate seems to feel it rather noticeably.

Those are the only examples I can cite right now offhand cuz I don't actually own a 360 (used a friend's for a year when he was in China). I played a lot of Gears, Mass Effect, DIRT, and Forza 2. Texture pop-in and DVDROM noise are other traits that stick clearly in my mind .

I can't wait for the day when 360 is such old news that the devs will talk about the hardware. I'm sure that there are tons of issues with it, considering its rather unique design compared to the other unified GPUs. And with how R600 didn't turn out so well, especially with respect to AA.

Anyone see bilinear in recent games?

 

[Boxercide]

Isn't this stretching it a little bit? Isn't AA more common in 360 multiplatform games than the competition? And AF, I'm not sure but it seems to be commonly deficient on both consoles, not just 360. And I dont imagine 7800GTX has any problems with it on PC, so it's other factors at play.

Actually, a common note i see on grandmaster's videos is that the PS3 has better AF. Not saying every game or blah blah, just for whatever reason, AF is implemented on PS3 with seemingly more regularity.

Off the top of my head, Oblivion springs to mind.

 

[Aeoniss]

^COD4 as well.

Along with noted exclusives like KZ2 etc.

 

[NRP]

With AAA titles using occasionally using bilinear and rarely using AA, I think some guesses can definitely be made about what Xenos can and can't do well regardless of what was marketed in tech write ups early on.

I'm inclined to agree with this. The proof is in the games, and . . . well . . . we haven't seen a lot so far in the AA and AF dept on 360 games. Devs have talked at some length about the AA situation, but I don't recall much talk about (why there seems to be a lack of) AF in 360 games.

 

[Iron Tiger]

I'm inclined to agree with this. The proof is in the games, and . . . well . . . we haven't seen a lot so far in the AA and AF dept on 360 games. Devs have talked at some length about the AA situation, but I don't recall much talk about (why there seems to be a lack of) AF in 360 games.

I was reading a flamebait comment on why Metal Gear Rising will look better on the 360. "...the 360 renders textures better." It took a great deal of restraint for me to not spout off about how the PS3 renders textures better because it uses AF, so even if the 360 has more memory for higher res textures, you might only notice the difference in the rare instances when they're up close and parallel to the camera. I feel like the bigger person for not having said that.

I've run 16xAF on every game with every generation of hardware since I turned red with the Radeon 8500, so I'm perplexed why the 360 seems incapable of using AF.

 

[Rangers]

Actually, a common note i see on grandmaster's videos is that the PS3 has better AF. Not saying every game or blah blah, just for whatever reason, AF is implemented on PS3 with seemingly more regularity.

Off the top of my head, Oblivion springs to mind.

He also said AA in the original comment I was responding to, which the Xenos seems more proficient at. So right away, 50% of my comment is being ignored, in favor of the 50% easier to skewer?

Also I'm not really sure about the AF thing. Without some sort of evidence beyond anecdotes.

Also I think his comment misses the point, as beloved as AA and AF are at B3D, most people probably dont notice, and I doubt theyll ever be standard on consoles. Just like 60 FPS (or even locked 30) will never, ever be standard, because better graphics are more important (save for classes that demand 60, like racers).

I guess if AA and AF are that important to you, you should stay on (much more expensive) PC. Although I noticed AF doesnt play nice with Crysis (you can either have AF or POM not both). It's typically benchmarked without AA and AF, on the highest end $600 cards and $1000 CPU's.

 

[Rangers]

While we're on this subject, can anybody tell me an easy way to ascertain AF by eyeballing a game?

I can see tearing and AA (obviously). But I really cant tell AF. I've seen examples, but they usually involve grid patterns that may not be in a given section of a game. Is there a way to tell even from textures without a grid pattern?

 

[homerdog]

Also I think his comment misses the point, as beloved as AA and AF are at B3D, most people probably dont notice, and I doubt theyll ever be standard on consoles. Just like 60 FPS (or even locked 30) will never, ever be standard, because better graphics are more important (save for classes that demand 60, like racers).

This is simply not true. Most people can't tell you why the game with AA and AF looks better, but they can surely tell the difference. I've done blind tests in HL2 on my technology illiterate friends and every one of them chose AA+AF. They just couldn't say why. Lots of comments like "I don't know it just looks better".

 

[Boxercide]

He also said AA in the original comment I was responding to, which the Xenos seems more proficient at. So right away, 50% of my comment is being ignored, in favor of the 50% easier to skewer?

Also I'm not really sure about the AF thing. Without some sort of evidence beyond anecdotes.

Also I think his comment misses the point, as beloved as AA and AF are at B3D, most people probably dont notice, and I doubt theyll ever be standard on consoles. Just like 60 FPS (or even locked 30) will never, ever be standard, because better graphics are more important (save for classes that demand 60, like racers).

I guess if AA and AF are that important to you, you should stay on (much more expensive) PC. Although I noticed AF doesnt play nice with Crysis (you can either have AF or POM not both). It's typically benchmarked without AA and AF, on the highest end $600 cards and $1000 CPU's.

I'm sorry, did you just say grandmaster's analyses are "anecdotal"?! I can't discuss anything with you when you say stuff like that.

 

[Graham]

The reason they don't use Logluv is the same reason they didn't choose FP16 - lack of hardware alpha blending. (Then again, the fillrate gets a massive hit from the dual render targets... :s )

Take this as my 'not a developer with access to a devkit' understanding, but from what I know, the way xenos is setup the MRT hit may not be as large as expected. (AFAIK) It's edram has 4 memory channels - as to not require colour compression when using 4xaa. One channel is used when you aren't using msaa. However, the extra channels can be used when using MRT, so in the case of their two RT rendering - in theory at least it may still be quite efficient.

As for why they didn't use R11G11B10... who knows... It could have been the colour precision or usage of destination alpha. Or maybe it wasn't available in time during development. :s

Bungie released a Gamefest paper called 'HDR the bungie way'. It went into a bit of detail as to why they did what they did. The gist was they wanted loads of dynamic range without banding. And... well.. basically, they did it for the bloom.

 

[swaaye]

I've done blind tests in HL2 on my technology illiterate friends and every one of them chose AA+AF. They just couldn't say why. Lots of comments like "I don't know it just looks better".

I think I've been a bit misunderstood here. Yeah AF is awesome, but with 360 I've seen bilinear filtering. As in not even trilinear. As in you can see mipmap boundaries like it's 1999. That was just so jarring in Forza 2. I couldn't believe that I was seeing that in a game that was touted as "graphically impressive and redefining blah blah". Even N64 could do better filtering than that lolz.

Yeah it's anecdotal evidence and isn't proof of anything other than the devs simply didn't implement even trilinear filtering. It's not everywhere in the game either. I remember that it was off the road. The road itself was a blurfest, probably using trilinear or maybe even 2x AF (been a long time since I played it).

 

[liolio]

It's probably more a question of not taking advantage of something that's practically free.

The format does support blending. Sebbi explained it awhile back. Destination alpha isn't necessary for blending because there is still access to the source alpha, the output value from the pixel shader... Think of it as... immediate use? Frame-to-frame it's not going to matter because the output is always changing, but say for... multi-pass rendering where the value should be kept for each pass , but there are alternate solutions for a lack of destination alpha anyway.


Other examples would be... foliage, chain-link fences, explosions, smoke, fire, shields, shield bubble.

Thanks for the explanation

It was GDC this year. I'm guessing you also got the idea for environment tessellation from the Halo Wars presentation?

Yes, for those interested the original topic (with all the slides) is here (Again thank alstrong it was a good one ).
I'm currently rechecking the presentation by Bungie and the Microsoft halo team.
OT
I like this one "Mixed Resolution Rendering" from AMD
/OT
EDIT
I re-read the "zen of mulit core rendering" and while it shows what Bungies/Ms halo team want to implement in the future (read next gen). But some time they give hint stating some effects or this could be feasible with actual hardware.

 

[zed]

While we're on this subject, can anybody tell me an easy way to ascertain AF by eyeballing a game?

I can see tearing and AA (obviously). But I really cant tell AF. I've seen examples, but they usually involve grid patterns that may not be in a given section of a game. Is there a way to tell even from textures without a grid pattern?

the easiest way is to look at polygons at obliquish angles to the camera (ground,walls, or in this case also the gun)

16xAF vs no AF

 

[upnorthsox]

the easiest way is to look at polygons at obliquish angles to the camera (ground,walls, or in this case also the gun)

16xAF vs no AF

But I think his question is how do you quantify it? It's obvious the top one has alot more AF but what makes it 16x from looking at it?

 

[homerdog]

I have a real hard time spotting the difference between 8x and 16x AF unless I'm viewing something at a very extreme angle.

Anything less than 8x though and things get pretty obviously blurry.

 

[AlNom]

so in the case of their two RT rendering - in theory at least it may still be quite efficient.

Good to know! Thanks.

Bungie released a Gamefest paper called 'HDR the bungie way'. It went into a bit of detail as to why they did what they did. The gist was they wanted loads of dynamic range without banding. And... well.. basically, they did it for the bloom.

True, but they don't mention the R11G11B10 mode at all, just the 7e3 FP10 (1010102). :s On the other hand, I was wondering if the former had noticeably worse colour precision due to one or two bits less for the mantissa. I guess so.

 

[zed]

I have a real hard time spotting the difference between 8x and 16x AF unless I'm viewing something at a very extreme angle

the same is true with AA + resolution in prolly most things WRT graphics, the law of diminished returns, i.e. the higher they go the less impact they make
0xAA -> 2xAA is greater than 2xAA->4xAA
320x240 -> 640x480 is greater than 640x480->1280x960

I saw this screenshot in another thread, the ground would look a lot nicer with some decent AF (also it might be using billinear filtering instead of trilinear, but hard to say for certain)

edit - also the environment map looks to be low resolution as well, 128x128x6 pixels

 

[Rangers]

I was reading a flamebait comment on why Metal Gear Rising will look better on the 360. "...the 360 renders textures better." It took a great deal of restraint for me to not spout off about how the PS3 renders textures better because it uses AF, so even if the 360 has more memory for higher res textures, you might only notice the difference in the rare instances when they're up close and parallel to the camera. I feel like the bigger person for not having said that.

I've run 16xAF on every game with every generation of hardware since I turned red with the Radeon 8500, so I'm perplexed why the 360 seems incapable of using AF.

So you're saying the PS3 always uses AF and the 360 never does?


Uncharted 2 doesnt use AF according to Grandmasters analysis. Another I recall was Resistance 1.

These are just the ones I know for certain because I happened to see it discussed somewhere. Who knows the status of countless other games.

Without doing some at least half assed scientific study of games on Xenos vs RSX and their AF levels, all this chatter here is anecdotal imo.