Digital Foundry Article Technical Discussion Archive [2010]

[Sigfried1977]

Sorry, I meant to say it isn't using MT Framework (though DF says it's based off the older version for DR1), but it's not using the updated/current version Capcom is using now.

But yeah, even with the current version Capcom still screwed-up with RE5 and LP2 (with the exception DMC4). (Super)SFIV on other hand were pretty equal performance-wise on both consoles, but that used a completely different engine, it sucks that they allegedly won't be using again in the future.

I thought RE5 was fine actually. I'm switching back and forth between the PC version (which technically runs circles around both consoles in terms of performance and image quality) and the PS3 version, and even though Capcom toned down the PS3 version in spots it's barely even noticeable. Seriously, compared to previous console generations these multi-platfrom games are remarkably close in terms of quality, and even with downgrades RE5 remains one of the best looking games on the PS3 as far as I'm concerned. Plays just fine too (I platinum-ed it solo while having 0 troubles with the Sheva A.I. Saved my ass way more often than I saved hers in fact. Maybe I'm just weird that way)

Calling it a screw-up is, quite frankly, unfair. Judging by the screen-shots that kinda holds true for Dead Rising 2 as well. I'm really surprised that the 30% resolution drop isn't more plainly evident to be honest.

 

[Laa-Yosh]

Am I right that than SSAA yields the best solution?
Or is there an AA tech that yields better IQ than SSAA

There are two approaches, you can use full supersampling or you can decouple the AA for the geometry and for the shading (usually, geometry samples are cheaper, basically coverage tests for the polygons).

What becomes far, far more interesting is the distribution and pattern of the samples, to make the best possible use of your calculations. Ordered grid patterns are the least efficient, rotated grid are better, and quasi-random stochastic sampling is the best.

Adaptive (contrast sensitive) AA is another good feature to have (basically, compare samples and if the contrast is higher than a set threshold then increase number of samples and do more calculations).

MLAA only is a substitute solution until the hardware can do better, IMHO; but it may end up as an additional tool to enhance the results of the more traditional AA solutions. On its own, it still has a lot of weaknesses in my opinion.

 

[Billy Idol]

What becomes far, far more interesting is the distribution and pattern of the samples, to make the best possible use of your calculations. Ordered grid patterns are the least efficient, rotated grid are better, and quasi-random stochastic sampling is the best.

This is interesting. I use similar tricks, with different sample patterns, to de-alias my numerical simulation method - interesting that this is used in graphics tech as well! But I never heard of stochastic sampling - is the quality of this approach easy to control? Or is it prone to artefacts (I guess the accuracy highly depends on the number of samples...in such approaches, typically, the results converges only proportional to the root of the sample number)

Adaptive (contrast sensitive) AA is another good feature to have (basically, compare samples and if the contrast is higher than a set threshold then increase number of samples and do more calculations).

That adaptive AA is interesting. I know that it is possible to detect edges in pics, but is it really possible to adaptively detect high frequency texture aliasing for instance. Which criterium is needed to find those types of aliasing which are not rooted in an edge?

 

[Silent_Buddha]

This is interesting. I use similar tricks, with different sample patterns, to de-alias my numerical simulation method - interesting that this is used in graphics tech as well! But I never heard of stochastic sampling - is the quality of this approach easy to control? Or is it prone to artefacts (I guess the accuracy highly depends on the number of samples...in such approaches, typically, the results converges only proportional to the root of the sample number)

Well, ATI has been using quasi-stochastic random distribution for years now. It's not random from frame to frame, but the sample points chosen are a set semi-random pattern. I believe Nvidia has been doing something similar the past few years also, although theirs is a bit more of an ordered pattern I think.

ATI a few years back also experiemented with quasi-stochastic across multiple frames. They experiement both with 2 alternating frames with different sample patterns and 3 frames of differing sample points. The problem here was that unless the FPS was a rock steady 60 FPS or above, that you would notice shimmering in the aliasing, which kind of defeated the purpose of AA. On the other hand when it worked, it gave the impressive of much higher levels of AA for a relatively small cost. So for example, visually it may look like 8-16x while only taking the hit for 4x box MSAA. Additionally without any of the artifacts or corner cases where the AA falls apart as is the case with almost all custom AA schemes in use today. Well, except for the corner case if your FPS falls below a certain point and shimmering becomes noticeable.

The dawning of the mass exodus from CRT to LCD's pretty much made the tech irrelevant in the PC space.

And that's something that would be a problem for any purely random frame to frame stochastic AA implementation. You're going to need high sustained frame rates. Considering the relative rarity of 60 fps games on consoles, it's extremely unlikely we'll see much work in that direction there. And the prevalence of 60 hz LCD's in the PC similarly hampers implementation there.

It's a bit of a shame really, as the quality for performance is perhaps the best available for any given AA scheme.

Regards,
SB

 

[Trejser]

Face-Off: F1 2010

________________________

Interesting read, good that PS3 versions of EGO games performs better and better with every release. But they definitely should switch to MSAAx2 or MLAA, QAA works well for some games (KZ2, R2) but F1 isn't one of them

________________________

Grandmaster: Any info about Castlevania Face-Off?

 

[Yashamaru]

Face-Off: F1 2010

________________________

Interesting read, good that PS3 versions of EGO games performs better and better with every release. But they definitely should switch to MSAAx2 or MLAA, QAA works well for some games (KZ2, R2) but F1 isn't one of them

________________________

Grandmaster: Any info about Castlevania Face-Off?

R2 doesn't use QAA but it use 2x MSAA

 

[Gitaroo]

nice job on the article, really hoping for a Castlevania face off soon, like before the game release.....

 

[Billy Idol]

Well, ATI has been using quasi-stochastic random distribution for years now. It's not random from frame to frame, but the sample points chosen are a set semi-random pattern. I believe Nvidia has been doing something similar the past few years also, although theirs is a bit more of an ordered pattern I think.

ATI a few years back also experiemented with quasi-stochastic across multiple frames. They experiement both with 2 alternating frames with different sample patterns and 3 frames of differing sample points. The problem here was that unless the FPS was a rock steady 60 FPS or above, that you would notice shimmering in the aliasing, which kind of defeated the purpose of AA. On the other hand when it worked, it gave the impressive of much higher levels of AA for a relatively small cost. So for example, visually it may look like 8-16x while only taking the hit for 4x box MSAA. Additionally without any of the artifacts or corner cases where the AA falls apart as is the case with almost all custom AA schemes in use today. Well, except for the corner case if your FPS falls below a certain point and shimmering becomes noticeable.

The dawning of the mass exodus from CRT to LCD's pretty much made the tech irrelevant in the PC space.

And that's something that would be a problem for any purely random frame to frame stochastic AA implementation. You're going to need high sustained frame rates. Considering the relative rarity of 60 fps games on consoles, it's extremely unlikely we'll see much work in that direction there. And the prevalence of 60 hz LCD's in the PC similarly hampers implementation there.

It's a bit of a shame really, as the quality for performance is perhaps the best available for any given AA scheme.

Regards,
SB

thanks for the exlpanation, now I get it!

 

[Billy Idol]

Thanks grandmaster for the F1 face off!
And thanks a lot for having some vids using non TPS view as well!
I wonder if in like for like situations, the games perform differently with different views?

Could it be that the missing post processing (compared to Dirt2) is more favorable for the PS3 and thus put this version slightly ahead in terms of performance?
Maybe the postprocessing used in Dirt2 is a little bit more taxing on the PS3, so slight advantage for the Xbox360 - in F1, no post processing...slight advantage for PS3?

 

[Neb]

Nice article Richard. Though I do wonder why DX11 features in menu is missing considering Codemasters already had it in for DIRT2. Hopefully it will be implemented with patch into menu.

 

[Neb]

opcorn:

Castlevania Lords of Shadows DF comparision article.
http://www.eurogamer.net/articles/digitalfoundry-castlevania-lords-of-shadow-face-off

 

[Trejser]

Looks like 360 version have Triple Buffering... that's unexpected

 

[Billy Idol]

Thanks Richard for the comparison!

 

[dragonelite]

Cool still.
Hope they patch in a no vsync or soft vsync option because im not really sensitive for tearing when im playing. But i am sensitive to low frame rate going from 24 fps to 35 fps in Age of canon when i got a new laptop was like playing a different game for me.

 

[Billy Idol]

In the first comparison vid, rather at the beginning of the vid (starting at about 10 seconds), where you see the end part after the horse chase. He jumps with the horse over the cliff. The horse somehow ghostifies...in the PS3 version you see how the shining horse casts a light on both upper sides of the cliff (actually the heads of statues, nice detail), whereas in the Xbox360 version it is completely missing. I also have the feeling that the light is casted on the werewolf pack in the PS3 version, which is again missing in the Xbox version.

This does not look like absolute identical, like for like pixels...as claimed in the article??

 

[brain_stew]

Cool still.
Hope they patch in a no vsync or soft vsync option because im not really sensitive for tearing when im playing. But i am sensitive to low frame rate going from 24 fps to 35 fps in Age of canon when i got a new laptop was like playing a different game for me.

They're already using triple buffering, so disabling vsync isn't going to provide any significant increase in framerate, it'd just release some memory they have no use for. The game just runs that poorly.

 

[AlNom]

In the PS3 version you see how the shining horse casts a light on both upper sides of the cliff (actually the heads of statues, nice detail), whereas in the Xbox360 version it is completely missing.

The light radius/scale is lower. In the closer shot of the first statue head, it's still being lit albeit dimmer.

 

[Gitaroo]

got to love PS3 games that support DTS, but also hate that games have to force DTS (disable DD) in order to work I think ppl with receiver that support DTS and PCM audio should probably go with the PS3 version of castlevania easily, Im playing with DTS sound, and its crystal clear, going to try PCM later.

 

[SG79]

got to love PS3 games that support DTS, but also hate that games have to force DTS (disable DD) in order to work I think ppl with receiver that support DTS and PCM audio should probably go with the PS3 version of castlevania easily, Im playing with DTS sound, and its crystal clear, going to try PCM later.

It's a bit of a pet peeve, and again, Naughty Dog is the only developer that lets you choose your poison from the game's own options menu.

I'm playing DR2 on PS3 (since I have no other choice), and the default PCM option is incidentally the weakest. The surrounds are inactive and the volume is lower, but if I switch to DD, it's noticeably better.

 

[Gitaroo]

most of not all 1st party PS3 titles that support DTS will automatically choose DTS over DD by default, I remember Bad Company 1 was also default to DTS.