Digital Foundry Article Technical Discussion Archive [2010]

Status
Not open for further replies.
Gah, if that's MLAA, I have to say, I'm less than impressed. Definitely better than Quincunx certainly, but not much to write home about. Still a jaggy mess, just less of a jaggy mess.
You've confused me. Which of those LBP2 shots is a jaggy mess? The worst offender I can see are the ultra-high contrast glowing beams against a black background, and they're muted jaggies, and only where they occur. There are more inbetween step values than 4xMSAA, although the distribution favours a stronger, more visible last step I think. However, the near vertical orange beam is completely jag free. None of those LBP2 shots is anything like a jaggy mess that I can see! :???:

The tech differences are quite remarkable. Overall the lighting is much more intrinsic, you notice. Well, as the article says, light sources and shadowing are present as effects all over the shop, so even just adding fire, you get convincing lighting with it. I don't think the improvements are that obvious until you place the exampels side by side like this. I certainly didn't notice LBP2's fire was much improved when just playing the Beta.

I'm also surprised no-one has commented on that splat effect! Best special effect I've ever seen in a game! :oops:
 
Gah, if that's MLAA, I have to say, I'm less than impressed. Definitely better than Quincunx certainly, but not much to write home about. Still a jaggy mess, just less of a jaggy mess.

Regards,
SB
:LOL:
Your post made me laugh...thanks for this Mr Buddha, Sir!

I wonder how long it takes until big multiplat releases introduce MLAA?
It seems that it was to late for instance for the new Castlevania...however, it seems that smaller dev teams can include this without much sweat in a few days (see LBP2 and Under Siege as an example) - I don't understand the logic behind this to be honest: it must be even easier for big devs to incorporate such a tech into their engine, right?!
 
Last edited by a moderator:
Some items in the DF article that totally made my day:

"MLAA is now a part of the Edge tools available to all PS3 developers and it's effectively a drop-in component that can be added in an afternoon of coding - expect to see it in many more titles, not just those from Sony."...
That's Cell kicking major b@tt for everyone!


"The new "Splat" effect in LBP2 is excellent too. Media Molecule appears to have implemented a special effect buffer with its fluid dynamic applied, and when you destroy an object, it gets mixed in with the current buffer contents. Create a toxic cloud, put in an object and then dissolve it, and its cloud gets mixed in with the toxic cloud's turbulence."
I believe this comes as the biggest effect, a solid object getting liquified and flowing in front of players' eyes with complete fluid-dynamics... that's a big heads-up for soft-body physics enthusiasts like myself for sure!


"Another important change to LittleBigPlanet is that as a physics-based game it was surprising to see bounce absent (in the 1st one). In LBP2, bounciness is a material parameter you can :eek:tweak:oops:, so basketballs and footballs for example behave more realistically."
Yes, yes, give us more of that rigid-body beautiness alright!!!


"in the original LittleBigPlanet, 24 cloned creatures was the limit before you blew the thermometer. In LBP2 we made an army of 150 creatures using the old creature components with the thermometer only a little over half full."
have mercy MM, sweet sweeet mercy :)
 
Last edited by a moderator:
MLAA is not as simple as that comment seems to suggest, because the performance characteristics still have to be balanced around it to maintain a stable fps rate.
 
Well that's just what was said in the LBP interview earlier in the year.

Likely meaning if you have the performance and resources to spare, it can be dropped in with an afternoons work of course.
 
To be fair, the LBP1 shot has the camera higher up, and those edges aren't vertical, whereas in the LBP2 example they are. Thus with zero AA, they'd still be perfect jag-free lines! However, the sentiment is definitely true. LBP1 has those crawling jaggies; LBP2 doesn't. There are a few cases where AA isn't as strong, but it really is an incredibly clean game.
And first game is using MSAAx2 not QAA.
I think he meant versus QAA in general. Although I don't understand how he can view LBP as a jaggy mess, so maybe I'm wrong on this!
 
I'm not sure how that first shot could be described as a "jaggy mess". They're almost completely eradicated. Perhaps the almost-horizontal bright white against black of the second shot shows something not much better than the original though, but isn't that a worse case scenario?

By the way, what was the AA in LBP1?

The first shot in the analysis isn't a good one to use when looking for aliasing, as the MLAA shot has the most noticeable feature for AA as a near perfectly verticle line rather than the more diagonal line of the LBP1 shot. But even in that shot it's quite easy to see the aliasing in the high contrast areas. Low contrast area's are always going to be more effective at hiding aliasing (see Doom 3).

You've confused me. Which of those LBP2 shots is a jaggy mess? The worst offender I can see are the ultra-high contrast glowing beams against a black background, and they're muted jaggies, and only where they occur. There are more inbetween step values than 4xMSAA, although the distribution favours a stronger, more visible last step I think. However, the near vertical orange beam is completely jag free. None of those LBP2 shots is anything like a jaggy mess that I can see! :???:

The second set of shots is the worst offender obviously as it has the most high contrast edges, but it's evident in the other shots also. It's rather comparable to 2x MSAA and a bit better in some cases, but in looks it doesn't appear to come close to approaching 4x MSAA quality in the cases that matter the most (high contrast).

From all the "OMG, MLAA is the best thing since sliced bread posts" I've seen, I was quite expecting something between 4x and 8x MSAA. Or in other words something closer to what I expect from 4x or 8x MSAA when I use that on PC.

It's good for consoles certainly, as everything on console is pretty much a jaggy mess and less of a jaggy mess is at least progress.

BTW - if anyone has read my past posts on AA in the 3D hardware forums here, they'll know I'm hyper senstive to AA. Especially so in motion versus a still shot. As when it's in motion it's going to be constantly dragging my eyes away from the game's visuals to the aliasing. In other words, AA to me is generally far less offensive and noticeable in a still shot than it is in motion. So if I can see it in a still here, I can only imagine how bad it'll be in game (for me).

But again, that's nothing against MLAA, but probably more against all the freaking hype for it on the boards here which made me expect something better than what it turned out to be. As I said, it's less of a jaggy mess which is a good thing. And certainly a far better solution than Quincunx.

As for the tech, yeah, it's mostly a quite noticeable improvement over LBP1. Although it uses way too much of my most hated current gen 3D checkbox feature, DoF. But at least it's not so bad here, as it's obvious you're supposed to be looking through a camera at what's happening as opposed to say a FPS where presumably you're looking through your own eyes.

Regards,
SB
 
Interesting article although do I spot some downgrades?

I am talking about what looks like just a simple background blur a la UT3 style in LBP2 as opposed to more 'proper' DOF in LBP1. Also lightbeams in LBP1 seems to use more slices to form volume though they dont get occluded. Motionblur less samples/quality?

Otherwise nice ~moderate improvements.
 
Last edited by a moderator:
From all the "OMG, MLAA is the best thing since sliced bread posts" I've seen, I was quite expecting something between 4x and 8x MSAA. Or in other words something closer to what I expect from 4x or 8x MSAA when I use that on PC.

It's good for consoles certainly, as everything on console is pretty much a jaggy mess and less of a jaggy mess is at least progress.

...But again, that's nothing against MLAA, but probably more against all the freaking hype for it on the boards here which made me expect something better than what it turned out to be. As I said, it's less of a jaggy mess which is a good thing. And certainly a far better solution than Quincunx.

And the scenes have the backgrounds blurred out which tend to have the smallest details thus without DOF it would have been an even bigger offender (unless contrast is such that it makes it hard to spot jaggies). But when it works and where it works it sure is better than none or perfomance hit of 2-4xMSAA.
 
The first shot in the analysis isn't a good one to use when looking for aliasing, as the MLAA shot has the most noticeable feature for AA as a near perfectly verticle line rather than the more diagonal line of the LBP1 shot. But even in that shot it's quite easy to see the aliasing in the high contrast areas. Low contrast area's are always going to be more effective at hiding aliasing (see Doom 3).

The second set of shots is the worst offender obviously as it has the most high contrast edges, but it's evident in the other shots also. It's rather comparable to 2x MSAA and a bit better in some cases, but in looks it doesn't appear to come close to approaching 4x MSAA quality in the cases that matter the most (high contrast).

From all the "OMG, MLAA is the best thing since sliced bread posts" I've seen, I was quite expecting something between 4x and 8x MSAA. Or in other words something closer to what I expect from 4x or 8x MSAA when I use that on PC.

Do you have comparison pictures that show 4x or 8x MSAA with MLAA to show the differences ? It's easy to make a comment without visuals. According to the original MLAA tech article, it can be as good as 16x or even 64x in the best case scenario.

It's good for consoles certainly, as everything on console is pretty much a jaggy mess and less of a jaggy mess is at least progress.

I thought GoW3 has pretty good IQ ?

When someone felt compelled to bring in PC GPU performance to compare with a 5 year old console with limited memory, it already says something about the software technique used.
 
Do you have comparison pictures that show 4x or 8x MSAA with MLAA to show the differences ? It's easy to make a comment without visuals. According to the original MLAA tech article, it can be as good as 16x or even 64x in the best case scenario.

Follow me.
 
Only jaggies i can see are from the lantern in the background, looks better in the first pic.
 
Last edited by a moderator:
The second set of shots is the worst offender obviously as it has the most high contrast edges, but it's evident in the other shots also. It's rather comparable to 2x MSAA and a bit better in some cases.
Well, I just plain can't relate to that! Noticeable jaggies if you look or them, maybe. But jaggy mess? What do you call 0xAA then? And as for MLAA performance, these shots obviously haven't gone out of their way to focus on MLAA comparisons, but if you get a block in the editor and rotate it, jaggies are extremely hard to spot, such that it's hard to tell if an edge is perfectly vertical/horizontal or not, because you can't measure by the edge tearing.

As for the tech, yeah, it's mostly a quite noticeable improvement over LBP1. Although it uses way too much of my most hated current gen 3D checkbox feature, DoF. But at least it's not so bad here, as it's obvious you're supposed to be looking through a camera at what's happening as opposed to say a FPS where presumably you're looking through your own eyes.
You're not safe - the movie-type camera has a DOF setting that allows most of the level to be blurred out. :p
 
Any video showing the Splat effect ? Artistically, I like the old smoke effect better even though some may think it's sparse. Can the new smoke effect be used as a gameplay element ? e.g., place monsters in thick cloud. Or is it for show only ?

I'm most happy with their improvement in gameplay. Together with the improved user tools, the more efficient use of thermometer memory is the best news for me.
 
Interesting article although do I spot some downgrades?
Yeah, there's a few tradeoffs. As well as the motion blur taking a hit, it seems the DOF is running on a similar change, and isn't as refined in the LBP2 Beta as LBP.

Also lightbeams in LBP1 seems to use more slices to form volume though they dont get occluded.
LBP2 doesn't use slices. It's a volume light, like (AFAICT) CryEngine 3's GI solution with a single volume and no nested volumes. These comparisons have used the same level created in LBP imported into LBP2, and the default spot-light fog setting from LBP isn't as strong in LBP2 which can be cranked up. You'd then see the smoothness of the shadows. They have some optical inaccuraces, like not occluding the whole length of light beam, but they are very convincing and without the artefacts of typical cascaded shadow samples.
 
Any video showing the Splat effect ?
It's in the full Eurogamer article. I love it!!
Artistically, I like the old smoke effect better even though some may think it's sparse. Can the new smoke effect be used as a gameplay element ? e.g., place monsters in thick cloud. Or is it for show only ?
It's the new rendering effect for the old Toxic Fog, so functions identically - you can have things walk through it it stick it on objects to make them dangerous. It'll look out of the place with orignal story levels though, and the ghost spewing fluffy clouds instead of their eerie haze!
 
Status
Not open for further replies.
Back
Top