Digital Foundry Article Technical Discussion Archive [2009]

[Dr. Nick]

This also appears to be a very inaccurate statement and once again shortchanging the PS3 version. The writer claims that the PS3 version is more or less strictly limited to around 6 enemies on screen, while the 360 version can handle "far more enemies." Here's a video that shows the PS3 version handling 12 enemies on screen at a time (with perhaps some slight slowdown, although it's difficult to ascertain in the video):

http://www.youtube.com/watch?v=MA7Runk_lSc

I think you missed the video portion of the article where they compared similar situation across consoles and it still ended up showing more enemies on the 360. The problem with this comparison is that there are so many variables at play between the two versions. There is also a different design philosophy at play. It would have been great and I was partly hoping for a like for like situation between Sigma2 and NG2 even after the removal of blood but that didn't happen. The two games are different which makes a comparison very hard to do.

I will say based on what is known about both of these systems what he did say in the article is more sound than most of the stuff you will hear about the games. They might have tried to not get the Z rating in Japan but NG2 got an M rating over here in the US and tolerable ratings elsewhere. Why not turn on the blood for people over here? They made changes with previous NG's for different territories why not do the same with this game especially when you have people complaining about its removal? Why when your cut scenes are limited to 30fps do you remove extra models? Nobody is playing these scenes I understand but why not leave them the same. They are pretty much the same just with less models in them now. It would have been really impressive if they got the old stairway scene working at 60fps and would have been a way to stick it to the director of NG2 that they managed to do everything that the 360 did only much better. Why didn't they try to do that? If everything was designed like for like just with more polish and balanced gameplay and the other extras including things that the PS3 is good at no one would think twice about Ninja Gaiden Sigma 2’s technical superiority but they ended up removing or toning down things that the 360 is believed to be good at and the PS3 not so good at they locked cutscenes at 30fps. (I think old Team Ninja should have done the same thing though for cut-scenes and used the extra power to make the cutscenes look better. They kind of needed it) In the end the game is still more polished on the PS3 so it is the better version between the two but there were somethings in the NG2 that people could question if the PS3 would have had a easy time pulling off at the same quality.

We might get the new Team Ninja side of the story since they seem more willing to talk the tech stuff but we won't be getting that from the old Team Ninja. They didn't bother doing it even when they were given the chance to when they did the Softimage casestudy.

 

[AlNom]

This I can't possibly concieve since none of us have any idea how much SPU utilisation the NGS2 code has. And you say that even if it has some, the 360 could probably replicate it? I'd really like to know how? Since using the SPUs on PS3 would imply pretty much maxing out the RSX, and so i'd imagine it would likely be close to doing so on Xenos (at the PS3 version resolution with same AA). So without Xenos what else can the 360 use to do the job of the SPUs on CELL?

I gather that it was the implication of using Cell to help out RSX with certain loads such as culling and feeding data to the VS units, which it should be clear that it is needed (from developer comments and presentations). That they're capable of doing post-process effects at a lower resolution (at higher precision) shouldn't be taken for granted because RSX is naturally faster at it. The use of doing said effects on the SPE's is a matter of balancing frame render times. You'd get that idea from the KZ2 docs out of GDC even.

For instance, RSX will be able to do post-process effects much faster, but they're trying to keep within 33ms render time. Offload the extra work to the SPE's in parallel whilst the other work is being done so in the end everything will be done on time. It ends up being a win because the brunt of the serial work (post-process calcs) has been taken care of. But this is highly simplified and isn't the scenario in all games.

 

[liolio]

Back to my rant about vertex shaders
Ninja gaiden renders @1120x585 which is 655,200 pixels, NGS 2 @720p which is 921,600 pixels.
The 360 deals with only 70% of the pixels the ps3 deals with or the other way around the ps3 deals with a pixel shading bigger of 40% (as Alstrong pointed).
Disclaimer: gross logic is to follow
X= ps3 vertex load Y=ps3 pixel load
A 360 vertex load B=360 pixel load
B=Y x 0.7 // X+Y=A+B => X + Y= A + 0.7 Y
We can't say that xenos has 30%more resources to deal with vertex load. In all likelyness Y>>X B>>A and A>X. Dealing with 30% percent less pixels may free multiple times the resources actually used by X to deal with A. I feel iffy about it, I can't see some extra 10k or less polygons monsters/encounters consuming those extra resources if we consider only the vertex load.

I read again Epic presentation about the gore effect to try to find answers. I actually had a tough time because I don't how each effects affect perfs. The presentation is here
First, the meshes and how they "split". Epic went for a data driven model to save performances. I think it's the sae for Temco (ie body are not randomly chopped).

Undamaged mesh with no overhead
Gore mesh with additional information
Completely independent of the undamaged mesh
Pre-cut with all gore pieces separated
Full freedom to hand-model cuts and guts
Can use different/additional textures and shaders

Basically the engine switch to the more complex mesh went needed. If I understand properly they trade RAM for better perfs most of the time. IN NG2 I don't how many time "dismembering" happens but I feel like it's more often (extra insight on the game welcome). Overall Temco would make a clever move by using this tricks as even if there are more characters on screen I'm not sure that you see that many different model at a time.
sum-up:
Gore cost memory and perfs (both on CPU and GPu side) in a variable ratio depending on how you handle it.
It may have an extra cost if the engine keep a bunch of pieces around and you can interact with them both on memory and CPU side I guess (I'm not sure that actual rendering of some/many limbs is that much of a concern perfs wise has models/meshes and textures are likley to stay in RAM anyway).

Then come this:

Many different techniques used in combination:
Projected blood decals
Screen-space blood splatter
World-space particle effects
Surface-space shader effects
Fluid simulations
Morphing geometry

That's for Geow and some may not have equivalent in NG2 (or are only used in Geow for really specific scenes).
I enlightened those that seems relevant to me. I' not sure I understand properly how this work but from my understanding it looks like a mixed bags in term of workload.
For decals CPU have to decide where they are (tied to a mesh or a static object) and their evolution then on the GPU side it looks to me like extra texturing and shading and a fair amount of alpha to coverage (hence Mazinger dude take on bandwidth and Edram).
Particles are well particles.
Overall I would not want to go in details I may not understand or misunderstand and I'm sure some members could extrapolate for Epic presentation more than me more accurately.
Overall gore effects seems to stress the an engine and a systems in pretty various from muy low level understanding.
I'm still not sold on Mazinger Dude conclusion even if I can see some truth to it. Overall I'm in lost in the middle between design choices to reach a wider audience? memory and badwidth constrains (extra memory for some effects vs extra memory for the frame-buffers, lot of alpha to coverage)? In the end I would not a pretty bad use of CPU resources especially on the PS3 (which may translate in no SPU culling, bad instancing as some one pointed, particles not done on cell, etc, etc.) Actually I lie I'm in the dark and I do not think that the game make the best use of available resources no matter which system you consider which make comparison on resource utilization even more iffy.

End of the rant

 

[MazingerDUDE]

The writer mentions the greater vertex capabilities and polygon-pushing power of the 360, and how the PS3 would be unable to handle it. Yet many of the comparison screens shown in the article itself totally contradict this argument of the 360 version having superior polygonal detail, with many shots showing greater polygon detail in both the environment and the character models on the PS3 side

The difference in polygonal detail in both the models and environment are clear in many of these screenshots... in FAVOR of the PS3 version.

Yes, there're parts with added geometry on NGS2, but there're far more with reduced ones.

http://www3.telus.net/public/dhwag/NGS2_polydecuction_01.jpg

http://www3.telus.net/public/dhwag/NGS2_polydecuction_02.jpg

http://www3.telus.net/public/dhwag/NGS2_polydecuction_03.jpg

These are rather subtle changes done with extra care and added shader details, but there're tons of examples like these across the entire game. I picked the most obvious ones to put up at DF, it'll take almost forever to cover every single details in the game. The Colosseum crowd scene is not just a cut scene, the part is actually a background for boss battle in that chapter. So is the cave scene with tons of removed objects.

The set of foliage shots has got nothing to do with the poly count as it's just a 2D texture. In fact 360 version should be more heavy on GPU with overdraws as it displays more on what's behind.

This also appears to be a very inaccurate statement and once again shortchanging the PS3 version. The writer claims that the PS3 version is more or less strictly limited to around 6 enemies on screen, while the 360 version can handle "far more enemies." Here's a video that shows the PS3 version handling 12 enemies on screen at a time

The scene was artificially set up by gathering ninjas from 3 sections. You won't see more than 6 in each section where 360 version could easily throw 50~100% more at smoother frame rate. And this is only place in NGS2 where you can gather so many as in other places, following ninjas will just disappear when you reach the next section. (this could be game breaking, so they sometimes add barrier to block you from moving any further without killing them all)

 

[Shifty Geezer]

Okay, new proper tech thread for DF. I've spliced out the latest NG2 tech talk from the old DF thread, although I didn't do a particular exacting job, so don't be surprised if your highly relelvant and insightful post hasn't made it across!

 

[assurdum]

http://www3.telus.net/public/dhwag/N...cuction_01.jpg
http://www3.telus.net/public/dhwag/N...cuction_02.jpg
http://www3.telus.net/public/dhwag/N...cuction_03.jpg
Please don't said that is for a technical reason...even the first ngs have these slight differences compared to xbox version of ng. Rewrite a code not necessary imply the exactly position of every details, we don't talking of simple code porting. ngs is a different engine to ng 2 on 360.

 

[obonicus]

Links are showing with ellipses (...).

 

[semitope]

Couple problems I want addressed are the polygon claims of the writer and how the provided evidence could truly claim that.This isnt really directed at the DF guys as i just want to know if # of polygons can be assessed by pictures and videos alone.

My second concern is the idea that njs2 can show fewer enemies than nj2. The only thing the test shows IMO is the maximum number of enemies you can gather in that particular spot and has nothing to do with the engine.

Is there any thread on B3D confirming the authors vertex claims?

 

[Silent_Buddha]

Yes, there're parts with added geometry on NGS2, but there're far more with reduced ones.

http://www3.telus.net/public/dhwag/NGS2_polydecuction_01.jpg

http://www3.telus.net/public/dhwag/NGS2_polydecuction_02.jpg

http://www3.telus.net/public/dhwag/NGS2_polydecuction_03.jpg

Those are some pretty drastic changes. The windows look almost naked without the shutters.

And replacing 3D objects with 2D textures or just getting rid of them entirely? I hadn't realized just how much 3D detail was cut in the port.

That said at least they added some texture detail to compensate somewhat. Still I hate when 3D games use flat 2D textures for world geometry.

Regards,
SB

 

[makattack]

Please don't said that is for a technical reason...even the first ngs have these slight differences compared to xbox version of ng. Rewrite a code not necessary imply the exactly position of every details, we don't talking of simple code porting. ngs is a different engine to ng 2 on 360.

It seems pretty clear that when you substitute bit mapped images / textures for 3D models, you're doing that for technical reasons. Even if this is a new engine, why go through the effort of changing the base content / object that runs on this new engine? Unless the new engine can't handle it with reasonable or desired results.

Someone (an artist perhaps, or modeler) went through the trouble of mapping/making a 3D object representing fences, windows, etc... to convert that to a flat image is work. Why do the extra work instead of reusing what you have available -- a bunch of numbers in a file? Someone had to convert those numbers into an image file, or flattened using some tool that modeled the original object(s) and surface mapped the appropriate textures.

 

[SG79]

You won't see more than 6 in each section where 360 version could easily throw 50~100% more at smoother frame rate.

Except when the engine was stressed and choked, and that happened frequently on higher difficulties where the idea was "let's throw more at the player". The NGS2 team was certainly smarter to avoid such situations, and it's too bad that it doesn't make much financial sense to patch up the 360 game.

One infamous crowding glitch was in chapter 11. If you simply avoid fighting enemies from the start and up to the bridge in the village, the enemy AI locks out. You'll literally have punching bags (at what looks like sub 25Hz frame rate).

Of course, the single incident that backs up your argument is the staircase ambush where the game was at is ugliest and was revised greatly in the port. It was one of those moments where you simply shook your head and wondered what they were thinking

 

[Dr. Nick]

It seems pretty clear that when you substitute bit mapped images / textures for 3D models, you're doing that for technical reasons. Even if this is a new engine, why go through the effort of changing the base content / object that runs on this new engine? Unless the new engine can't handle it with reasonable or desired results.

Someone (an artist perhaps, or modeler) went through the trouble of mapping/making a 3D object representing fences, windows, etc... to convert that to a flat image is work. Why do the extra work instead of reusing what you have available -- a bunch of numbers in a file? Someone had to convert those numbers into an image file, or flattened using some tool that modeled the original object(s) and surface mapped the appropriate textures.

I was going to say the same thing. In cases where this extra stuff doesn't bog down the system why remove it? They could make it look prettier or leave it the way it is. They instead chose to remove or remodel certain things with less polys and the remodeling I'm sure takes time to do. Ninja Gaiden Sigma 2 is certainly a better looking game but that doesn't mean that the removal all of these things were done to make the game look better and in some cases the removal of certain things makes it look worse like the colisium scene and other gatherings.

With that said I found Ninja Gaiden 2 to certainly not be the best looking game in Team Ninja's arsenal even before Sigma 2 annoucement and even after excluding Sigma 1. According to Itagaki's calms it might have been the best the 360 could go but it did it in an ugly buggy way.

 

[obonicus]

There really is nothing about NGS2 that can't be done on 360. In fact, they could probably add more alpha effects (blood, explosions) on top of it. In terms of theoretical shader performance, RSX & Xenos are roughly equal with Xenos being just more flexible with its unified structure. And you've got EDRAM for free MSAA and alpha as well.

You seem to have completely misunderstood my followup question. It was: if the PS3 had exactly the same capabilities as the 360, do you think we'd be seeing the exact same singleplayer game the 360 version got? There are unquestionably things that came from technical decisions, but coincidentally several points of contention technically were also major sources of complaints in terms of gameplay.

 

[psorcerer]

I think I have some misunderstanding here, I want to ask couple of questions regarding the article:

1. Game with lots of overdraw is it good or bad technical decision?
2. Game with lots of alpha textures instead of screen space effects, is it good or bad technical decision?

 

[Shifty Geezer]

1. Game with lots of overdraw is it good or bad technical decision?
2. Game with lots of alpha textures instead of screen space effects, is it good or bad technical decision?

They depend entirely on your hardware. If your hardware is good at overdraw, these are good design choices. eg. On PS2 they were the way to get things done! But if your hardware has trouble, then you want to avoid overdraw and go with single-pass effects wherever possible. If you're creating a multiplatform engine that needs to run on different hardwares, you'll want to avoid a technique that works on one box but not the other, or have alternative implementations. eg. Porting a PS2 multipass renderer to XB would result in a dog of a game.

 

[psorcerer]

They depend entirely on your hardware. If your hardware is good at overdraw, these are good design choices. eg. On PS2 they were the way to get things done!

Err... and how exactly this was THE WAY on PS2?

But if your hardware has trouble, then you want to avoid overdraw and go with single-pass effects wherever possible.

In my book you never want to draw things 10 times, you may do this because this is the only way to do specific effect on current HW, but this is not the "right" decision, this is a compromise.

If you're creating a multiplatform engine that needs to run on different hardwares, you'll want to avoid a technique that works on one box but not the other, or have alternative implementations. eg. Porting a PS2 multipass renderer to XB would result in a dog of a game.

In my book you never want to go crazy with overdraw, you always need to optimize for as little OD as possible.
In my book you never want to do things that can be done in screen space - in world space, because it's always less efficient.

 

[Richard]

In my book you never want to go crazy with overdraw, you always need to optimize for as little OD as possible.

Including back-to-front?

 

[psorcerer]

Including back-to-front?

If you can on your HW, why not?
At least do some occlusion.

I think the math is very simple: for 720p you need 1280 * 720 / 4 = 230,400 polygons on screen per frame, maximum. If you're pumping more: you're using your GPU inefficiently.

 

[Shifty Geezer]

Err... and how exactly this was THE WAY on PS2?

Because it couldn't apply multiple textures in a single pass. If you wanted an object to have 4 textures, you had to draw it 4 times. The sysmte was designed for very fast throughput of lots of simple triangles, which were layered into more complex images, in contrast to modern GPUs which are designed to do as much as possible in one pass.

 

[psorcerer]

Because it couldn't apply multiple textures in a single pass. If you wanted an object to have 4 textures, you had to draw it 4 times. The sysmte was designed for very fast throughput of lots of simple triangles, which were layered into more complex images, in contrast to modern GPUs which are designed to do as much as possible in one pass.

Ok, so this was not THE WAY, it was a compromise to do something that was impossible to get by other means.
But here, in article about NG2 we hear that there was some "architecture strength" exploited and not a compromise for an architectural weakness.