*Game Development Issues*

[upnorthsox]

Where did you get this insane idea?!

LOL! I knew that'd get you to respond on a Tuesday morning, hope you didn't spill your coffee.

Just have a look at that table, this makes more sense.
Obviously not having the ability of directly reading back vram from Cell at decent speed is retarded, on the other hand the workaround works perfectly fine, just makes things more complicated.

Indeed.

 

[Arwin]

I'd guess SPU pixel shading (deferred) and post processing would be much simpler (and cheaper).

But can someone tell me, as far as I remembered you could have the framebuffer in XDR memory, where SPU post-processing should be fine. Can't the RSX even send its output to an SPU more or less directly?

 

[T.B.]

But can someone tell me, as far as I remembered you could have the framebuffer in XDR memory, where SPU post-processing should be fine. Can't the RSX even send its output to an SPU more or less directly?

Yes, you can output to XDR which is fine for SPEs. This is however not the most efficient, which may or may not matter in your case. You can also render to DDR and then have RSX push the data to XDR, which gives you all kinds of "tiling" advantages, when it comes to your XDR budget.

About the "deferred on the SPUs"-thing, I'd really like to see someone do that, especially with cubeshadowmaps for pointlights. I've been throwing that around in my head for some time, but couldn't really come up with an efficient caching system for that. Without shadows, it's certainly doable. I'd guess 1080p at 30Hz if you have maybe four or five pointlights on screen.

 

[Cyan]

This is really a key point. You *do not want* one version to be inferior to the other. People will be less likely to buy your game for one platform if it looks better on the other platform. Nobody wants to play a gimped version. And we do want the money. Simple as that.

Another thing people often falsely believe is that there is a cell-optimal-model and an SMP-optimal-model. This is not really true, at least not for games. Almost everything you do runs faster in the "cell-model", even if you run it on Xenon. It also results in much cleaner code, which I guess people who haven't worked on large games cannot really understand. A lot of code is written with unmissable deadlines two days away, so if people simply multi-thread around to fork off some component and run it in parallel, you will see some pretty fun race-conditions.

So this is not really a 360-vs-PS3 thing. Both architectures need this model *badly* to perform well. 360 is a bit more friendly towards the old ways, which is an advatage early in the cycle. It might even be an advantage for the entire cycle, at least for performance-light titles. It's no free lunch, however.

Do you mean by that that the 360 is better at things you see and the PS3 is better at things you don't see?

Two news regarding the subject of this thread. The PS3 is the lead platform for Mirror's Edge development -it was already pointed out by jandlecack, but it went unnoticed-

Mod Note: Scans not allowed. -AlS

While Bethesda still prefers the "ease" of development on the 360's SDK platform

http://pc.qj.net/Bethesda-prefers-Xbox-360-as-native-dev-platform-hints-Fallout-4/pg/49/aid/122837

Peter Hines of Bethesda Softworks was recently in London to show off the upcoming Fallout 3 (Xbox 360, PlayStation 3, PC). While doing so, he was asked a few questions that revealed some interesting answers.

In a conversation with Tech radar, Hines was quoted as saying that the devs over at Bethesda prefer the Microsoft Xbox 360 as the model platform where they make their games first. The PC and the PlayStation 3 came in later.

Hines explained that the primary reason why they didn't develop natively on the PC is because they wanted to avoid guessing what configurations people already have on their systems. Combinations of RAM sizes and graphics card setups would have been troublesome to deal with for the devs.

"Obviously we are more familiar with the Xbox because we are familiar and the other thing is that the Xbox is much easier to take to tech shows," said Hines as he explained that it was handy to have the hard drive pop in and out off Xbox 360s so they wouldn't have to carry the whole thing while they travel.

 

[T.B.]

Do you mean by that that the 360 is better at things you see and the PS3 is better at things you don't see?

I'm not really sure how to interpret that question, to be honest. What do you mean with things that you "do or don't see"?

 

[woundingchaney]

I'm not really sure how to interpret that question, to be honest. What do you mean with things that you "do or don't see"?

Im assuming he is referring to estimated gpu and cpu hardware power with gpu being things you see and cpu being things you dont see.

 

[jandlecack]

One could argue that with the Cell SPE's being so powerful for number crunching and the like, the physics in any PS3 game can reach new levels in terms of how dynamic, realistic and advanced they are.

I don't know if that's something you couldn't "see" but it's definitely different from more AA.

 

[Carl B]

Guys I want to emphasize that this thread isn't for the purpose of comparing multiplatform titles to see which is better, it's for discussing actual multiplatform development. So... I'd say keep the back-and-forth to the Games forum inside the Star Wars thread, and use this thread for asking why something is the way it is, or adding insights where possible.

 

[Cyan]

I'm not really sure how to interpret that question, to be honest. What do you mean with things that you "do or don't see"?

Sorry for the delay. Well, I meant if it's easier for developers to untap the power of the GPU (X360) but, what about the perception and the things you hardly see in a simple image, more evident in other circumstances? :smile:

I mean, if I understood developers correctly, there's problem when the developer manually determine how much pixel shaders and how much vertex shaders he or she will use, as this causes a loss of 20-25% in performance, depending on the given situation, while in the case of X360 there shouldn't be any performance hit because of the US.

Anyways, back on topic, what about lighting, flexible and deformable objects' physics and so on, relativistically moving on a three-dimensional plane, like most games nowadays in a realistic way, or showing shockwave propagation in a real life like explosion, collisions.., or gravitational point lens with a great lighting engine? It does sound to me like the Cell still has some untapped potential in that regard, if I understood correctly.

Cheers!

P.S: The GPU in the XBox360 is quite unknown to many developers, though. Other features on the Xbox360 will eventually be exploited..., like tessellation (afaik, only Rare makes good use of that great feature) but they have to figure out how to use it and of course it needs to be enabled and present in the SDK for them to even figure out how to use it, since, imo, it's the future.

 

[T.B.]

Sorry for the delay. Well, I meant if it's easier for developers to untap the power of the GPU (X360) but, what about the perception and the things you hardly see in a simple image, more evident in other circumstances? :smile:

I think it's been stated quite a few times that Xenos is a lot easier to tune for than RSX. Basically both have a set of knobs that you can tweak to achieve maximum performance. Only RSX has like three times as many. So the chances of you missing a knob and messing up your performance are much larger on RSX. This has alot to do with understanding the hardware and your software and how they interact.

I mean, if I understood developers correctly, there's problem when the developer manually determine how much pixel shaders and how much vertex shaders he or she will use, as this causes a loss of 20-25% in performance, depending on the given situation, while in the case of X360 there shouldn't be any performance hit because of the US.

Well, 20% seems like a pretty random number. Of course, RSX can't move processing power from fragment processing to vertex processing, so you may be vertex-processing limited. In that case, PS3-wisdom tells you to move some of your vertex load to the SPUs (assuming you've really optimized the shaders, of course). Skinning is a good example of that. Of course, you will have to write a skinner, or use Sony's or whatever. This is of course extra work.

On the 360, Xenos will move processing power over to vertex processing, which *does* reduce fragment processing power. So it's not free. It's just automatic.

Anyways, back on topic, what about lighting, flexible and deformable objects' physics and so on, relativistically moving on a three-dimensional plane, like most games nowadays in a realistic way, or showing shockwave propagation in a real life like explosion, collisions.., or gravitational point lens with a great lighting engine? It does sound to me like the Cell still has some untapped potential in that regard, if I understood correctly.

Not sure you want relativistic physics in your games. It messes with the head.
Same with gravitational effects on light, really. Once your rays are bending, matrices just don't do it anymore.
Anyway, parity is king (unless you're exclusive), so don't expect gameplay relevant changes. Maybe someone makes a nicer water-surface or particle physics. Expect people to speed up rendering with the SPUs.

P.S: The GPU in the XBox360 is quite unknown to many developers, though. Other features on the Xbox360 will eventually be exploited..., like tessellation (afaik, only Rare makes good use of that great feature) but they have to figure out how to use it and of course it needs to be enabled and present in the SDK for them to even figure out how to use it, since, imo, it's the future.

I think by now developers have had enough time with Xenos to have a good idea about what it can and cannot do. Of course there will be new techniques comming up, but it's not like it's an unknown quantity.

 

[AlNom]

Basically both have a set of knobs that you can tweak to achieve maximum performance. Only RSX has like three times as many. So the chances of you missing a knob and messing up your performance are much larger on RSX. This has alot to do with understanding the hardware and your software and how they interact.

hm... if you carefully tweak some of those "RSX knobs", would it also benefit Xenos (any high-level/non-specific examples you may share ) ? Or are all/the major 'knobs' more platform specific, regarding the different handling of fragment or vertex processing loads? I mean in a similar way that the stricter coding practise on Cell can benefit development on other platforms.

 

[TrungGap]

Better as in realistic and representative of what real reflections look like.

I'm not sure what is artist real intention, however there are many kind of materials with different reflection characteristic. Unless you know the artist true intention of the material, what you're saying is wild conjecture.

https://www.seewald.com/images/Italy/venice-floor_reflections.jpg

http://adampolselli.com/photos/floorreflection.jpg

http://i.pbase.com/g3/97/654797/2/66626902.KGV916V7.jpg

http://www.heavenlytouchstonecare.com/adminpanel/images/P1010099a.jpg

 

[T.B.]

hm... if you carefully tweak some of those "RSX knobs", would it also benefit Xenos (any high-level/non-specific examples you may share ) ? Or are all/the major 'knobs' more platform specific, regarding the different handling of fragment or vertex processing loads? I mean in a similar way that the stricter coding practise on Cell can benefit development on other platforms.

It's pretty much all over the place. Of course, general performance-coding practices apply to all architectures, but apart from that there is no general yes/no here. Deep down inside, on the raster and execution-unit level, they behave remarkable similar, so there you can get ideas for one architecture from the other. Everything that accesses memory is vastly different, so you have to tweak for each system. The 360 simply has a much simpler memory system, so a lot of the balancing things just don't apply here. Same for all things render-target, where you have VRAM vs. EDRAM and a resolve system. Pretty different stuff.

I'll keep the actual examples to myself (NDA and all), but while I'm writing this, I'm waiting for a profiling run to finish, as our deferred shader runs at a good 100% ALU load on Xenos but at 85% on RSX. Fixing that will get the PS3 to 30fps@1080p, for a baaad scene with 11 pointlights. Fun stuff.

 

[thatdudeagain]

It's pretty much all over the place. ...

I also believe that NAO stated that through using some tricks on the RSX he was also able to speed up performance on the Xenos....

 

[Squilliam]

How does the Xbox 360 Tessellator compare with the technique on the PS3s Cell for vector geometries and can these two different approaches be used simultaniously to improve the visuals of multiplatform games whilst maintaining visual parity?

 

[Cyan]

It's pretty much all over the place. Of course, general performance-coding practices apply to all architectures, but apart from that there is no general yes/no here. Deep down inside, on the raster and execution-unit level, they behave remarkable similar, so there you can get ideas for one architecture from the other. Everything that accesses memory is vastly different, so you have to tweak for each system. The 360 simply has a much simpler memory system, so a lot of the balancing things just don't apply here. Same for all things render-target, where you have VRAM vs. EDRAM and a resolve system. Pretty different stuff.

I'll keep the actual examples to myself (NDA and all), but while I'm writing this, I'm waiting for a profiling run to finish, as our deferred shader runs at a good 100% ALU load on Xenos but at 85% on RSX. Fixing that will get the PS3 to 30fps@1080p, for a baaad scene with 11 pointlights. Fun stuff.

Pretty interesting, as usual. It's a weird scene, as you describe. The 360 should have the edge there. The great advantage of the USA (this stay for unified shaders architecture) is that when vertex or pixel computing STALLS, it can dynamically give more resources (the ones that are waiting) to the heavy process (adding those to the ones that are working).

And yes, for the litlle I know (I'm not an insider, keep this in mind people please), is possible tune up the process ( I think this is something like if (ALU stall time > $maxtimevar) { _commuteALU('pixel'); } )

Did you try this?

As for the parity, while both consoles behave remarkable similar on some situations, I'm against it because every 360 multiplatform game costs the same as the PS3 version. I find it completely unfair, to be honest. Sadly, I just own one console, and I'm not wealthy either.

I don't see much advantage to paying the same as I don't understand why I have to pay the same for the good games, or at least my favourite games, regardless the costs of the game.

Minimal technical differences aside, if the game is an AAA title on both consoles, PS3 and 360 users would be pretty happy but I wouldn't be as happy as most PS3 users.

I mean..., why do I have to pay 70€ for a version of a game when the developers did put a lot less more resources on the version I'm playing? I feel like I'm paying the extra costs and I am not receiving any benefit at all in return for my monetary contributions because I have the same game as PS3 users.

 

[T.B.]

Pretty interesting, as usual. It's a weird scene, as you describe.

Used to be our main-menu backgound. Ugly as hell, but a nice test scene.
The new one is much prettier, but less extreme with the light...

The 360 should have the edge there. The great advantage of the USA (this stay for unified shaders architecture) is that when vertex or pixel computing STALLS, it can dynamically give more resources (the ones that are waiting) to the heavy process (adding those to the ones that are working).

Slight misunderstanding here. Both platforms basically process the same shader. The difference is that the ALU on RSX is stalling 15% of the time, while Xenos doesn't. So the 360 wins. (Or used to. )
For the most part, 100% load is good. It means you're getting your money's worth somewhere in the pipeline.

I mean..., why do I have to pay 70€ for a version of a game when the developers did put a lot less more resources on the version I'm playing? I feel like I'm paying the extra costs and I am not receiving any benefit at all in return for my monetary contributions because I have the same game as PS3 users.

I think you have a few things mixed up there, sorry. First of all, the price of a full-price title is not based on development costs. If it was, nobody would have been able to pay for GTA4.
Second, programmer time isn't all that much of a money-sink. Programmers may be hard to get, but when compared to art teams and things like that, we are few. So most cost is shared between platforms anyway.

So at the end, you'll always pay full price. If the game cost 1M to make or 100M doesn't matter. As long as we can market it as a full-price title, we will. Hopefully, the 100M title sells more units, though.

The extra price of development is our problem. If we think one version is too expensive, we'll cut it. But that will never make a game cheaper for anybody.

Just as a thought experiment: If I'd tell you I'm making your favourite game, but it runs much smoother on the PS3, would you want me to put more resources into the 360 version, or do you want me to stop, as soon as both versions had the same amount of my attention?

 

[Cheezdoodles]

I think you have a few things mixed up there, sorry. First of all, the price of a full-price title is not based on development costs. If it was, nobody would have been able to pay for GTA4.

Of course we would have (Costs/expected sales)
is rather simple math. A very pessimistic estimate for GTA4 would be 10million sales, so unless the costs where $ 0.6 billion or more, paying for the costs would be cheaper than paying the price we pay now. Even when we factor in RRoI

So at the end, you'll always pay full price. If the game cost 1M to make or 100M doesn't matter. As long as we can market it as a full-price title, we will. Hopefully, the 100M title sells more units, though.

There are budget titles that are released on budget prices, just not enough of them.

The "full" price is just some standart price that the publishers and manufacturers have agreed on, just like any other cartel. (not necessarily "agreed" in the littural meaning, but just like in any business with limited competition, prices stay higher than they should longer, because its rarely in anybody's business to be competitive on price (depending on industry).)

If this business was under optimal market conditions, games would be sold at margin. Sadly it is not, but with digital distrubution on its way, devs can become more and more independent from their publisher and market powers may eventually begin to come into swing.

 

[Quarke]

Pretty interesting, as usual. It's a weird scene, as you describe. The 360 should have the edge there. The great advantage of the USA (this stay for unified shaders architecture) is that when vertex or pixel computing STALLS, it can dynamically give more resources (the ones that are waiting) to the heavy process (adding those to the ones that are working).

And yes, for the litlle I know (I'm not an insider, keep this in mind people please), is possible tune up the process ( I think this is something like if (ALU stall time > $maxtimevar) { _commuteALU('pixel'); } )

Did you try this?

As for the parity, while both consoles behave remarkable similar on some situations, I'm against it because every 360 multiplatform game costs the same as the PS3 version. I find it completely unfair, to be honest. Sadly, I just own one console, and I'm not wealthy either.

I don't see much advantage to paying the same as I don't understand why I have to pay the same for the good games, or at least my favourite games, regardless the costs of the game.

Minimal technical differences aside, if the game is an AAA title on both consoles, PS3 and 360 users would be pretty happy but I wouldn't be as happy as most PS3 users.

I mean..., why do I have to pay 70€ for a version of a game when the developers did put a lot less more resources on the version I'm playing? I feel like I'm paying the extra costs and I am not receiving any benefit at all in return for my monetary contributions because I have the same game as PS3 users.

I'm sure many more PS3 users have asked why they're paying full price for an inferior 360 port than the other way around. The PS3 version of multi-platform games, are for the most part, compromised because of the 360. A 360 DL DVD has about 7GB of capacity and the PS3 version is usually the same size so the quality or amount of the art assets has already been compromised. If the game uses streaming, then the size of the world as well as the quality of the textures has severely been compromised. Additionally, some developers don't even bother to improve the quality of the sound and just transfer over what they mastered for the 360 version.

 

[Cyan]

Used to be our main-menu backgound. Ugly as hell, but a nice test scene.
The new one is much prettier, but less extreme with the light...

Slight misunderstanding here. Both platforms basically process the same shader. The difference is that the ALU on RSX is stalling 15% of the time, while Xenos doesn't. So the 360 wins. (Or used to. )
For the most part, 100% load is good. It means you're getting your money's worth somewhere in the pipeline.

:smile: Sorry, I just thought it was the long way around (that the RSX with a 85% load-out had a 15% of its power unused while the 360 was running at a 100% load, putting the console on its knees and stalling, as I know very little). Nice to know you got the problem fixed T.B ., which is great, with some extra work on the PS3, I'm guessing here, excuse me if I'm wrong...

I think you have a few things mixed up there, sorry.

Telling you the assumption I've made leads me to the next point. I tried to tell you this before but I couldn't find the right words, as it sounded like a trifle.

First of all, the price of a full-price title is not based on development costs. If it was, nobody would have been able to pay for GTA4.

GTA IV is a totally different beast as its development costs reach stratospheric levels, although it would be nice to know the individual resources they put on each version.

Second, programmer time isn't all that much of a money-sink. Programmers may be hard to get, but when compared to art teams and things like that, we are few. So most cost is shared between platforms anyway.

So at the end, you'll always pay full price. If the game cost 1M to make or 100M doesn't matter. As long as we can market it as a full-price title, we will. Hopefully, the 100M title sells more units, though.

The extra price of development is our problem. If we think one version is too expensive, we'll cut it. But that will never make a game cheaper for anybody.

In essence, what I tried to say is just that it looks like for developers doing their job well means bringing in a lot of money on the PS3 version. For other users it's pretty damn great. For me, if they sell me my version (360) at the same prize as another (PS3) version, it's not that great. Is that a job well done or bad done?

Each platform has its virtues. Whenever Rage (this is just an example, I'm not sure if I want to buy it) comes out PS3 owners will play it on a single disc. I'll play it on two. I'd like developers to hand over the same dosh than in the PS3. After all, for all the good it is to me, if they are going to bleed me dry the same €....

But they don't, they put less resources on my version. What can't be right is the fact that they place all the developers on staff to "squeeze" the PS3 tight and dry while they leave the cleaning lady with the 360 version and see what happens.

Looks to me it's a common practice these days.

Also, well.., it isn't justifiable paying different prices for the quality, since the final quality is the result of many other different things, like the programmers' talent, as well as the machine or the tools, but it's the same game in the end.

We are back to square one then the PS3 version should be costlier because there are more people programming on it, after all once we got to the point, if we are going to put the economic aspects on the scales, let't put them all.

I'll probably buy all the games I like regardless the price while thinking of more debt very soon to come, as I am a very weak person and easily influenced anyways.

Just as a thought experiment: If I'd tell you I'm making your favourite game, but it runs much smoother on the PS3, would you want me to put more resources into the 360 version, or do you want me to stop, as soon as both versions had the same amount of my attention?

Fair question. You tighten my neurons there, I have to admit. I think I understand your point. You have a schedule and you don't want to miss the deadline and launch the game in time, and it probably means you are working overtime because the PS3 version is lagging behind. It adds stress to your well being.

Moreover, if we take your game as a whole you'll probably need -let's say- 10 new programmers to work on the PS3 version, while the other version isn't getting any attention at all.

That's what I don't find fair for multiplatform programmers like you who are working more than anyone without that much benefit in the end, except if you find happiness in making others happy, which is nice. By others I mean PS3 users. I, as a X360 user, don't care much. If the the game is great, it'd make me happy too, of course

Cheers T.B!