Would you rather have the Cell work on Graphics or Physics?

[BenQ]

I can't go into any thread these days without hearing people speculate about the ways in which the Cell can help out the RSX.

I don't quite understand why though. My understanding is that the RSX will be an extremely powerful GPU, designed perfectly for its role.... rendering graphics.

Do some people have the idea that the RSX is somehow weak or limited? Why would it need any help from the Cell? Don't you feel the Cell could be better used doing things that CPU's are more traditionally good at?

I would like to see the Cell put to work at absolutely AMAZING Physics and AI. Would the Cell not be better equiped to handle these things rather than graphics? ( This is an honest question, I don't know ).

It just seems to me that I would MUCH rather be playing PS3 games with mindblowing physics and AI, than playing a game that has 5-10% better graphics, but "been there done that" physics, because the Cell had it's hands full helping the RSX at a job that it's ALREADY very good at.

I was under the impression that physics is an area that the Cell was supposed to be VERY good at. Did that turn out not to be the case?

 

[Powderkeg]

I think the majority of talk about Cell helping out the RSX is a waste of time. Very few developers are even going to consider that option, much less actually implement it. It's costly, time consuming, and the benefits are minimal at best.


From a developers point of view, it's better to do "good enough" with the graphics and get your game out sooner and cheaper than it is not invest the money, time and effort it would take to use Cell for rendering.

Especially if it's a multiplatform game, which most are.

 

[leechan25]

I think don't a developer has to choose. There is eough Cell to go around.

 

[scooby_dooby]

It's more in reference to the "Hardware Defining Games" like MGS2 was at it's time, or GT4, or Halo.

Usually 1st party, always exlusive, the occaisional game will absolutely push a system to it's limits, these would be the games, if any, that might use CELL to help with graphics.

 

[lip2lip]

because the evaluation physics package includes collision detection, basic vector physics requirements, refraction, and so on, it will probably be integrated into graphics creation at cell proccessing time. This is good, because with physics and data movement concerns being handled on the cell, other more linguistic commands can be handled by the ppu.

 

[Titanio]

Re: Would you rather have the Cell work on Graphics or Physi

Do some people have the idea that the RSX is somehow weak or limited? Why would it need any help from the Cell?

In most instances, it's not a matter of need, but want/desire. People always want to do more, and graphics remains a favourite of a lot of people. RSX doesn't mark a cap on what people want out of graphics.

Don't you feel the Cell could be better used doing things that CPU's are more traditionally good at?

Cell is not a traditional CPU. It could be good/great for some graphics tasks. In a lot of instances it's not about replacing work done on the GPU, but supplementing it..doing more on top of. But remember that GPUs can't do EVERYTHING you might ever want to do with graphics.

I would like to see the Cell put to work at absolutely AMAZING Physics and AI. Would the Cell not be better equiped to handle these things rather than graphics? ( This is an honest question, I don't know ).

I can't be sure either, but my guess is that graphics (some tasks, at the top end of the pipeline at least) and physics are both things Cell would be good/great at. AI is tougher to say, since there are so many different techniques.

It just seems to me that I would MUCH rather be playing PS3 games with mindblowing physics and AI, than playing a game that has 5-10% better graphics, but "been there done that" physics, because the Cell had it's hands full helping the RSX at a job that it's ALREADY very good at.

That's an interesting figure you've got there (5-10%). It'd be interesting to see how you quantified that..

RSX, like any GPU, has limits. CPUs are more flexible than GPUs. There is some stuff you could do on Cell that would be very difficult to do on RSX, and then some things you could do on Cell that would be in addition to what RSX was already doing.

I was under the impression that physics is an area that the Cell was supposed to be VERY good at. Did that turn out not to be the case?

Sounds like it's turning out well for physics, although there is little definitive comment.

All this aside, I think you could be doing some graphics work on Cell and still have "kick-ass" physics. Just not as much as you could have if you were doing just one or the other..but since when does a CPU ever just have one thing on its plate?

Ultimately it's up to the developer to decide where the priority is for their game.

 

[dukmahsik]

I think the majority of talk about Cell helping out the RSX is a waste of time. Very few developers are even going to consider that option, much less actually implement it. It's costly, time consuming, and the benefits are minimal at best.


From a developers point of view, it's better to do "good enough" with the graphics and get your game out sooner and cheaper than it is not invest the money, time and effort it would take to use Cell for rendering.

Especially if it's a multiplatform game, which most are.

agreed

 

[Farid]

What about both?

There's seven 3,2GHz processors, there's plenty raw performance for both graphic and physic computations.

A few months (Weeks, thanks to Anand stupidity), people didn't even know what they were going to do with a Dual Core CPU. Now that (some) people are finally positive about multi-cores, and start looking for innovative way to use theses cores, you'd want them to forget about it, and just throw everything into physics?

All games do not need a lot of calculations for physics. What those games are going todo with the raw power left unused?

BTW, using SPEs in addition of RSX doesn't mean anything about RSX other that it can work in parallel with SPEs for rendering tasks.

 

[leechan25]

I like cell because it become what a developer needs it to be.

It can help with graphics,

Act like a PPU

now we know it can preform task like an RPU

Sound and Video are no problem.

It give PS3 the best of both world by being close hardware with a full programable CPU.

 

[blakjedi]

I think the question started when Sony's presentation of RSX virtually always included the CELL. Cell and RSX are seen as a graphics combo accroding to Sony which includes not just teh framework of how they work together but also the numbers game... (51 billion dot poducts is not the RSX but RSX plus Cell for example)

 

[Megadrive1988]

RSX can be and is more independant as a graphics chip than Graphics Synthesizer. RSX does not really need any help from Cell for processing graphics. this was not true of the GS since the GS was totally dependant on the Emotion Engine.

with that said, RSX and Cell can work together on graphics. as others have said, there are plenty of resources in Cell to do both graphics and physics at the same time.

Cell could provide more data for RSX than RSX would be able to deal with, or the bus architecture would allow, and at the same time, be doing tons of physics processing

I think memory bandwidth will become the limitation before Cell runs out of processing performance.

 

[ralexand]

The funny thing to me how people just throw it out there as a trivial solution. "Need AA+HDR, just get the SPEs on the cell to do it." If it was so trivial why aren't most developers using SPEs now in their development. Almost every developer that have commented on their ps3 development have said they are using the ppu and the gpu only currently so using the SPEs obviously isn't trivial even in a traditional approach of doing the things that cpus do currently ie. physics, AI etc. It really is a cost vs. return on investment scenario, sure people could use the cell/RSX in a load balanced way to output graphics but does it really give you that much more over using a traditional approach vs. the cost in time and effort.

 

[Nite_Hawk]

The funny thing to me how people just throw it out there as a trivial solution. "Need AA+HDR, just get the SPEs on the cell to do it." If it was so trivial why aren't most developers using SPEs now in their development. Almost every developer that have commented on their ps3 development have said they are using the ppu and the gpu only currently so using the SPEs obviously isn't trivial even in a traditional approach of doing the things that cpus do currently ie. physics, AI etc. It really is a cost vs. return on investment scenario, sure people could use the cell/RSX in a load balanced way to output graphics but does it really give you that much more over using a traditional approach vs. the cost in time and effort.

The idea is right but your argument isn't so great. Most of those developers had very little time to actually get anything going on cell, not to mention they are working without completed development kits. What someone can easily do on Cell now compared to what someone can easily do on Cell 2 years from now is going to be vastly different.

I wouldn't doubt that there will be custom and perhaps even off-the-shelf libraries that will be built to do exactly these kinds of tasks on the SPEs. The problem is that no one has actually written them yet.

Nite_Hawk

 

[Titanio]

The funny thing to me how people just throw it out there as a trivial solution. "Need AA+HDR, just get the SPEs on the cell to do it." If it was so trivial why aren't most developers using SPEs now in their development. Almost every developer that have commented on their ps3 development have said they are using the ppu and the gpu only currently so using the SPEs obviously isn't trivial even in a traditional approach of doing the things that cpus do currently ie. physics, AI etc. It really is a cost vs. return on investment scenario, sure people could use the cell/RSX in a load balanced way to output graphics but does it really give you that much more over using a traditional approach vs. the cost in time and effort.

Well, first off, I think few enough people ever throw out ANYTHING as being purely trivial and simple. Usually these things are raised as questions and there's quite a debate that follows about feasibility. So I think that characterisation is wrong.

When things like this are discussed, I've always got the impression that it's more an issue of possibility than how frequently it will be done. I think it should be obvious that most games will do graphics on the GPU, the rest on the CPU, fullstop, or at least for near-term.

It's always the case with every system, every generation, that a relatively small subset of games really pushes the hardware and does more ambitious things, usually later on (but it seems more games get more ambitious as the console's lifecycle wears on). But they're the games that stand out and become exemplars for what can be done on a system. IMO, it's natural for people to get excited by, and to discuss, what the best games, technically, could do.

 

[liverkick]

It really is a cost vs. return on investment scenario, sure people could use the cell/RSX in a load balanced way to output graphics but does it really give you that much more over using a traditional approach vs. the cost in time and effort.

Can't we just say time will tell and leave it at that? Developers are only using the PPU at this point, and did anyone expect anything different? Folks are just getting their feet wet with this thing, which is what I thought everyone predicted in the 1st gen. The bandwith on the devkits is extremely limited from what I understand. Those SPEs will get the knock on the door soon enough.

Its seems when it comes to Cell, B3D either leans extremely pessimistic or perhaps overly optimistic.

Either way 2, 3, 5 years down the line Id be surprised if there arent methods used on this machine that achieve results that no one had remotely thought of in 2005.

 

[Shifty Geezer]

Like Leechan I believe the Cell is a versatile solution provider that can doa lot of things. Not necessarily all at once, but can turn it's hand to it. So we can talk about SPE's doing maybe physics, and talk about them doing AA, and ray tracing of sorts, but that doesn't mean all those features will be going at once. It just gives a selection for devs to pick and choose what works best in their product.

eg. For a racer, there's only so much physics you can model in a car. Okay, they could go all out and have volumetric gas combustion and friction on every moving part... But really, when you've got the cars done there'll likely be plenty to spare on maybe augmenting RSX with raytraced reflections or some graphics enhancement. Likewise a fighting game, where you'd perhaps have physical modelling of the fighters movements and cloth simulation etc., but maybe there's a bit left over to add AA. And on a flight sim, there's so little to it that Cell could...I dunno, realtime raytrace a landscape from 2 different textures . And Tetris! Well, real-time raytraced with full HDR Global Illumination.

Whereas a huge FPS might take up all Cell's resources on destructable environments and AI, and then it's *only* poor little RSX all on it's own. Which isn't bad! It's a meaty GPU and it'll turn out great visuals. But if you've got spare Cell-cycles, might as well burn them on something.

The limiting factor is mostly getting to use those cycles, and that'll likely come with the development of library functions. In a couple years time there'll quite possibly be off the shelf solutions for Cell functions, covering the gamut of it's capabilities, and devs will pick'n'mix what suits them.

 

[ralexand]

It really is a cost vs. return on investment scenario, sure people could use the cell/RSX in a load balanced way to output graphics but does it really give you that much more over using a traditional approach vs. the cost in time and effort.

Can't we just say time will tell and leave it at that? Developers are only using the PPU at this point, and did anyone expect anything different? Folks are just getting their feet wet with this thing, which is what I thought everyone predicted in the 1st gen. The bandwith on the devkits is extremely limited from what I understand. Those SPEs will get the knock on the door soon enough.

Isn't the bandwidth between the ppu and spe going to remain the same. I thought it was other bandwidth that would get upgraded. Either way it sounds alot like what we're hearing from MS and what we heard concerning the vu in the ps2.

So we can talk about SPE's doing maybe physics, and talk about them doing AA

Was it ever fleshed exactly how cell will do AA?

 

[Titanio]

Isn't the bandwidth between the ppu and spe going to remain the same. I thought it was other bandwidth that would get upgraded.

He was referring to the "other" bandwidth, I think (CPU-to-GPU). Although internal bandwidth in Cell should get a boost too, since it's underclocked right now.

 

[seismologist]

We've discussed this before.

It depends on the game. For something like Devil May Cry, cell should be used to push the graphics beyond what's possible with just the RSX, which I'm sure it will be.

 

[ralexand]

Isn't the bandwidth between the ppu and spe going to remain the same. I thought it was other bandwidth that would get upgraded.

He was referring to the "other" bandwidth, I think (CPU-to-GPU). Although internal bandwidth in Cell should get a boost too, since it's underclocked right now.

My point though was that the bandwidth in the chip shouldn't hamper from using spe in development now since the structure of the chip will likely remain unchanged.