• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

Sony's CELL and console design philosophy.

jimbo

Banned
The more I read about the current generation CPU's and GPU's and also looking back at the PS2 it seems what Sony is trying to achieve, goes against current trends in graphics. Because I just get the impression that Sony still stuck on the idea that the more polygons a system can push the better the graphics.

For example with the first PS2 Sony designed a CPU that was extremely powerfull for its time, could push a lot of geomtery, yet did not account for the GPU intensive developing that began taking place in the PC world. Graphics depended quite a bit more on the abilities of the GPU than the CPU. They forgot to add a number of features to their GPU that not only left them behind, but also bottlenecked a very powerfull CPU.

With the PS3 it looks like they were trying to do the same thing at first. They wanted to design a chip so powerfull, that it could push out trillions of polygons but didn't care too much to spend the same amount of money on a GPU as powerfull or at least make sure the CELL had the same capabilites as well. It's like they're trying to make a system where instead of using textures, vertex and pixel shader effects, they are trying to get to the point where they can just model every little bump on a road surface with real polygons. However when they realized CELL just wasn't going to be able to do the features needed in a GPU, they went to Nvidia.

It seems as if Sony is still trying to make pretty graphics with a CPU instead of a GPU.

I'm sure eventually we'll get to a point where we have so much power that we CAN individually model every single ridge, crack and chip of every single brick of a wall. But even when we do, the fact still remains, it's a waste of resources in the first place, and second it's a LOT more time consuming than using other methods of achieving the same thing through textures, bump mapping, displacement mapping, etc. Whose going to sit there and develop things like that when they can just let the GPU do it for them automatically?

Discuss please.
 
Everything will be done procedurally eventually, but I'm not so sure what we should discuss about :)
 
jimbo said:
It seems as if Sony is still trying to make pretty graphics with a CPU instead of a GPU.

If that's the case then the PS3 would have 4 CELLs linked together :lol I don't understand what we're supposed to discuss here; the PS3 will have a GPU which would be slightly more(or less) powerful than the Xbox 360/Revolution's GPU. It might have a more powerful proccessor but that doesn't mean its going to waste; AI and physics would benefit from whatever power is bottlnecked by the GPU.
 
TTP said:
Everything will be done procedurally eventually, but I'm not so sure what we should discuss about :)


Basically if you agree and if you think this is flawed or if it is the future.

sly said:
If that's the case then the PS3 would have 4 CELLs linked together.

Two actually. And yes, that's really what it was supposed to have.
 
As layman scum trying to deduce things as best I can, the "Nvidia concession" or whatever you want to call it looks like it'll be nothing but a positive for Sony's machine. Cell is a bad ass gaming processor, and while RSX may be a middle ground of sorts (as far as Sony ditching Plan A or whatever), the combo may end up being the best of both worlds when its all said and done.
 
hukasmokincaterpillar said:
As layman scum trying to deduce things as best I can, the "Nvidia concession" or whatever you want to call it looks like it'll be nothing but a positive for Sony's machine. Cell is a bad ass gaming processor, and while RSX may be a middle ground of sorts (as far as Sony ditching Plan A or whatever), the combo may end up being the best of both worlds when its all said and done.


Oh I don't doubt that, I'm sure Nvidia is going to do a great job at putting together a hell of a chip in a very short amount of time. What I'm talking about is Sony treating GPU's as an afterthought. Imagine what they could have come up with had they put the same amount of effort they put in CELL into a custom designed GPU!
 
"Let's make crazy expensive pie-in-the-sky chip designs, and then by the time they actually come out they'll be about the same level as shit we could have pulled off the shelf"
 
Kleegamefan said:
I like cheese....
me too
cheese.gif
 
I tend to agree somewhat..Sony (to date) has pushed for high polygon counts with stuff like textures and AA taking a backseat.... This is Sonys first console with an honest to god fully featured GPU... about damn time.
 
Things in computing tend to move towards general solutions. Eventually, cell's model of general purpose parallel units will be the way to do graphics. But obviously even Sony has made the concession that this is not the case just yet.
 
"cell's model of general purpose parallel units"

But but! MS sai the XCPU had more general purpose* processing power than cell!!!




*: pr bullshit, but we dont care, we gobble it up all the same!
 
Apple Computer said the cell wasnt powerful enough. Said power pc would be a better chip.

I would post the link if I had it, but read this info just within the past week or 2.
 
jimbo said:
For example with the first PS2 Sony designed a CPU that was extremely powerfull for its time, could push a lot of geomtery, yet did not account for the GPU intensive developing that began taking place in the PC world. Graphics depended quite a bit more on the abilities of the GPU than the CPU. They forgot to add a number of features to their GPU that not only left them behind, but also bottlenecked a very powerfull CPU.
GPUs developed the way they did in the PC world as an endrun around the limitations of CPUs more greatly constrained by the need to support legacy architectures. Game console design doesn't have such a burden in that regard so console manufacturers are more free to pursue greater advances in CPU design and overall system architecture (unless of course the console manufacturer opts to use PC parts...) which means potentially less emphasis is required on building a beefy GPU specifically to compensate for legacy weaknesses in the rest of the system design.

But putting slightly less emphasis on GPU design doesn't mean that teams of engineers from both Sony and Toshiba just "forgot" to add features. A little perspective here:

Playstation 2: Arrived on store shelves for the first time in Japan in March 2000
GC: Arrived on store shelves for the first time in Japan in September 2001
Xbox: Arrived on store shelves for the first time in North America in November 2001

PS2 went on sale a year and a half before competition arrived that presumably has the GPU features you're claiming Sony and Toshiba engineers simply "forgot" to include. Why are you overlooking such a simple factor that more greatly accounts for the disparities in GPU design between the systems rather than reaching for these scatterbrained theories of forgetful engineers completely missing what's going on in the PC world?


I'm sure eventually we'll get to a point where we have so much power that we CAN individually model every single ridge, crack and chip of every single brick of a wall. But even when we do, the fact still remains, it's a waste of resources in the first place, and second it's a LOT more time consuming than using other methods of achieving the same thing through textures, bump mapping, displacement mapping, etc. Whose going to sit there and develop things like that when they can just let the GPU do it for them automatically?
Normal maps and displacement maps involve starting with high poly models in the first place and if a next gen game makes extensive use of such techniques you're hardly going to see development teams avoid the work involved in creating models with much larger amounts of polys. Not to mention that the PS3 and Xbox360 are certainly NOT going to be slouches in the poly-pushing department no matter what other GPU features they may or may not emphasize. Every console next gen is likely to be capable of pushing tens of millions more polys than this gen's entrants did. So nobody's avoiding working with significantly more polys than they did this gen.
 
You know why Sony's doing this? Because they want to crush the 360 before he PS3 gets released with standard dumbass EB talk like what happened with the DC. It doesn't matter if it doesn't deliver, that hasn't stopped Sony before...
 
"PS2 went on sale a year and a half before competition arrived that presumably has the GPU features you're claiming Sony and Toshiba engineers simply "forgot" to include. Why are you overlooking such a simple factor that more greatly accounts for the disparities in GPU design between the systems rather than reaching for these scatterbrained theories of forgetful engineers completely missing what's going on in the PC world?"

Uhm I'm not saying they forgot necessarily but rather that they never tried to push the envelope on GPU's. They want to jump ahead of the curve with the CELL, and their PS2 CPU, but not the GPU. They're happy to use whatever GPU technology is available at that time, instead of trying to add new and revolutionary features.
 
kaching said:
GPUs developed the way they did in the PC world as an endrun around the limitations of CPUs more greatly constrained by the need to support legacy architectures.
That was how it started, but that was before it was realized that a GPU workload requires a different set of optimizations than CPU workloads tend to.

Now of course I don't think that it won't swing back in the other direction someday, but a GPU today definitely looks nothing like a CELL or a regular CPU internally -- and that's for a lot of very good reasons.
 
Norse said:
Apple Computer said the cell wasnt powerful enough. Said power pc would be a better chip.
CONTEXT. Apple said Cell wasn't good enough for their needs, and Apple needs something different than what Cell offers. That and the PPC core is actually the weakest part of the Cell, and weaker than an actual G4/G5. It's really just meant to be a coordinator for the real workhorses.
 
PS3 brute force approach was probably the best option at the time.

Since then, GPUs have come to the fore in the PC area. However, they are clearly designed around their need to be in a PC. Mostly they have lots of their own ram, their own T&L, etc. Almost like 'leave me alone you piece of shit PC, just let me get on with it'.

You could think of a console like a PC graphics card, but with a CPU on it. Lots of fast ram, good bandwidth etc. So the environment is different and you could argue you need a different solution.

For me, the best combination would seem to be Xenos + CELL. CELL does your T&L very efficiently (its basically a vector monster), and Xenos does the pixel shading, dedicating all its transistors to the task.
 
Top Bottom