2 Cell Processors and an RSX GPU (Hypothetical)

[aldo]

I had heard that originally Sony had planned to go with 3 Cell processors instead of 1)Cell + RSX. We have also read recently that the PS3 specs are not written in stone and there will be some minor upgrades, albeit most likely to memory, but for curiosity's sake:

According to some the Cell costs Sony anywhere from $30 to $100 each. If it costs anywhere around the $30 mark and Sony decided to change the PS3 spec to 2 Cell processors before release would this be a benefit or just more CPU power or cores that may never be used? Would this be potentially too much power in proportion to the GPU? Is that possible?

Would it be easier to code across the two PPUs as opposed to a PPU and an SPE on the same chip? How powerful would it be using only the 2 PPUs if that were the case as opposed to the SPEs?

What would the impact be on developers developing for a single Cell PS3 and then finding out they updated to 2 Cells? Would this be a plus, minus, or could they just ignore the chip on games currently in development and it be a wash?

-aldo

 

[pcostabel]

I had heard that originally Sony had planned to go with 3 Cell processors instead of 1)Cell + RSX. We have also read recently that the PS3 specs are not written in stone and there will be some minor upgrades, albeit most likely to memory, but for curiosity's sake:

According to some the Cell costs Sony anywhere from $30 to $100 each. If it costs anywhere around the $30 mark and Sony decided to change the PS3 spec to 2 Cell processors before release would this be a benefit or just more CPU power or cores that may never be used? Would this be potentially too much power in proportion to the GPU? Is that possible?

Would it be easier to code across the two PPUs as opposed to a PPU and an SPE on the same chip? How powerful would it be using only the 2 PPUs if that were the case as opposed to the SPEs?

What would the impact be on developers developing for a single Cell PS3 and then finding out they updated to 2 Cells? Would this be a plus, minus, or could they just ignore the chip on games currently in development and it be a wash?

-aldo

Never going to happen. One Cell is plenty powerful. Hypotetically though, multiple Cells should be handled transparently by the OS, as long as you have enough threads to use all the SPEs.

 

[borntosoul]

i was thinking this might happen as well (that was quite some time ago). the cost is the main thing here, if they could afford to do it. if they need another one to help with rsx. one of the demos at e3 ws run using two cell cpus only. who knows how or what will happen. buts its not likely - it sure would be a shock to everyone if it did.

 

[randycat99]

Maaaayan, would a dual-Cell+RSX be one cool mo-fo! Who really knows...maybe that tiered yields strategy puts some new opportunities as far as budgeting Cell into a console. (Hey, it's ok to let your mind run free once in a while, as long as you are fully admitting that is what you are doing )

 

[LunchBox]

i thought the cell-based gfx engine was a 4-cell design???

 

[standing ovation]

I had heard that originally Sony had planned to go with 3 Cell processors instead of 1)Cell + RSX. We have also read recently that the PS3 specs are not written in stone and there will be some minor upgrades, albeit most likely to memory, but for curiosity's sake

Your original hunch is probably right. Memory, after all, is one of the slowest links in the electronics food chain. So it only stands to reason that improvements in this area are likely to have a noticeable impact on performance.

Speaking of performance ...

Because of its knack for accelerating operations, Cell behaves more like a general purpose GPU than a traditional CPU. And like most graphics chips, keeping it saturated will be, according to experts, not a trivial undertaking.

Besides, IBM seem to be having enough trouble engineering single-Cell PS3s that can be cooled with fans.

 

[aldo]

I submitted the initial post as I was heading out the door from work and was somewhat hesitant to come back and look at the replies tonight. I'm surprised and glad to see that you guys took this in the right vein as simply hypothetical.

I am of the opinion that if Sony made a sudden design change like this, the initial response would be like randycat's that it would be perceived as one cool mo-fo! by techs and Playstation enthusiasts, but then devs would start chiming in saying, "We can barely figure out how to access the power of one Cell let alone 2" and wondering why Sony didn't put the extra money toward resources that would benefit everyone immediately like more memory or a hard drive .

Still, Neil Young at Electronic Arts’ comment that they are "trying to hold to a [budget] increase of 50% over the current generation" for code-powered effects, as there are only 4 SPE’s left for “physics, particles, AI, and so onâ€￾ makes you wonder if some devs would be elated to have a second Cell on board.

-aldo

 

[London Geezer]

The problem with that configuration is that 2 Cell chips will probably not give you double the performance of 1. Not even close.

I think performance would be greatly improved by increasing other aspects of the architecture, since i'm not sure PS3 games will be CPU limited. Or, if anything, the existing Cell could be souped up. Making it OOO for example seems to be a big deal to some developers. They're already nagging about having to code for one CPU and 7 SPEs, i can't imagine what they'll say if there were double that!

Personally, i'd rather have Xenos instead of RSX, and one Cell should be enough, if i had to choose.

 

[Guden Oden]

I need another hypothetical topic on this subject about as much as I need my penis cut off...

The ONE combination of hardware that did NOT do the rounds on the internet before the unveiling of PS3 was 3 cells + GPU. That particular setup must be the only one nobody speculated over.

Besides, we know now more than two cells in their current implementation requires an external crossbar switch to route traffic between the chips (there is only one XIO interface on each cell, so it either connects to another cell or to a switch which in turn connects to other cells.

As for "standing ovation"'s speculation about cooling, that isn't likely to be much of an issue really. In IBMs cell presentation, the 3.2GHz cell runs at 1.1V and dissipates no more power than any other modern high-performance general computing processor. Actually, compared to a 3.2GHz P4, it dissipates considerably less. So so much for that FUD...

 

[aldo]

I need another hypothetical topic on this subject about as much as I need my penis cut off...

Sorry to hear about your penis...on many levels, but I will guarantee this is my first and last hypothetical topic.......that I start.

BTW, interesting point on the necessity of an external crossbar switch. Is that a major hurdle? Cost, design, both?

-aldo

 

[Guden Oden]

BTW, interesting point on the necessity of an external crossbar switch. Is that a major hurdle? Cost, design, both?

It's already been shown by IBM I think, in their 4-cell rackmount demonstration boards. Or it might have been 2-cell too, but I'm pretty sure it was 4.

It's not really as much an issue of cost or design issues, but more one of practicality. Consoles have to be as streamlined as possible (though some may argue the saturn annd PS2 were two exceptions from that rule ). Multi-chip cells and bridges and such clutters up the design, in addition to taking up more space, costing more money, drawing more power and dissipating more heat.

Besides, PS3 design has already been announced. It's 1 cell + 1 GPU, and I'll wager already mentioned penis on it staying that way until launch. Possibly, if sony decides to change any hardware specs, they might increase XDR RAM or something like that, but I wouldn't expect them to add another entire cell core. That'd just be too big of a change. Besides, considering how big the cell chip package + memory layout is, one might wonder if it could even physically fit in a device the size of the announced PS3 minitower case...

 

[standing ovation]

The problem with that configuration is that 2 Cell chips will probably not give you double the performance of 1. Not even close.

I am not so sure about that ...

Because of their penchant for distributed computing, multi-Cell systems will probably yield order of magnitude increases in performance. The problem, however, is sustaining this level of corporation in the real world.

Getting several dedicated Cells to work together will undoubtedly increase the complexity of software development. More processing power on tap essentially means more operations can execute simultaneously. So anything that has a remote hand in scheduling / vectorizing apulets (compilers, operating systems, etc.) will have to consider this. Ultimately, applications running in a multi-Cell system must be more parallelizable than in a single-Cell one.

Who knows? Maybe IBM were thinking about software designers when they decided to go with one of the simplest single-Cell designs for PS3?

 

[Megadrive1988]

having 2 RSXs would be nicer than 2 CELLs, IMO.

 

[standing ovation]

As for "standing ovation"'s speculation about cooling, that isn't likely to be much of an issue really. In IBMs cell presentation, the 3.2GHz cell runs at 1.1V and dissipates no more power than any other modern high-performance general computing processor. Actually, compared to a 3.2GHz P4, it dissipates considerably less. So so much for that FUD...

This is the basis of that argument.

Cell operates in a range from 0.9 to 1.3 volts, Kahle of IBM said, but the developer trio hopes to keep it at the low end of that range to get power consumption down to something that can be cooled with a fan. Kahle would not disclose performance or power consumption figures.

"I once told them I estimated the architecture would consume 80 watts at 4 GHz, and they didn't blink," said Microprocessor Report's Krewell.

Source: EETimes.com

 

[London Geezer]

Is that such a big issue? X360 is using some form of watercooling (last time i checked). So PS3 could do the same.

 

[Josh378]

having 2 RSXs would be nicer than 2 CELLs, IMO.

You just might have an interesting idea there...2x RSX in SLI? Would that be sick or what?

Maybe 384 MB of DDR3 ram for the 2x RSX in SLI and 256 for the CELL?

It could work....I'll stop here tho for wishful thinking....


-Josh378

 

[Tahir2]

..

What about 2 PPE's each with 4 SPU's and 1 RSX?

I believe a similar setup was previously speculated but wouldn't such a design balance the system and keep costs lower than the proposed hypothetical 2xCell+RSX?

 

[Guden Oden]

This is the basis of that argument.

The colored section you quote is not an actual quote of anything Kahle said, or anyone else for that matter. It's the writer of the article's own interpretation, and it may or may not be very accurate...

Again, cell does not have a monstrous power draw. It draws less than a P4 at target frequency (3.2GHz), just go check IBM's PDF on the subject, there's a nice table there which shows voltage and wattage for various clock speeds.

You're fudding, for whatever reason. Either because out of ignorance because you simply don't know any better, or else because you're trying to badmouth the console, I'm not sure which and frankly it doesn't even matter. Again, just go check the PDF, it'll clear up all your confusion on the subject.

 

[Barbarian]

I'd be happy with just getting the 8th SPU back.

 

[standing ovation]

The colored section you quote is not an actual quote of anything Kahle said, or anyone else for that matter. It's the writer of the article's own interpretation, and it may or may not be very accurate...

True. But considering the source and subject matter, I doubt this is an Arts major pontificating about science. The author is a lot more credible than you give him credit for. As is the Editor-in-Chief of the Microprocessor Report.

So if I am guilty of anything it is assuming that electrical engineers know watts up.

Network Systems DesignLine is CMP Media's sixth in a series of vertically micro-segmented Web sites for design engineers involved in designing electronic systems. CMP's Electronics Group has a high degree of expertise in the networking systems design area. In addition to Annie Lindstrom's broad knowledge of this sector, CMP's Robert Keenan, Loring Wirbel, John Walko, and Rick Merritt* are distinguished editors in the space. In addition, the site has access to a deep library of how-to technical articles, product information, and news content through CommsDesign.com.

SOURCE: CMP Media Press Releases

[size=-2]* Rick Merritt, the journalist in question, is EE Times' Advanced Systems Architecture Editor-at-Large.[/size]