Playstation 3's cell processor

[prochembro]

This is a new type or processor being developed by Sony, IBM and toshiba.

It's supposed to be the central chip in the PS3. I saw on some sites that this thing is capable of 1 tera flop per second processing power.


Is it just me or is that insane?
Anyone else know more about this chip?

 

[Mephistopheles]

From the little info I saw it seems as if these cell processors are capable of running in parallel very well, hence the name "cells". Therefore, you could just plug in 2, 4, 8, 16... cells, as many as you want.

They're also highly versatile in the sense that you can also use them to process video data, as far as I understand it... if anyone here has any more info, please tell me too!

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>

 

[poly4life]

A console that can execute 1 teraflops! You must understand, I am skeptical about that claim, since IBM has released no real-time benchmarks for this chip, as it is still in the making, which I presume. Computers that can calculate and execute many teraflops are VERY EXPENSIVE. It is hard for me to believe that IBM has developed a cell processor, specifically for the PS3 (this is what I've read), and the average consumer will pay no more than a few hundred dollars for Sony's darling video game machine. If IBM is worthy of their claims, this would be a monumental event. The P4 3.2E, for example, can, on average, execute between 4-8 GFLOPS. Basically, if IBM can prove true on their claim, I'll glady eat my words.

 

[Mephistopheles]

I think that what they meant is probably that the cell architecture can be parallelized so as to execute over 1 teraflops easily. They didn't say how many cells would be required. So if you can put together 128 cells with the same output as the 3.2E, then you'd have a teraflop-level machine....

Apart from that, I agree: around a teraflop <i>for each cell</i> would be a monumental breakthrough indeed. I don't think it would be easy to achieve...

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>

 

[poly4life]

Hmm, perhaps this technology is not completely unlike a processor with multiple cores. This is complete speculation on my behalf, but maybe this Cell processor has the POTENTIAL to run at 1 TFLOPS. So, it's not that each cell can execute 1 TFLOPS, but a multitude of cells - whatever that number may be - can execute 1 TFLOPS in (pseudo) parallel.

Continuing on, IBM would release this processor for the PS3, albeit all but a couple of cells removed. The maximum amount of cells that can be integrated into the processor would be put to use in the supercomputer business, for the technology to implement many, many cells at the consumer level is too costly. Again, this is complete speculation.

 

[Mephistopheles]

Yes, this is what I mean. In any case, it's kind of "build your own MCM on the fly" with these small pieces (cells) of silicon.......

MCM, BTW, are the big-caliber chips like power5 - they're not from a single wafer, they're a fusion of pieces... So cell processor groups would not be much unlike MCMs, except that it's more the end user, not the manufacturing company, that scales the chip/processing power according to their needs....

(complete speculation, of course)

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>

 

[P4Man]

> I saw on some sites that this thing is capable of 1 tera
>flop per second

That is based on a misunderstanding. insiders have said it will be a "terrible flop", not teraflop 😉

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[Mephistopheles]

ROFL! terrible flop... :lol:

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>

 

[jim552]

While the "Cell Processor" being devloped by IBM/Sony/Toshiba, may end up being completely different I think that it is to close to overlook the name similarity between it and the "Cellular Computering" project(s) IBM already has underway.

I am convinced the IBM/Sony/Toshiba "Cell Processor" is a scaled down version of IBM's "Cellular Computing" project(s).

So with my stated position in hand this is what I think.

The processor configuration to be used is based on the research and concepts from IBM's Supercomputer research. (Specifically "Cellular Computing".)

The thought process is, "As computers have gotten faster, access to memory has become more important, and has become a bottleneck. Therefore to fix this, let's distribute the processing power, to computer cells where each cell has processing resources, memory resources, and I/O resources and then interconnect the cells."

This is really nothing more than taking what Supercomputers in general have become, (i.e. Entire systems connected to extremely fast LAN's), and shrinking that concept down to the chip level.

In it's full glory "Cellular Computing" will enable cells to pass instructions, and data back and forth over their communications bus. The communications structure between cells will enable cells that have failed to be bypassed. (Systems based on this are essentially "self healing" to a dregree.) Data will find cells to reside in, instructions will find cells to execute in. Additionally, there is talk about eliminating the boundry between cells that are contained within a specific system, and cells that are contained on other systems connected VIA traditional networks.

Does this concept exist now?

Yes.

The current IBM "Blue Gene" Supercomputer is the fist implementation of a system based on IBM's "Cellular Computing" Research.

Currently, I think as of June, the IBM Blue Gene system in place has 8,000 processors and is ranked 4th fastest in the world of supercomputers. When complete this system will have 131,000+ processors total. (Possibly, the Blue Gene or it's sister Blue Gene/L project will end up with just over 1 million processors.)

Each processing element in the Blue Gene Supercomputer is on a single chip with two processing elements, some math helping elements, and 4mb of RAM.

The eventual distributed RAM, among all of the processors in IBM's Blue Gene will be 512gb.

IBM is also working on the "Super Dense Server".

The "Super Dense Server" will have 60,000-80,000 processors and a total memory,(YES RAM), capacity of about 40 terabytes.

With 40 terabytes of RAM, it needs "AT LEAST" 40 terabytes of disk just to load up a full database?

(NOTE: As these systems are considered to be "research projects" the number can change dramatically due to what they find as they go along.)

How much of all of this will be implemented in the PS3?

Just like always, we won't know until it comes out, but to fit in with the concept the new PS3 will integrate processing elements, RAM, and I/O into cells, and connect each cell through a form(s) of a communications bus. These cells will reside with multiple cells per chip or maybe all cells on a single chip. There will be a communications infra structure where cells can exchange packets of information that may contain instructions, data, control information, or combinations of these.

Possibly more importantly, if the PS3 is going to have large numbers of processors, how are their programming tools going to handle this? There are inumerable questions that can arive from this concerning communcations, data storage, instruction distribution, and on.....

I see this whole thing under three possibilities. Either Sony is using the Cell Concept as "hype", and they will just have a better game box.

The other two, are that Sony is really trying to implement the Cell Concept and will either revolutionize consumer computing in general, or flop.

Sony, and IBM and Toshiba, are being pretty tight lipped about this.

IBM has a lot of it's prestige invested in this as well. They are the ones that sold Sony on the concept. If the design comes through, they get a few feathers as well as future business.

Time will tell.....

 

[Mephistopheles]

Wow, this does sound like a great idea.... Scaling will work marvellously... But won't manufacturing costs be too high for a high-performance cell?...

Very promising idea, though! Very intriguing.

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>

 

[jim552]

Even if this is the model that the IBM/Sony/Toshiba "Cell Processor" is based on exactly what is included in it will not be known until it is released for use.

What is included will be what defines costs, for the most part.

It is VERY likely, that even if the PS3 configuration is based directly on this research, what comes out for "consumer use" will be much more scaled back/down.

After all it is VERY unlikely that, at least in the near future, anyone will need 130,000+ processors on their desktop. So alot of complexity and a lot of cost can be reduced just from simplicity such as this.

Hopefully they will not need to sacrafice anything that enables what is implemented to be easily scaled.

Again, I just think they the two are VERY much related.

 

[ChipDeath]

Have you seen any information on the actual processing power of 1 single cell though?

After all it is VERY unlikely that, at least in the near future, anyone will need 130,000+ processors on their desktop.

if each cell had the equivalent processing power of a ZX81, then most people <i>would</i> probably need 130,000 of them :lol: .

But seriously, this does sound interesting though. Wonder just <i>how</i> they'd use this sort of stuff in the PS3? Would they stick with the 'classic' architecture and simply have a few cells for the processor, but still get some specialist graphics processing tech in there for the visual stuff, or will they simply just stuff it full of cells & use some of them as graphics hardware? The latter option is very interesting, because you could then have some games which are graphically simple but require a lot of processing power (Like simulating the behaviour of every inhabitant in a Simcity type game or something), and some games which are all about the graphics (but use less CPU-time), and both types could run equally well on the same hardware.

Of course this kind of gets back to the idea of using the graphics hardware for non-graphics tasks I guess... We'll have to wait and see what they come up with I suppose.

---
Epox 8RDA+ V1.1 w/ Custom NB HS
XP1700+ @200x10 (~2Ghz), 1.4 Vcore
2x256Mb Corsair PC3200LL 2-2-2-4
Sapphire 9800Pro @412/740