Is the Cell a FrankenChip?

[Urian]

A FrankenChip is an scalar microprocessor made from an one or various existant microprocessor, the idea is that you can get an specialized microprocessor removing and adding elements from an existant microprocessor.

It can sound weird, but I believe that the Suzuoki patent of 2001 is only a concept of that STI want for the Cell and the microprocessor started to be designed really when the PowerPC 970 started to be designed from the Power4.

 

[Megadrive1988]

I wish Cell had more than one PowerPC core ~ PPE

 

[Edge]

I wish Cell had more than one PowerPC core ~ PPE

That's like saying you wish the orchestra had more than one conductor.

 

[ERP]

That's like saying you wish the orchestra had more than one conductor.

By that analogy more like a bigger string section.

 

[McFly]

I wish the PS3 would have the PS4 CPU and GPU ... or make it PS5 CPU and GPU and I'm sure everyone will be happy ... for at least a few minutes.

Fredi

 

[standing ovation]

A FrankenChip is an scalar microprocessor made from an one or various existant microprocessor, the idea is that you can get an specialized microprocessor removing and adding elements from an existant microprocessor.

I don't know about you, but I think STI put too much into the Cell coffer to simply reinvent the wheel. Given all the dough they are throwing around I am almost certain they are trying to finance a revolution.

 

[Fafalada]

By that analogy more like a bigger string section.

Well it would be enough if string section had an ace soloist.

 

[randycat99]

OKAY, ENUFF!!! How about the cowbell- what part will be analogous to that?

 

[TurnDragoZeroV2G]

Guess I can now look forward to how people are going to interpret those comments.

Puerto Rican ghettos present and accounted for.

Edit: apparently I was waiting in the wrong thread. *cough*

 

[Urian]

I don't know about you, but I think STI put too much into the Cell coffer to simply reinvent the wheel. Given all the dough they are throwing around I am almost certain they are trying to finance a revolution.

I don´t believe that the Cell is a reinvention of the wheel. Only an architecture for having a multimedia based supercomputer on one chip but it won´t rewrite the rules of the microprocessor design. It only shows the versatility that the engineers can have designind a microprocessor.

My idea of the Frankenchip is that I ever watch the news about Cell since 2003 and the white papers of it and for me Cell was originally a concept of a microprocessor and when STI wanted to make the idea a real thing they used existant elements of another microprocessors for making the Cell concept a real microprocessor.

 

[Carl B]

I don't know about you, but I think STI put too much into the Cell coffer to simply reinvent the wheel. Given all the dough they are throwing around I am almost certain they are trying to finance a revolution.

The big money went into the fabs, fabs that would have been built regardless of *what* architecture Sony went with. So you may consider the Cell project a fool's undertaking or you may not, but at the end of the day the vast majority of the expenditure has utility beyond Cell.

 

[Edge]

CELL to me is a General Purpose CPU with a bank of ultra powerful DSP's, with lots of localized memory, massive internal bandwidth, and with massive external bandwidth.

I will describe each one in turn:

1) ultra-powerful DSP's - extremly high clock rate for DSP's, as no DSP by any other company runs anywhere near 3.2 GHz. So these specialized processors were designed to run at high clock rate. This was a key architecture decision in designing CELL.

2) localized memory - 256 KB of SRAM, must faster than eDRAM, but also taking up four times the number of transistors, but the cost is worth it, as you have the speed of cache, but in a signaficant amount to run a fair size algorithm or algorithms. Each processor can operate at full speed within it localized memory, while having ZERO impact on the rest of the system. This eliminates huge amounts of wait states, where multiple processors is contending for the same cache. Of course there is still data access contention in the system as all these processors have to be coordinated in getting fed. This makes CELL more difficult to program than a traditional processor.

3) massive internal bandwidth - very important, especially for algorithms that need to stream data from one processor to the next, and with such a large number of processors, seven of them, you have to keep them fed, even if they are not doing streaming algorithmic work.

4) massive external bandwidth - 25 GB/sec to main memory, and 35 GB/sec to a GPU. Data can come in from main memory, and be worked on by a bank of processors, and then instead of being written out to main memory, can be passed directly to the GPU. This kind of through put with so many processors, and especially for any data that can be worked on in streams, would provide huge benefits. Could easily be 10 to 30 times more powerful than a single traditional processor in this regard.

I know there are pluses and benifits to this design, but in the end, many processors with localized memory can either be operating on a single task and having ZERO impact on the processing ability (except data contention) of another processor in the system, or each processor is employed to work on a small piece of a single problem, before passing it on to the next processor. There is flexibility in the design, with the localized memory being big enough to support individual jobs, and with the massive internal and external bandwidth supporting these units being fed.

While I describe the SPE's as DSP's they are much more than DSP's, and more akin to something between a traditional DSP and a general purpose CPU. I only use the term DSP because DSP are designed to be fast at math operations, just like the SPE's in CELL.

CELL is very forward looking, especially one could argue that CELL in it's present incarnation exists to teach programmers about data contention, and how to break up algorithms between many processors. I think PS4 will contain a single tranditional processor again, but even more SPE's, probably 32 to 40 of them. Unless a breakthrough comes in preventing current leakage, I don't see the clock rate going up by much, with 3.6 GHz probably being tops.

 

[Fox5]

Considering those SPEs are more general purpose than modern day GPUs, it may be a disservice to compare them to DSPs. How about math coprocessors?

 

[Edge]

Considering those SPEs are more general purpose than modern day GPUs, it may be a disservice to compare them to DSPs. How about math coprocessors?

I think the argument for defining the SPE's would be equivalent to defining if Pluto is a planet or not. In the end there is no definite description of these things. I think the SPE's could easily be defined as either one of these: CPU, DSP or math coprocessor. Take your pick.

 

[centerofadmiration]

Why not call them what IBM and Sony always call them? That being Vector Processors, or better yet simple Vector Computers the size of a microchip? Each one has its own SRAM, CPU and logic so its not far fetched to call them that.

 

[standing ovation]

I don´t believe that the Cell is a reinvention of the wheel. Only an architecture for having a multimedia based supercomputer on one chip but it won´t rewrite the rules of the microprocessor design.

In light of the book, FrankenChip means to reanimate something that is dead ...

As far as fitting into society goes, reuse is what made the "monster" in Frankenstein seem so out of place. Likewise, a bunch of computer chips revived from a deceased generation of processors would appear crude to us as well.

Don't get me wrong. Ingenuity brought the corpse in Frankenstein to life; and this would probably be the case with a FrankenChip. However, its impact (as in the book) is likely to be perceived as shockingly regressive.

But Cell's DNA sounds strikingly progressive -- not something you would come up with using preexisting technologies alone. If successful I think the project will not only rewrite the rules of microprocessor design but programming as well.

 

[standing ovation]

The big money went into the fabs, fabs that would have been built regardless of *what* architecture Sony went with. So you may consider the Cell project a fool's undertaking or you may not, but at the end of the day the vast majority of the expenditure has utility beyond Cell.

Without Cell, I doubt Sony would have built any fabrication facilities at all. A broader knowledge base and much higher economies of scale give existing chipmakers an obvious advantage, making it more cost effective to simply outsource designs.

With Cell, I think Sony, under the leadership of Ken Katuragi, would have bet the farm on a radical notion: a one-size-fits-all microprocessor. Since his demotion, I doubt the company has been putting as many eggs into one basket. :neutral:

 

[Carl B]

Without Cell, I doubt Sony would have built any fabrication facilities at all. A broader knowledge base and much higher economies of scale give existing chipmakers an obvious advantage, making it more cost effective to simply outsource designs.

With Cell, I think Sony, under the leadership of Ken Katuragi, would have bet the farm on a radical notion: a one-size-fits-all microprocessor. Since his demotion, I doubt the company has been putting as many eggs into one basket. :neutral:

It's true that without Cell the build-out would have been less likely, but that's mainly due to volume assurances. If nothing else, it can be said that Cell will reach high volumes of production if for PS3 alone, and truthfully I'm sure whatever other chip would have been chosen/designed for PS3 would have led to a similar build-out.

And with the deed being done, they have increased fab capacity for any future projects they may undertake. Also bear in mind now that Kutaragi is no longer head of semiconductors it hasn't actually led to a slackening off of Cell investment; on the contrary Sony is looking to invest a further $3 billion on it's fab capacity.

On the side, here's the PDF from the semiconductor group's recent roadmap: http://www.sony.net/SonyInfo/IR/info/Semiconductor/2005.pdf

 

[standing ovation]

It's true that without Cell the build-out would have been less likely, but that's mainly due to volume assurances. If nothing else, it can be said that Cell will reach high volumes of production if for PS3 alone, and truthfully I'm sure whatever other chip would have been chosen/designed for PS3 would have led to a similar build-out.

I do not think "Cell" is the nomenclature for a particular processor as much as it is an emerging standard in microprocessor design.

If its marketing is to be believed, then Sony's new guidepost could fix a lot of the problems associated with the old (x86) standard and ultimately kick Intel to the curb. Seizing control of the world's chip production -- from simple electronics to supercomputers -- is something Mr. Katuragi could not pass up. Hence the rationale behind Sony's spending spree.

Apart from Cell's applicability, it does not make cents for Sony to invest heavily in manufacturing facilities just to become another competitor, particularly when its electronics division has been less profitable than its entertainment divisions -- music, movies and videogames.

(Apparently, Sony Electronics has serious personnel issues.)

 

[Carl B]

Apart from Cell's applicability, it does not make cents for Sony to invest so heavily in manufacturing facilities to become just another competitor, particularly when its electronics division has historically been less profitable than its entertainment divisions -- music, movies and videogames.

You see, but that's the point exactly: to insource what has traditionally been a very outsource heavy aspect of Sony's CE operations, thereby increasing profitability (assuming high utilization rates at the fabs). As for x86 - yes it fixes problems - but primarily targeted at media and imaging it seems, and other FP-intensive ops. Never billed as a replacement for that architecture by any means.

I think the fab build-out was a good move on Sony's part; that's a personal view.

Cell, whether successful in shifting the paradigm or not, provides the 'insurance' to make such a build-out relatively secure in terms of risk (via the PS3), and in that light I think regardless of eventual production needs, having the extra fab capacty will in most instances be a net positive.

I agree that Kutaragi was (and is) probably possessed of a 'grand vision' for Cell and it's future, but at the same time I don't think him so reckless that he didn't put the Semiconductor division on path for an acceptible 'Plan B' scenario should Cell not find itself in everything. Afterall, the EmotionEngine has shown us that 'failure' for the architecture in the market does not mean failure for the initiative from a manufacturing standpoint as long as internal demand alone can compensate for the investment.

As for the nomenclature, well I think 'Cell' is the name for the singular present example of the Broadband Engine/Cell architecture (kind of like a self-titled album), but there have been discussions on this very point before that really didn't resolve.