Toshiba to produce Cell in 2004

[Vince]

12/16 20:51
Toshiba Will Begin Making `Cell' Chip in 2004, Executive Says
By Yoshifumi Takemoto

Tokyo, Dec. 17 (Bloomberg) -- Toshiba Corp. will begin commercial production of a radically new processor designed with Sony Corp. and International Business Machines Corp. as early as 2004, an executive at Japan's largest chipmaker said.

The chip, called ``Cell,'' will be made at a new chip plant Toshiba will build in Oita prefecture, in southern Japan, Toshiba Vice President Yasuo Morimoto told reporters attending a yearend gathering on Monday. Toshiba said Friday it will invest 350 billion yen ($2.9 billion) on the Oita plant and another for flash- memory chips it will build in Mie prefecture.

Dubbed a supercomputer on a chip, ``Cell'' is a departure in design because it combines so many functions on a single piece of silicon. The chip may be used in Sony's next video-game console, as well as a host of other consumer electronics, such as a new generation of slimmer, more powerful cellular phones. Elements of the chip's design may also be used in IBM's server chips.

``Cell'' is an outgrowth of an agreement reached in April by Toshiba, Sony, the world's largest maker of video-game consoles, and IBM to collaborate in a four-year project to find ways to produce smaller, less power-hungry and faster chips on 300- millimeter (12-inch) silicon wafers.

Larger wafer sizes may reduce production costs by as much as 30 percent compared with standard 200-millimeter wafers, analysts and investors estimate.

Toshiba, in collaboration with Sony and IBM, wants to develop microprocessors capable of handling moving pictures broadcast over the Internet.

The ``Cell'' chip may give the partners an edge over rivals such as Intel Corp. because it adopts an entirely new chip design developed by engineers at the three companies to specifically handle complex graphics, Morimoto said.

``We have an edge because Intel may find it difficult to abandon the architecture of their current microprocessors,'' Morimoto said, adding Intel's processor designs favor computer- centered applications, rather than graphics.

Morimoto expects the two new plants Toshiba announced Friday, both of which will use 300-millimeter production technology, to each cost between 100 billion yen and 200 billion yen.

The Tokyo-based company, which had 311 billion yen in cash as of Sept. 30, will fund the projects using money earned from operations. Still, Toshiba would ``welcome investment from partners,'' Morimoto said.

To share costs, Toshiba will ask Sony to invest in the new Oita plant, the Nihon Keizai newspaper reported earlier.

http://quote.bloomberg.com/fgcgi.cgi?T=marketsquote99_relnews.ht&s=APf6DDBRkVG9zaGli

Seems there is alot of Fab space being devoted to Cell based solutions. Oita and Fishkill are both suppose to be quite large.

 

[V3]

Toshiba, in collaboration with Sony and IBM, wants to develop microprocessors capable of handling moving pictures broadcast over the Internet.

Is that a hint ?

 

[TTP]

No. That's just "I dunno what I'm talking about but I guess putting the words 'moving pictures', 'broadand' and 'internet' close together makes my article more cool"

 

[PiNkY]

The chip may be used in Sony's next video-game console, as well as a host of other consumer electronics, such as a new generation of slimmer, more powerful cellular phones.

That just doesn't make sense to me, mobile phones need cool running, low power chips, not XXXWatt space heaters as found in PCs and consumer devices.

 

[Vince]

That just doesn't make sense to me, mobile phones need cool running, low power chips, not XXXWatt space heaters as found in PCs and consumer devices.

The idea around cellular computing is that at the most basic level, there is a common 'core' thats inherient in the design. It's this fundimental functional unit (probobly composed of processing and on-board storage, communincations) that is used concurrently to produce the higher-powered deviced - ie. Cell in PS3.

Thus, each individual core is built to be very effecient in tranistsor vs. preformance and heat, size ratios. Ideally, this new form of architecture would be scalable not only in hardware, but software.

Cool idea in theory anyways.

 

[PiNkY]

I did not want to tease you Vince, my comment was in regards to the "The chip" statement in the quote. No questioning the scalabilty of such an approach. Yet in a way, thats basically what MIPS bussiness model has been about for years now (supply a simple expandable core ISA).

 

[KnightBreed]

The ``Cell'' chip may give the partners an edge over rivals such as Intel Corp. because it adopts an entirely new chip design developed by engineers at the three companies to specifically handle complex graphics, Morimoto said.

``We have an edge because Intel may find it difficult to abandon the architecture of their current microprocessors,'' Morimoto said, adding Intel's processor designs favor computer- centered applications, rather than graphics.

Intel's chips aren't made to "specifically handle complex graphics." Which is precisely why Microsoft went with nVidia for the graphics chip.

 

[MfA]

Would be so nice if they quoted the important parts of such interviews.

 

[fbg1]

The idea around cellular computing is that at the most basic level, there is a common 'core' thats inherient in the design. It's this fundimental functional unit (probobly composed of processing and on-board storage, communincations) that is used concurrently to produce the higher-powered deviced - ie. Cell in PS3.

So, for example, a cell phone would a use single-core Cell chip while a PS3 might use a 32-core Cell chip?

 

[speng]

It's not only about how many chips, it's about how the chip performs doing the tasks of video image processing and dealing with broadband content.

It should perform with lower power consumption, way less heat and with equivalent power to handle many processes that todays regular pc's do.

So, your palm-phone will probably be able to handle on demand video with some Wi-Fi access, or take pictures and edit/crop and send them. Or a short movie of your trip sent to your family live.

With the PS3, I think the core of everything will be based on games, with plug-ins to enable Ti-Vo functionality and any other set-top box requirements.

Speng.

 

[V3]

So, for example, a cell phone would a use single-core Cell chip while a PS3 might use a 32-core Cell chip?

I think something like that. Intel is going that route as well it seems. Their plan for their 1 billion transistor chip is to have something like 4 cores and very large cache.

 

[Vince]

I did not want to tease you Vince

Heh.. you didn't tease me

Yet in a way, thats basically what MIPS bussiness model has been about for years now (supply a simple expandable core ISA).

Usually don't like to make my predictions public (as they can come back to bite you), but watch and see whose core (albeit stripped to shit and built up) is at the base of each cellular core - MIPS 64.

 

[Panajev2001a]

Vince, it might be MIPS64 stripped down or an ARM stripped down... still it is a 57 instructions ISA, talk about it beign stripped down ... with no FPU ( shared ) and no L2 ( shared amongst the Thread Units )...


but as far as Blue Gene/L Cellular Architecture goes we know something about the Thread Units, which are these "cellular cores" you are talking about... of course the CELLs as seen in Blue Gene might differ from what Sony, IBM and Topshiba have in mind for the future of CELL, for its next step... although seeing that many of the slides about CELL that Kutaragi has used were straight out IBM papers mentioning Blue Gene as well..

I know you are very familiar with the CELLs in Blue Gene... 8 Thread Units with L1, 1 FPU, 1 shared L2, 512 KB of e-DRAM, an Arbiter, etc...


You can find this quote at IBM's Blue Gene site ( "Blue Gene in the news", link to the Wired 9.07 story... )...

Coupled with this DRAM is the new type of processor that is Denneau's particular pride and joy. Each processor "core" will contain a number of separate subprocessors, each very simple and not startlingly fast. These will share higher-level resources such as the local embedded DRAM and the complex circuitry dedicated to making floating point calculations. Everything is stripped of unnecessary complexity: Denneau is taking what's called reduced instruction set computer (RISC) ideas a lot further than anyone else at IBM, constricting the number of different things his chips can do to a bare minimum. "We looked at program traces in IBM chips - at which instructions are actually executed in applications," Denneau explains, "and we found an amazing thing. There are hundreds of instructions, but all we ever saw used were the same 50 or 60 or 70 instructions over and over again. We found instructions in the processor that the compiler didn't know how to make use of."

Much of the problem, as he saw it, was that commercial chips run only one "thread" of calculations at a time. To move the thread along as quickly as possible, the chips spend a lot of effort anticipating what might be needed at some time in the future, constantly second-guessing the thread's destination to make its execution quicker. Rather than trying to rush through a single thread quickly, Denneau's processor - stripped down to 57 instructions - runs four threads, which can all take their time.

 

[PiNkY]

The big problem though is not to build such a chip, but to make the targeted applications multi-threaded in an efficient way. Games especially have hitherto been mostly single-threaded, and the rare examples that are multi threaded currently do not show really vast performance gains (the Quakes, (2 threads, i think) being the most prominent examples, offer about 20% more performance with SMP when cpu limited and hardly count as the worst coded games on pcs). Of course you gain additional performance by executing the different os services in parallel, but esp. consoles have not too much of those and i guess hardly spent much cpu time on them. They also sport dedicated hardware for most of the more complex ones (IOP, both DMACs in PS2 for I/O, Sound and Network DSP in XBOX "southbridge", etc.). I am really looking forward to what they will come up to solve these problems, as general solutions/paradigms will imo represent a much greater (sw) engineering feat then just putting lots of simple cores on a single die (and i am sure that they have some promising ideas, otherwise they would not spend the money to build such a part).

 

[PiNkY]

Though if one distances oneself a bit from the blue gene setup, and think of a cell system mainly composed of sp-fp capable cores (similiar to PS2 VUs) with int cores in between (for load/stores & adress ops.), i`d imagine that a cell chip would make quite a flexible fragment/vertice unit. Couple that with a decent CPU and you'd have something that makes sense to me.

 

[Panajev2001a]

pinky, that is not a bad idea, but it would break the big picture... the implementation of the FP units and specific parts of the CELL for PS3 might differ from other CELL based producs, but they have tp share similarities... like it happen in the x86 world... the core ISA must be constant in all the CELL chips... as much as possible... think about what they said about broadband OS for CELL, about having two CELL based machines work as a single machine...