Zig&Zag: Now Sony management change, layoffs

[Vince]

You're forgetting Cell's scalability. Sony would be daft to put a 1:8 BE into a TV, but they can use a 1:1 Cell chip, and print them off at 3x the density of the BE at 1/3rd the price.

Actually, it would be prudent (not daft) for Sony to put a 1:8 (eg. 1:8 implying the 250mm2 IC) Cell IC in their CE products. As a semiconductor manufacturer, they pay per SOI wafer for their ICs, not per yieldable IC as you can do with some foundries (like IBM did).

With this in mind, they need to produce N quantity of ASICs for PlayStation3 -- they must output N 7:1 capable Cell's per wafer. Cell is a larger chip, over 50% of it's surface is composed of redundant structures. Crank through the probability and it's favorable (eg. off the top of my head... ) that there will be some sort of characteristic size distribution applicable to errors per die, perhaps a power-law distribution.

As a corollary to the above, it's obvious that the more products they can use Cell in, the more diluted the fixed costs become. As stated, if it's some sort of power-law distribution, they will have many more yieldable ICs (eg. those without fatal errors) that have 1 working SPE than 2, and much, much more that have 2 working than 7.

So, the costs are already on Sony's balance sheet due to PlayStation3. It's in their best interest to stick a 1:8 Cell into everything they can. But, while it is a super-linear increase in yield as you decrease die size, you then have to incurr costs to redesign the entire back-end of the IC and you have recurrent costs in both having divergent manufacturing lines as well as the opportunity costs that's lost. Just sell the ones with 8 working SPEs for servers at immense mark-up, the 7 working ones go to PS3. The 6 working ones are CMPed into Home Servers or HDTVs, etc.

It has occurred to me that my Linksys router is nothing but a MIPS computer and as such is basically future-proof when to comes to computing the closed-set of functions possible under the 802.11g standards and the hardwired functions.

Using the same paradigm, I'd love to see Sony put Cell into their ES series receivers; a Cell with 2 or 4 SPEs can easily decode any audio standard that's physically feasible. Put an Gigabit Ethernet port on it and do the same with their Blu-Ray recorders, HDTVs, DVRs, etc.

 

[avaya]

OK say they put a PS3 redundant CELL; 4-6SPE’s, in a Bravia W-Series LCD TV. We know they could leverage its power to improve picture quality and audio fidelity while decode multiple streams for preview. However could they not also use the extensive processing horsepower that would be left over to implement some sort of innovations in the TV space?

Maybe some form of voice command/recognition ability? Although I assume the software elements involved would likely complicate that idea.

Another idea could be the ability to browse the internet via the TV, after all at HD resolutions this is now a feasible idea.

This is all conjecture on my part.

 

[Shifty Geezer]

Leveraging the power is something that's been on the cards from the off. The Cell partners have spoken of networking Cell based equipment and sharing resources. It's hard to imagine this in action given the available power and the rather limited scope for applying it, but an imaginary use might be...

On a TV a program is transmitted with a digital content page that incorporates a realtime 3D model. Say an archaeology program with a 3D skeleton. On your TV it updates at 15 fps when you view it. Connect up your PS3 and it's rendered at 60 fps (or 30, or 59.94, or interleaved 30 half-frame, 60 fields per second, depending on what magic these screens use). One of the designs of Cell software was on the principle of Apulets that are sent to any available SPE's to process on the network. Add more SPE's and you can use them. That's the reason for one of those gigabit network ports. I imagine a Cell TV will come with a Gigabit port as that'll be the communications highway for Cell processors in different devices.

But as I say, it's hard to imagine what processing a TV or HiFi will be doing that'll require more processing power than a 1:6 Cell where you'll actually see a benefit with adding another Cell device on the network.

 

[ihamoitc2005]

CELL dev cost

CELL development cost = $400 Million

 

[ecliptic]

The entire Cell thing is a gimmick. There is absolutely nothing it can do that some other processor cannot do.

 

[AlphaWolf]

The PlayStation business hasn't performed well for a while now. Weighing it with so much responsibility will be risky over the next several years.

Well this times article disagrees.

The falling prices have created an unusual situation where former bread-and-butter products like televisions and computer chips have become money losers. The biggest earners are now the popular PlayStation 2 video console and nonelectronics operations: films, music, insurance companies and a bank that Sony runs in Japan.

 

[gokickrocks]

what happens when IBM develops an enhanced version of CELL without Sony

 

[AlphaWolf]

CELL development cost = $400 Million

For Sony alone? or split 3 ways with IBM and Toshiba.

 

[Bobbler]

The entire Cell thing is a gimmick. There is absolutely nothing it can do that some other processor cannot do.

That is a rather silly thing to say...

There is nothing an Athlon64x2 can do that a Pentium 100 can't, but it doesn't mean I'd consider getting a Pentium 100 when I could get an A64 -- There are some things the Cell is exceptionally good at (streaming type applications), that's what it was made for... I'd bet that a Cell would cream an A64/P4 in HD decoding situations, yet an A64/P4 would probably destroy it in other situations.

 

[Bobbler]

CELL development cost = $400 Million

I have heard much higher numbers than that for the entire cell development cost (a couple billion) -- where did you get that number?

 

[Bobbler]

what happens when IBM develops an enhanced version of CELL without Sony

Well Cell is by STI (Sony, Toshiba, IBM), so I imagine IBM can't just create a suped up version without talking to STI first. Not sure why IBM would want to just go behind the backs of Toshiba and Sony for something that IBM probably wouldn't be in a position to use very well.

 

[randycat99]

Just sell the ones with 8 working SPEs for servers at immense mark-up, the 7 working ones go to PS3. The 6 working ones are CMPed into Home Servers or HDTVs, etc.

I tell ya- that whole tiered usage scheme just brings a friggen tear to my eye! Just pulling a number out of thin air, but the idea of a chip getting an effective yield of over 90% right out of the gate, across the entire range of targeted end-use products, is just breathtaking. It's great for the environment, great for the bottomline. It takes the idea of scaleability to a whole new level (not just scaling a design to meet a wide range of targets, but exploiting yield effects to make one design meet a wide range of targets).

As exciting a future PS3 looks to have, I have to admit it is equally exciting (for me) to anticipate how well this tiered yield strategy will take Cell...

 

[Shifty Geezer]

The entire Cell thing is a gimmick. There is absolutely nothing it can do that some other processor cannot do.

Mmmm, I love the smell of flamebait in the morning!

No there's nothing Cell can do that other processors can't. Any processor can perform an 256 point FFT. Just that it's strengths in certain areas mean it can do things faster and cheaper. Any processor can perform an FFT, whereas how many other than Cell can perform an FFT with 42 GFlops useable power in an affordable single-chip solution with inherant scalability? And you're point that 'some other processor cannot do' flies in the face of everything thus far said inthis thread. Yeah you can grab a custom chip to perform video work, a different processor to drive a mobile phone, another processor for you TV and a fourth for your console, but you'll need different chips for different jobs with the high overheads of different development platforms. And that's the way the industry currently is. Sony's buying in different components for different uses. Cell overcomes this with a single chip solution that they can potentially attain incredible cost reductions if it's be used in such vast numbers.

Regards Cell RnD, there has been figures of Billions to date, but AFAIK that's for fabrication plants. I've heard elsewhere RnD of the chip alone was $400 million but don't know where an official figure comes from.

 

[avaya]

I have heard much higher numbers than that for the entire cell development cost (a couple billion) -- where did you get that number?

The $400million is the initial R&D costs, including the research centre in Austin TX

The billions are for the fabs at East Fishkill and Nagasaki.

 

[PC-Engine]

Mmmm, I love the smell of flamebait in the morning!

No there's nothing Cell can do that other processors can't. Any processor can perform an 256 point FFT. Just that it's strengths in certain areas mean it can do things faster and cheaper. Any processor can perform an FFT, whereas how many other than Cell can perform an FFT with 42 GFlops useable power in an affordable single-chip solution with inherant scalability? And you're point that 'some other processor cannot do' flies in the face of everything thus far said inthis thread. Yeah you can grab a custom chip to perform video work, a different processor to drive a mobile phone, another processor for you TV and a fourth for your console, but you'll need different chips for different jobs with the high overheads of different development platforms. And that's the way the industry currently is. Sony's buying in different components for different uses. Cell overcomes this with a single chip solution that they can potentially attain incredible cost reductions if it's be used in such vast numbers.

Regards Cell RnD, there has been figures of Billions to date, but AFAIK that's for fabrication plants. I've heard elsewhere RnD of the chip alone was $400 million but don't know where an official figure comes from.

A fridge doesn't need 46 GFLOPS of computing power. It doesn't even need a 1PPE 1SPE CELL...

It only needs a cheap $3 8-bit IC you can get at Radio Shack and even those are the smart fridges. CELL is a nice concept, but concepts don't always work out as expected.

 

[Shifty Geezer]

Please link me to the article saying a 47 Gflop Cell is being incorporated into a fridge.

 

[PC-Engine]

Please link me to the article saying a 47 Gflop Cell is being incorporated into a fridge.

Please link me to an article where it says a mass market device needs a CELL chip instead of a cheap dedicated IC other than a game console.

 

[Shifty Geezer]

I haven't a link to hand, and admit maybe I just imagined all those announcements from Sony and Toshiba about wanting to include Cell in CE good like TVs. Perhaps Cell will only ever be used in PS3 then, and Toshiba just wasted loads of resources on it and are never going ot use the technology themselves?

I don't see why anyone would look at a programmable media processor like Cell and want to use it in a fridge though and don't know anyone but yourself who even suggested such a ridiculous notion.

 

[epicstruggle]

The PlayStation division had been losing the company money in recent quarters.

anyone have info on the breakdown of divisions and their revenue/loss over at sony? I find it hard to believe that their playstation division is losing money recently.

thanks in advance.

epic

 

[Shifty Geezer]

Just Google up Sony's financial reports. You'll get quarterly reviews if not an annual summary.