IBM to manufacture Cell on 45nm SOI, Sony/Toshiba to move to 45nm bulk process

[Arwin]

I think we're talking here very specifically of the use of Cell in small devices. The PPE unit is too much of a power drain for that. Instead, we'll see some more applications for the SPE based units with lighter more specialised or even maybe without managing cores, much like the Magic Mirror stuff that Toshiba developed using four SPEs and some other chips (if I remember correctly).

We're obviously not talking about Cell in the context of gaming or heavy-weight super computers and such that the science, military and medical departments seem to be quite enthusiastic about.

 

[Shifty Geezer]

I think we're talking here very specifically of the use of Cell in small devices.

Nothing in Nakagawa's statements suggests he's talking only about small devices. Power draw isn't an issue for TVs and players. Cut-down, underclocked Cell's would be plenty cool enough to be fitted in CE devices even if not handheld devices. Without an exact quote though, it's hard to infer his true meaning without editorial bias. Still, the fact they're not mass-producing Cell for use in Sony's CE goods goes to show a major reason for Cell has flopped off the radar.

 

[one]

Still, the fact they're not mass-producing Cell for use in Sony's CE goods goes to show a major reason for Cell has flopped off the radar.

Before coming to a conclusion we have to know how much investment is required to build a plant with state-of-the-art 45nm/32nm lines, and how many chips they have to manufacture in it to make it a more worthy investment than using other foundries. They can use Cell in their CE products even when using other foundries to manufacture them if it does make sense in terms of cost and product requirement. A larger volume makes chips cheaper in that case, too.

 

[Shifty Geezer]

Process cost isn't the issue. They wide-scale adoption of Cell justified the funding 65 nm, production facilities, but didn't (hasn't) happened. Now they're saying Cell wasn't really suited to those tasks and there isn't the mass market need for Cell, making their capacity redundant and giving cause to sell off a lot of it. The reason for the reduction on production is capacity is because they found they didn't have any use for it, with Sony's CE goods not using the products of Sony's chip foundries - the Cell. They haven't used different variations of Cell with reduced SPE's for cost. They're not using down-clocked cool variations. They're not using the defects from normal Cell production. They're not using Cell at all in TVs, movie players, any upcoming Sony products, not because it's pricey (they can use leftovers) or because they're power hungry (doesn't effect plug-in devices and lower clocks solve cooling issues) but because we're being told Cell's not suitable for those jobs. It seems the CE division didn't share Kutaragi's vision and don't think much of Cell for their purposes.

 

[patsu]

Toshiba did make a variant of Cell though for CE and laptop though. The SPUREngine contains only 4 SPUs (no PPU), probably to meet the necessary cost and price/performance ratio. IBM has a double precision Cell for scientific computing. The original Cell is still being used for PS3.

Nakagawa is an exec who controls his money tightly (flies via economy class), and only act on good numbers. So I wouldn't be surprised if he cuts back Cell investments even if it made technical, marketing sense but are too risky financially.

The Cell architecture will have to prove itself in the gaming market. Should it do well, there will still be chance for reuse and extension (or to convince Nakagawa with real numbers years later). It is just one of the many tests Cell will need to go through to establish itself... or not. And it will take time.

 

[Shifty Geezer]

Toshiba did make a variant of Cell though for CE and laptop though. The SPUREngine contains only 4 SPUs (no PPU), probably to meet the necessary cost and price/performance ratio. IBM has a double precision Cell for scientific computing. The original Cell is still being used for PS3.

Sure, and I wasn't suggesting Cell was a dead end. Only that the promise of Cell powered CE devices seems like a dead end at the moment. Toshiba have been talking about it for years, but producing nothing until SPURSengine, still not finding real applications. Sony were going to use Cell, part of the whole reason for Cell being its flexibility, and then gave up even making promises, only to finish with this statement that they don't rate Cell as worth bothering with. Cell in CE devices seems a no go. Why is that? Cost reasons doesn't seem sensible when you already have Cell throw-outs to play with, that you can put into high-end gear at a premium price. There must be plenty of 4 SPE Cells from existing Cell production to satisfy some high-end Bravia use, for example.

 

[patsu]

Sure, and I wasn't suggesting Cell was a dead end. Only that the promise of Cell powered CE devices seems like a dead end at the moment. Toshiba have been talking about it for years, but producing nothing until SPURSengine, still not finding real applications.

According to Toshiba, SPUREngine will be incorporated into their high end laptop line, together with the "Magic Mirror" application.

Assuming they keep to their announcements, it will also be used in their high end TVs.

Sony were going to use Cell, part of the whole reason for Cell being its flexibility, and then gave up even making promises, only to finish with this statement that they don't rate Cell as worth bothering with. Cell in CE devices seems a no go. Why is that? Cost reasons doesn't seem sensible when you already have Cell throw-outs to play with, that you can put into high-end gear at a premium price. There must be plenty of 4 SPE Cells from existing Cell production to satisfy some high-end Bravia use, for example.

Yes. Actually, they have said the same thing a year ago. As for the reasons behind not using it for high end CE devices... may be HD TV price and all CE prices are falling like a rock. So everyone is focusing on cost reduction (High End TV sells too little to be worthwhile ? See HD-DVD vs Blu-ray players for eroding price, ...). Perhaps only game consoles can sell at a decent enough number to justify the investments, like one alluded to). So they need to anchor that part first.

OTOH, the SPUREngine should be considerably cheaper to mass produce due to its simplicity.

 

[nAo]

CELL has always been a "glimpse into the future" kind of technology; when you sample things to come much earlier than anyone you know in advance that your architecture, no matter how good or bad is, is going to be a transitional one.
But maybe I'm slightly OT, I'd better open another thread on this..

 

[Shifty Geezer]

If it's a matter of cost and not functionality, why isn't there cause to use left-overs from existing Cell production? Or do we think Cell production has had such high yields that even after two years of 1:8 CBE production there aren't hundreds of thousands if not millions of low-end Cells available that'd sit nicely in 2000+ Euro TVs? And could have been incorporated over those past two years too? From Toshiba's POV, Cell in it's current build wasn't worth bothering with either, though we know they didn't care for PPE's inclusions. They've been the most vocal of using Cell in CE devices, but up until creating their offspring chip haven't included it, and aren't set to do a huge amount with it ewither from the looks of things. A SPURSEngine in an expensive laptop isn't going to appeal to much beyond the scientific and custom-apps industries. That's hardly mainstream CE devices!

Okay, let's simplify the whole discussion to its root. Why hasn't Cell in simple forms (1:4 and downclocked, say) been introduced into CE products over the past couple of years, and how does that tally with Nakagawa saying Cell isn't suited for CE products, such that he's pulled out of producing Cell for their own CE devices?

 

[patsu]

If it's a matter of cost and not functionality, why isn't there cause to use left-overs from existing Cell production?

I asked the same question before. The answer was Sony will need complementary components for the PS3 Cell (FlexIO ?). So the final cost will be even higher.

There is also Cell OS and programming need to consider. We should probably look at the entire package and see where the holes are.



As for Nakagawa's decision, I think it is very tied to his personality and business approach. The traditional hardware mentality is very much cost and scale based. So my guess is, him saying "Cell isn't suited for CE products" probably stems from those areas.

 

[fireshot]

Anyone remember this?

I think if Sony had kept plugging at it, cost and suitability do not pose a problem.

It is all politics imho. Kutaragi character is not well loved by Sony bigwigs. At the first signs of trouble, they hired a gaijin shield to dump the man and looks likely to dump his ideals. Which is a waste as Cell could do make Sony the leader in the ultimate unified experience at home. They cannot decide to jump in and out in this type of industry. The time lost is gone.

 

[patsu]

'Which is a waste as Cell could do make Sony the leader in the ultimate unified experience at home"

Cell does not necessarily contribute to a unified experience. Software does. e.g., XMB for cell phones and HD TV is already there. PS Home for cell phone has been demoed. Ironically, the power of Cell is also highly dependent on software (Note all the recent complains about delays and cross platform development).

The beauty of Cell is its peak power at consumer price. The economy of scale in reusing Cell helps to drive down the cost further. Without which, it means that Sony will need a focused strategy to sell more PS3.

EDIT: Note that the SPU algos/programs will still be "reusable" across the board, from Road-Runner to PS3 to SPUREngine.

 

[Shifty Geezer]

The beauty of Cell is its peak power at consumer price. The economy of scale in reusing Cell helps to drive down the cost further. Without which, it means that Sony will need a focused strategy to sell more PS3.

The unified software interface would be a huge win too. CE devices wouldn't be independent systems, but variations on the one system. One OS; one firmware across all platforms; one interface that updates uniformly at no effort; one set of image processing algorithms that can be applied from tiny handheld screens to HDTVs; one set of HCI interface systems - all of it scaling up seemlessly with new devices. It may be cheaper to use a huge mix of different hardware systems and rewrite the software for all of them, but Sony's never been on the 'cheapest solution' end of the market. They have always positioned themselves as superior products at a higher price. They could take up the cost of Cell to simplify their operations and create a very strong Sony buying incentive, where the mobile phone/digital camera/HDTV/media recorder/games console of choice is the one that works in exactly the same way as the other Sony devices you already have. CE devices are getting stupidly complex, and I think a lot of market share can be won by targeting user-friendliness above all else.

If the cost of implementation Cell (due to XDR etc.) is too high, than that was a foolish oversight by the design team, rendering all throw-out Cells worthless, as well as the investment in Cell fabrication.

 

[patsu]

I think essentially, the uniform software part lies in the SPU.

The unified user experiences needs to be carried over multiple architectures.

 

[one]

If the cost of implementation Cell (due to XDR etc.) is too high, than that was a foolish oversight by the design team, rendering all throw-out Cells worthless, as well as the investment in Cell fabrication.

I suspect heat is still a problem even with some SPEs disabled if you compare it with hardware chips. Kutaragi hired Transmeta. PPE is there unlike SpursEngine that has Toshiba's tiny 32bit CPU. Also it has a very high pin count. It needs a southbridge too.

Then there comes the problem of what kind of product you should create with Cell. Basically all throw-out Cell can do are what PS3 can do already with a potentially lower cost including marketing and whatnot. You can plug in a camera to PS3 too. Applications that can be accepted by the mass market are things like BD player and recorder, but it seems these dedicated devices seem profitable and quiet with their hardware chips. If you want smooth UI, you need to clock it up until it starts to require expensive PSU, motherboard and fan etc. Basically you only need Cell for devices that need the computer-like ever-changing nature, or when you need an absolute power that may be overkill for the current CE market. The latter is for the high-end market that needs full chips rather than throw-out Cell, and it doesn't need the output of a whole fab to be profitable. If assigning people and money to such a high-end area makes them more money than trying to use throw-out Cell, why bother? And that high-end includes the PS3 business itself, relatively speaking. Shareholders don't like risky moves, Sony is not in a position to declare another venture.

Meanwhile Matsushita/Panasonic is solid in its business, their in-house platform UniPhier chip manufactured in the 45nm process at Renesas can play Blu-ray in software and is actually deployed in a product (BD recorder DMR-BW800).

http://techon.nikkeibp.co.jp/english/NEWS_EN/20071003/140095/
http://www.prnewswire.com/cgi-bin/stories.pl?ACCT=109&STORY=/www/story/10-31-2007/0004694861&EDATE=

 

[22psi]

oh wells, i just bought a 40gber tonight with a buncha movies... couldn't wait until another revision.

 

[sevanig]

Look at that micro Cell...
Can you say psp2

 

[one]

http://it.nikkei.co.jp/business/news/index.aspx?n=AS1D220A8 22112007

According to Nikkei, Toshiba begins the mass production of 45nm system LSI in the second half of 2008. RSX and chips ordered by Xilinks are among the first candidates. Toshiba already installed the latest manufacturing units in the Oita plant. 45nm samples are expected to be shipped in the first half of 2008. And one tidbit, the mass production of RSX in 65nm begins this December.

 

[patsu]

Cool ! Thanks for posting this news. What about the South Bridge chip ?

 

[Heinrich4]

Agree thats very cooll!

And with RSX 65nm in production soon ,maybe they can reduce console price before march(end year for business/companys etc)... who knows?

(my guess ...console at less 100 total wattage)