What Are The Fastests Cells Out in "The Wild"?

[darkblu]

Phooey! You're just peeved 'cause your Al box and pile of LEDs are out of date..

you mean that good-for-nothing, arranged-in-a-480x272-matrix pile of LEDs? ;p

 

[dodo3]

The Inquirer Attacks!

http://www.theinquirer.net/default.aspx?article=32978

"Cell yields are in the toilet." - The Inquirer

 

[Heinrich4]

I have a heard(thanx my friend) a cell 3.2GHz since feb/06 in sdks.

 

[Carl B]

I think this says it all as to the level at which Charlie D tries to understand things...

I can't see how 40 per cent will ever make the PS3 economically viable.

...indeed Charlie, indeed.

 

[Bobbler]

Haha, I was going to actually post that in this thread -- I thought it was kind of funny, in the most tragic sense.


I wonder what the initial yields of the EE were -- that was a rather colossul chip at the time and had no redundancy to speak of.

 

[archie4oz]

Quite contrary to popular modern mythology (e.g. Cell GPUs and such), the EE didn't particularly have many yeild issues... The GS OTOH...

 

[SPM]

Surely the yields of Cell are likely to be quite a bit higher than Xenon, since both are similar in size, and because the SPEs take up about half the chip area, the defect rate using 7 SPEs will be about half what it would otherwise be.

 

[Carl B]

Surely the yields of Cell are likely to be quite a bit higher than Xenon, since both are similar in size, and because the SPEs take up about half the chip area, the defect rate using 7 SPEs will be about half what it would otherwise be.

You know which raises a point. It was my understanding from the original interview that the 'yields' refered to were for 8-SPE chips, such that 'logic redundancy' led to a 20-40% yield rate for 8-SPE chips, and that 7-SPE chips should obviously yield a lot higher. Thoughts on this? I didn't read the 'logic redundancy' as SPE redundancy though.

 

[Acert93]

Worse case scenario is that Cell is too big and too hot to obtain great yields, takes it on the chin this fall, and transitions to 65nm early in 2007 and the problem is solved. I honestly would not be surprised if a number of factors, including Cell and BluRay (both standards issues and production issues) along with software, tools, and api development made the Spring release unrealistic financially and that the major resolutions to these issues will come in 2007. We have seen how just a couple minor parts crippled the 360's production. Ramping these things up is pretty difficult.

But all the poo-pooing ignores that Sony has released 3 other consoles and a host of consumer electronics and that even given a worse case scenario 2007 will bring pretty quick relief for when the sales begin to really matter. Obviously any launch of this magnititude is difficult and there are bound to be some shortages, but it is not like Sony is the new kid on the block here. My guess they are pretty serious about the 6M number mark and the detailed information (2M at launch, 2M more by the end of calendar 2006, 6M total by March 31st, 2007) indicates we will see closer to 6M units than 3M.

 

[Bobbler]

Quite contrary to popular modern mythology (e.g. Cell GPUs and such), the EE didn't particularly have many yeild issues... The GS OTOH...

Where was EE fabbed? Where was GS fabbed?

I thought GS was about half the size of EE? What was the culprit in GS yielding issues?

Fabbing seems to be a fickle beast...

 

[Acert93]

Surely the yields of Cell are likely to be quite a bit higher than Xenon, since both are similar in size, and because the SPEs take up about half the chip area, the defect rate using 7 SPEs will be about half what it would otherwise be.

Xenon is 168mm^2 while Cell is 235mm^2. Cell is 40% larger in die area than Xenon.

 

[Carl B]

Where was EE fabbed? Where was GS fabbed?

I thought GS was about half the size of EE? What was the culprit in GS yielding issues?

Fabbing seems to be a fickle beast...

Oh no, the original GS was a giant to be sure.

(and both were larger than present-day Cell)

 

[Bobbler]

You know which raises a point. It was my understanding from the original interview that the 'yields' refered to were for 8-SPE chips, such that 'logic redundancy' led to a 20-40% yield rate for 8-SPE chips, and that 7-SPE chips should obviously yield a lot higher. Thoughts on this? I didn't read the 'logic redundancy' as SPE redundancy though.

Wonder if we can get any sort of clarification from someone at IBM in regards to the comment -- the Cell guys over at the IBM forums seem to be pretty good about answering questions. I wouldn't be surprised if it was 20-40% with 7SPE, but I'd love for it to be wrong, if only to spite Charlie for his continually dubious behavior.

 

[Acert93]

(and both were larger than present-day Cell)

That is something to chew on: Cell+RSX take up ~ the same die area as EE+GS did after their first shrink to 180nm in 2000. Further, Sony will be shifting to 65nm very early in 2007.

We may see similar die areas for Cell and RSX within 6 months which took Sony 2 years to accomplish with the EE and GS.

 

[Bobbler]

Oh no, the original GS was a giant to be sure.

(and both were larger than present-day Cell)

Holy cow, that is a rather giant chip -- especially back in 99.

 

[TheChefO]

That is something to chew on: Cell+RSX take up ~ the same die area as EE+GS did after their first shrink to 180nm in 2000. Further, Sony will be shifting to 65nm very early in 2007.

We may see similar die areas for Cell and RSX within 6 months which took Sony 2 years to accomplish with the EE and GS.

Interesting indeed - So the $100 price drop they produced in the two year time frame may be possible in 6 months without bleeding the company further.

So we could see a very aggressive:

2007 (May?) $400-500
2008 (May?) $300-400
2009 (May?) $200-300

However this would not give them much room for profit on ps3 hardware which they seem keen on this gen so it will probably be a bit less aggressive.

Perhaps:

2007 (May?) $450-550
2008 (May?) $400-500
2009 (May?) $350-450

The die size in relation to process improvement this gen seems likely to allow a rather quick price drop but I don't know if Sony will be quick to pass this price drop on to consumers or if they will use the extra cash to help 'balance the books'.

 

[Carl B]

Let's try to keep this from becoming another pricing thread; it's so much better off as a yields thread.

 

[archie4oz]

Where was EE fabbed? Where was GS fabbed?

EE : Oita TS Semiconductor
GS : Sony Kokubu Corporation (initial run), from there after SCEI Nagasaki Fab 1 and 2 (all now officially called Sony Kyushu Semiconductor Corporation).

 

[Bobbler]

EE : Oita TS Semiconductor
GS : Sony Kokubu Corporation (initial run), from there after SCEI Nagasaki Fab 1 and 2 (all now officially called Sony Kyushu Semiconductor Corporation).

I guess that is sort of expected, although I would have thought EE would have been at Nagasaki and GS at OTSS -- not for any technical reason, but Toshiba was responsible for GS (?), so I figured they'd have some hand in fabbing.

On a side note, is OTSS stated to fab any PS3 parts? It's a 300mm plant now, if I recall? (or maybe there was just talks and it isn't yet? not really sure)

 

[archie4oz]

I guess that is sort of expected, although I would have thought EE would have been at Nagasaki and GS at OTSS -- not for any technical reason, but Toshiba was responsible for GS

I'm always amused that the Toshiba/GS connection keeps coming up... I wonder why this myth continues to thrive... Anyways the Nagasaki fabs do fab the EE+GS chips though...

On a side note, is OTSS stated to fab any PS3 parts? It's a 300mm plant now, if I recall? (or maybe there was just talks and it isn't yet? not really sure)

Supposedly OTSS is scheduled for 65nm Cell fabrication (I presume for PS3 although Toshiba might use some elsewhere (HD-DVD decks perhaps). The Nagasaki fabs are supposed to be spitting out Cells and RSXs (65nm and 90nm respectively)...

Oh no, the original GS was a giant to be sure. (and both were larger than present-day Cell)

Those GSs (279mm) and EEs (240mm) never shipped in massive enough quantities... They were only in the initial DTLs.