-
Hey Guest. Check out your NeoGAF Wrapped 2025 results here!
Howard Stringer CEATEC excerpts: 1/2 of Sony's future banking on CELL technol
- Thread starter Kleegamefan
- Start date
Shogmaster
Member
Good gravy... For Sony's sake, Cell better not turn out to be another Emotion Engine.
Kleegamefan
K. LEE GAIDEN
Shogmaster said:
O.K......I'll bite...
What do you mean by that Shog??
Straightballin
Member
I think sony, IBM, and toshiba know what their doing.
Shogmaster
Member
Kleegamefan said:
Don't you guys remember all the lofty plans Sony had with EE back in 1999/2000? Like taking the workstation PC market away from Intel and being embedded into DVD players and such (sounds familair, no?). I have the NextGen issue where Ken Kutaragi talks about all these plans.
None of those plans have come to pass, but I don't think Sony was "banking half of it's future" into the success of EE in those roles.
Shogmaster
Member
Varian said:
That's what worries me. Back in 1999/2000, there were forces within Sony upper management that prevented such reckless gambles by Kutaragi. Maybe with Stringer, Kutaragi has a boss that's willing to take a chance in such gamble.
If it doesn't pay off, it'd be big troble. But then again, all this could merely be Stringer rhetoric.
I don't think KKK has much discretion within Sony after the last reshuffle, and he certainly is no favorite of Stringer.Shogmaster said:
The whole Cell thing is tied to Stringer's general message to analysts that he is paring back unprofitable product lines and focusing on "champion" products. Unfortunately for him he didn't have too many really good product lines to highlight as "champions," so a couple of clunkers (Cell and their MP3 players) had to get thrown in with the truly good product lines (eg PS3 and the new Bravia LCDTV line).
Shogmaster
Member
BTW Here are scans I made way back of the NG article with KK about workstation plans for EE.
Shogmaster said:
Sony's conservative aproach hasn't exactly worked wonders for them. Say what you will about Kutaragi's wrecklessness, but it's a good part of the reason Sony isn't in the tank already. He sure gets bashed a lot for his ambiton and for wanting to push the envelope, and I'll be damned if it doesn't baffle me.
DCharlie
And even i am moderately surprised
lol - the gamble was already MADE. There is nothing Stringer can do ! The investment is pretty much done so what the HELL can he say at this point in time???
"We spent HOW much on this project?!? But... oh jesus, who the HELL was running this place... ?" ??
He _HAS_ to hype Cell ! it's got BOLLOX ALL to do with this imaginary bum-chum friendship you people think everyone has with Kutaragi! This has nothing to do with that and all to do with massive financial undertakings that he _HAS_ to hype Cell up.
Shogmaster
Member
bishoptl said:
Cell's a clunker if they force it into Workstation CPU role. Workstation chips of the future need to be OOOE chips as well as massively parallel. Cell fails on the first role.
Cell will excel as gaming CPU (as did EE) and media decoding chip, but hopefully Sony is not planning those half asses workstation plans for Cell like they did with EE.
Shogmaster
Member
AB 101 said:
Maybe it's all the Sony products I have in the house. A Sony monitor, a sony LCD Tablet Slim tower, a Sony HDTV, couple of Playstations and four $500 Sony Clies I've bought over the years clearly points to someone that hates Sony.
DCharlie
And even i am moderately surprised
And the point is that it's too late for "maybes" - the gamble is already made and Stringer is in a possition where he can do nothing but support it and hope it works.
the whole preceding tone was "Kutaragi was held back from making decissions, now under stringer..." my point being that Cell project was started ages ago, and it's too late for Stringer to do anything of note other than hype it along and try and get it to bring in some cash.
ah - missed your edit.
Sqrt_minus_one
Member
Shogmaster said:
Allow me to disagree with the bolded part, every thing else I pretty much agree with.
Shogmaster
Member
colinisation said:
OOOE is a feature/advancement in CPU design over in-order CPUs. Why would you want to go back to in-order chip design when you don't have to worry about transistor count?
The Take Out Bandit
Member
Shog may be right.
Sony may have to settle for ruthlessly dominating home video game entertainment for yet another generation.
Woe is Sony.
WOE IS SONY! :lol
Sony may have to settle for ruthlessly dominating home video game entertainment for yet another generation.
Woe is Sony.
WOE IS SONY! :lol
Shogmaster
Member
The Take Out Bandit said:
OMG!! TTOB MUST BE GETTING A CUT FROM PS3 BUSINESS! CONGRATS TO YOU!! :lol
sonycowboy
Member
Shogmaster said:
Because general purpose code isn't a necessity for the highly parallizable, floating point intensive code sent to the SPE's, as the PPE with it's 2 hardware threads are sufficient?
Obviously there are trade-offs, but given that IBM / Toshiba / Sony spent a few billion on this specific chip, I'd think they probably have a damn good reason and even better performance improvements that they based it on.
Or do you know better?
Error Macro
Member
Shogmaster said:BTW Here are scans I made way back of the NG article with KK about workstation plans for EE.
LOL @ the "1000x more powerful than PS2" stuff. 6400x4800 resolution! Pixel sized triangles! 120 FRAMES PER SECOND!! omg I can't wait for the PS3!!!
soundwave05
Member
I think Kutaragi's percieved "recklessness" is most likely the reason Kutaragi was not selected as Sony president.
CELL seems like cool technolgy, but I doubt it ever finds a major application outside of the PS3.
Maybe if they could have signed a deal with Apple for use in Macs, but the Intel deal kinda sunk that ship. And honestly, I just don't know why you'd need a teraflop of processing power in a TV or microwave.
CELL seems like cool technolgy, but I doubt it ever finds a major application outside of the PS3.
Maybe if they could have signed a deal with Apple for use in Macs, but the Intel deal kinda sunk that ship. And honestly, I just don't know why you'd need a teraflop of processing power in a TV or microwave.
Shogmaster
Member
sonycowboy said:
In-order nature of XeCPU and Cell was to reduce transistor count and bump clock speed (and why the latest dual core A64s are struggling to get near 3Ghz while the XeCPU can easily get over 3Ghz with 3 cores). It's a necessary compromise for CPUs that are headed for cost sensitive console use.
Shogmaster
Member
Error Macro said:
You missed the best part: 66 million x 1000 = 66 trillion. :lol
The Take Out Bandit
Member
Nope, sorry I still have to pay for my bus pass and beer. 
HomerSimpson-Man
Member
Geez talk about putting nearly all your eggs in one basket, but then again the Playstation line makes up like half their money and Cell is heavily tied to it.
Lord Error
Insane For Sony
My feelings exactly. Sony wouldn't be half what it is without SCE department, and that department probably wouldn't even exist without KK. Some people really seem to worry too much about financial status of these companies, and about some future so far down the line that they'll probably be too old by then to even play games, much less care who leads in "cosole wars".
I'm all for these corporations gambling with all the money they have, if that means much better tech for me.
It will probably be included in PSP2 at the very least, probably in some advanced iteration it will be used in future PS products also. I have a feeling it will find place in many of their products though, instead of various dedicated DSP processors they use now, for example. Too much money is spent on cell research and fabs to be used just for PS3.
Error Macro
Member
Shogmaster said:
Holy cow!
:lol
DAMN YOU, NEW MATH!!
Shogmaster
Member
Onix said:
I'm not 100% sure that this will be a sure fire wide penetration market for CELL. CELL in the PS3 is a 234 million transistor beast. Sure, it will let you stream 48 simultaneous MPEG 2 streams at once (while running 2.8Ghz IIRC), but it won't be used as is for use inside TVs and set top boxes.
Even when you cut the 4 SPEs out of CELL, you still have 130M transistor monster (single core A64s are smaller than that at around 105M). Still too big for such use compared to whatever else is available and can do the same job (No, most folks don't need your TV to decode 24 simultaneous MPEG2 streams, and 2.8Ghz is kinda too fast and hot for something that's gonna be inside your LCD TV).
So then you cut another 2 SPEs out of CELL to get to around 70M+ transistors. Maybe cut the clockspeed down to more managable speeds for such use (like say 1Ghz). Now all of the sudden, CELL isn't so damn amazing anymore, but it's still like 70M+ transistors (bigger than PPC970/G5's 55M.). Will OEMs other than Sony, Toshiba and IBM use it for decoding in TV sets and such? That's a big if IMO.
edit: corrected tranny count for PPC970 and some other math.
soundwave05
Member
I don't really even understand why you'd want CELL technology in a TV. What real practical purpose would it serve?
I would think you'd rather spend extra money on getting a TV with the best picture quality possible for starters.
The big thing now for TV manufacturers is getting those nice, high-end HDTV sets even lower in price, not shoving supercomputer chips into them.
I would think you'd rather spend extra money on getting a TV with the best picture quality possible for starters.
The big thing now for TV manufacturers is getting those nice, high-end HDTV sets even lower in price, not shoving supercomputer chips into them.
I think the plan is to start offering more and more features in TV's and other electronic devices.
While a lot of the power isn't needed (and can be pared down), far more than what is in current electronics is warranted for the features they envision. The reason to use CELL for such future feature-sets is two-fold.
1) It's a scalable architecture. Which means even when pared down, there is economy of scale. The chip becomes cheaper and cheaper to produce over time/production.
2) All CELLs, regardless of power, will have a common OS and interface. That will allow for the inter-connectivity they desire. The 'wired home' as it were. Basically everything can 'speak' the same language, allowing the user to access/control all sorts of content regardless of what device they're using in their house.
Number two specifically is what scares Microsoft, and (in part) led them to produce the Xbox. They know that PC's are now a commodity item (they must keep OS prices down), and that there is no longer huge growth potential due to slowing market expansion .
The next 'big market' is the connected home. Media Center and Media Center extenders were a quick fix to get a bit of a foot-hold in the networked home. But Microsoft is certainly aware that the complexity and user interface of a PC just won't cut it in the long run. People want some thing simplistic and elegant, and want everything to speak the same language. Just look at the sales of the iPod.
What makes Sony one of Microsoft's biggest threats is the combination of points one and two. Many companies have tried to make media devices that 'simplify' things. MP3 streamers, etc. where some early attempts to capitalize on this emerging market. What sets Sony apart is that they have joined up with IBM and Toshiba in order to create a scalable architecture dedicated to audio, video, and communications. Basically, Sony joined with the right companies to produce a powerful and scalable architecture, with an OS designed specifically to take advantage of the chips intended use, and with electronic production and distribution capabilities that can install a very nice user-base.
Then throw in the fact it will be part of the PS3 - a 'killer app' - and Microsoft is very worried that they will be left out of a market that stands to produce tons of revenue and profits - possibly even more than the PC market at some point.
While a lot of the power isn't needed (and can be pared down), far more than what is in current electronics is warranted for the features they envision. The reason to use CELL for such future feature-sets is two-fold.
1) It's a scalable architecture. Which means even when pared down, there is economy of scale. The chip becomes cheaper and cheaper to produce over time/production.
2) All CELLs, regardless of power, will have a common OS and interface. That will allow for the inter-connectivity they desire. The 'wired home' as it were. Basically everything can 'speak' the same language, allowing the user to access/control all sorts of content regardless of what device they're using in their house.
Number two specifically is what scares Microsoft, and (in part) led them to produce the Xbox. They know that PC's are now a commodity item (they must keep OS prices down), and that there is no longer huge growth potential due to slowing market expansion .
The next 'big market' is the connected home. Media Center and Media Center extenders were a quick fix to get a bit of a foot-hold in the networked home. But Microsoft is certainly aware that the complexity and user interface of a PC just won't cut it in the long run. People want some thing simplistic and elegant, and want everything to speak the same language. Just look at the sales of the iPod.
What makes Sony one of Microsoft's biggest threats is the combination of points one and two. Many companies have tried to make media devices that 'simplify' things. MP3 streamers, etc. where some early attempts to capitalize on this emerging market. What sets Sony apart is that they have joined up with IBM and Toshiba in order to create a scalable architecture dedicated to audio, video, and communications. Basically, Sony joined with the right companies to produce a powerful and scalable architecture, with an OS designed specifically to take advantage of the chips intended use, and with electronic production and distribution capabilities that can install a very nice user-base.
Then throw in the fact it will be part of the PS3 - a 'killer app' - and Microsoft is very worried that they will be left out of a market that stands to produce tons of revenue and profits - possibly even more than the PC market at some point.
Shogmaster
Member
I don't know dude. I still don't see why any OEM would choose 70M+ transistor 1 PPE + 2 SPE CELL over something like a cheap but fast ARM11 chip that's like a fraction of the tranny count, but can decode a single H.264 stream just fine.
Hell, even IBM has a direct competitor to CELL for that usage with the LV PPC970FX/G5 that's coming out soon (which has only 55M transistors). Wouldn't the G5 be a much more useful chip than CELL for general usage if these set top boxes and TVs are gonna need more than just decoding capablities?
CELL will be used extensively by Sony and to a lesser degree by Toshiba because they have so much money invested, but I don't see others jumping on to the bandwagon when there are much more reasonable choices of chips in the wild.
I mean will people care if a Sony HDTV can decode three or four H.264 HD stream over a Samsung HDTV that can only decode one or two when they realistically only have one such feed coming into the living room?
Hell, even IBM has a direct competitor to CELL for that usage with the LV PPC970FX/G5 that's coming out soon (which has only 55M transistors). Wouldn't the G5 be a much more useful chip than CELL for general usage if these set top boxes and TVs are gonna need more than just decoding capablities?
CELL will be used extensively by Sony and to a lesser degree by Toshiba because they have so much money invested, but I don't see others jumping on to the bandwagon when there are much more reasonable choices of chips in the wild.
I mean will people care if a Sony HDTV can decode three or four H.264 HD stream over a Samsung HDTV that can only decode one or two when they realistically only have one such feed coming into the living room?
Nevermind, I'm hooked on phonics.
Regarding the 1000 times more powerful comment, Kutaragi mentioned in an interview that when they asked developers how much power they would need to create CG quality graphics during the PSOne era they said '1000x the power of the PSX'. The PS2 was 36 or so times more powerful than the PSX and the PS3 is around the same relative to PS2, resulting in roughly 1000x PSX.
Of course current CG graphics piss all over what was available during the PSOne era, so the PS3 doesn't seem quite there yet. PS3 may not be 1000x PS2, but it does have the ammount of power that Kutaragi originaly desired. In a kind of round about way he got what he wanted.
Regarding the 1000 times more powerful comment, Kutaragi mentioned in an interview that when they asked developers how much power they would need to create CG quality graphics during the PSOne era they said '1000x the power of the PSX'. The PS2 was 36 or so times more powerful than the PSX and the PS3 is around the same relative to PS2, resulting in roughly 1000x PSX.
Of course current CG graphics piss all over what was available during the PSOne era, so the PS3 doesn't seem quite there yet. PS3 may not be 1000x PS2, but it does have the ammount of power that Kutaragi originaly desired. In a kind of round about way he got what he wanted.
Shogmaster
Member
Kangu said:
All that typing for nothing huh?
Shogmaster said:
Shog, come on bud... whats with the half-assed posts? Seriously, they're are getting old.
The current incaration of Cell | STI Broadband Processor Architecture are designed expressely for for PlayStation3. The OoOE comment is utter BS, which I'll get to in a bit. As for Cell being a "clunker," the only true problem with it at this moment is it's lack of full-speed DP-FP FPUs. Currently, the SPUs support a subset of the IEEE 754 specification; which works for them as in real-time, non-critical applications full-complience is unneeded and not used as it would artificially increase the logic count and decrease area effeciency.
Yet, what STI is currently doing down in Austin to the BPA (which is an architecture, not a specific incarnation) is replacing the FPUs in the SPUs with fully IEEE754 complient DP units. This IC will excel in the DCC workstation role or any type of workstation/supercomputer functionality which is heavily computationally bound. OoOE, which is primarily a way to mask L2 latencies, is just not applicable to the discrete marketplace. The move to x86 wasn't due to it's inclusion of OoOE, but rather the realization that x86's economies of scale are such that the [cost]:[preformance] ratio tips in their favor over discrete solutions. If you look at the Vector processing systems that NEC or Cray (which seems like basically an extention of the US Defense Establishment) are both producing systems using custom IC which drastically outpreform your OoOE x86 offerings, but they don't have economies of scale to compete with x86 -- Cell will.
Shogmaster said:
Transistor counts mean nothing in this [manufacturing] context. Nor would your idea of respinning DD2 into some PPE+4SPE IC be feasible, logical or even sensible. Sony Group is a semiconductor company, a very large one. They pay on a per-wafer basis. They pay their $1300 for a 300mm pd-SOI wafer and run it through their 90nm line; a few weeks later, on the otherside of the line, out pops a wafer with a bunch of LSIs etched into them. Not all will be functional due to intrinsic errors in the manufacturting process; but Cell is a design that has over 50% of it's area devoted to redundant structures. If you calculate out the probabilities, you'll see that it's likely that the log of the number of errors against the log of the number of yeildable ICs (at variable SPU levels) will follow a power law.
What this means is that STI, in their required desire to net a mandatory number of ICs for use in Sony's PlayStation3, will have a lot more ICs which are fabricated and only have lesser number of SPUs operational. These ICs can't be used in the PS3, but they can be used elsewhere; Sony payed for them when they bought the wafer and if they can use it somewhere, anywhere, they amoritize the costs. This isn't that complicated.
Basically. They buy the wafer. Whatever they yeild which is operational should go to use: it doesn't matter where. If it has 8 SPUs you sell it in a Blade server at high proft, 7 SPUs goes into a PlayStation3, 6 SPUs are SMPed and put into a homeserver, say 4 SPUs go into an ES series recierver, etc.
But, to actually propose that they create a set-piece PPE+4SPU ASIC at this point, when yeilds suck, is utterly asinine. Something like that requires a full redeign and backend synthesis, as well as fixed-costs for the tape-outs, respins, etc. It's just not smart.
Shogmaster said:
11 FO4; As Marco of B3D is fond of pointing out it's microarchitecture is amazingly effecient. Add to it an effecient ISA, common computing enviroment, ohh, and perhaps the preeminent secure computing enviroment offered anywhere.
Shogmaster said:
They are not competitors, each has a niche. Each will fill it.