End of Cell for IBM

[Deleted member 7537]

Don't think that sony will drop Cell that easily, they have spent a lot of money developing tools specifically for this arquitecture, not to mention the investment on the development of the processor itself.

Anyway, i'm sure that Hirai will talk about this soon.

 

[Crossbar]

In an optimistic view, this could allow Sony to tailor a "Cell2" to the needs of a Console without having to think about reuse of the same chip in servers.
Just look at how much unused diespace in the 45nm version of Cell is, surely you could place a couple of other chips there that the PS3 needs anyway, and the FlexIO has nearly half its lanes unused...

What do you mean with unused die space? Are you talking about the sides that don´t have any pins? Those will likely disappear in the 32 nm shrink.

Unused FlexIO lanes? Are you sure about that, I think they are all being used.

 

[Carl B]

He's talking about the 'dead' silicon that has increased in relative area with each shrink of the Cell, which is of course inefficient from a die size perspective. But there would likely have been a more thorough job done on the shrink at 32nm anyway; 65nm and 45nm were quick and dirty shrinks, but with the move to metal gates I think 32nm would have been (and might still be) a logical place to optimize.

 

[V3]

Well, I don't think you should boil it down to that - in fact if anything it even underplays Sony's own involvement. Toshiba was for the SPEs in terms of philosophy definitely, with IBM out for Power and its interests, but IBM engineers contributed hugely to Cell's development, down to the supported instructions of the SPEs. It was a team of volunteer engineers within IBM operating under a different aegis than IBM corporate 'central' we should remember, and a lot of them were onboard with the small/efficient/fast idea.

Yes I am pretty sure they all contributed to the project. Cell itself was impressive in what it was set out to do. PS3 was not. Toshiba Cell TV looks impressive.

The architecting was around ~$500M... which is no joke it should be mentioned...

That's shared between STI though. I remember around $400M but yeah around there. Where as NV, AMD or Intel would plunge alot more than that for R&D. Intel probably invested more than that on Larrabee alone. This is not much different to how much they invested for Emotion Engine and GS or technology in the original PSX. I think PSX investment was around $500M too.

The investment isn't out of the norm but PS3 not as successful as the previous Playstation is though for Sony. So the successor to PS3 may come much later before they want to finance another $500M.

and the capital investments in fabs and related were in the billions. They recouped some of that with the fab transfers to Toshiba, but... it definitely is a lot of money.

Kutaragi wanted to steer Sony in that direction at the time. Cell or no Cell, they were entering fabrication to provide them with some advantage. It would given them an advantage if it actually met their goal. But they couldn't bring 65 nm on time for Cell. So it's all kind of a waste of money if they can't get that kind of advantage by owning their own fab. They invested on eDRAM process too if I remember correctly but they went with RSX instead. Things just fall apart at the end.

He's talking about the 'dead' silicon that has increased in relative area with each shrink of the Cell, which is of course inefficient from a die size perspective. But there would likely have been a more thorough job done on the shrink at 32nm anyway; 65nm and 45nm were quick and dirty shrinks, but with the move to metal gates I think 32nm would have been (and might still be) a logical place to optimize.

I remember reading that they optimise the shrink for power consumption not size. Cell could be smaller. It's around 250M transistors. The PS3 only used 7 SPUs, when they restructure it for 32nm they should be able to remove one and make it smaller, there is no need to keep disabling one SPU for yield.

 

[AlNom]

with the move to metal gates I think 32nm would have been (and might still be) a logical place to optimize.

Who is in charge of die shrinks

 

[-tkf-]

Most annoying thing about these changes in CPU architecture is the loss of knowledge. Argh!

If Sony went with a Cell 2 with a higher clock and more SPU´s wouldn´t they be able to harvest most of their hard learned experience with everything from compilers to graphics engine design?

 

[rpg.314]

The PS2 knowledge and investment didn't stop them from throwing cell1 at the devs.

 

[Crossbar]

Who is in charge of die shrinks

Probably IBMs East Fishkill plant, together with engineers from Toshiba.

http://www.semiconductor.net/articl...liance_for_32_nm_Bulk_Process_Development.php

Toshiba will likely be responsible for the bulk production.

Here are some rumours of Sony (and Nintendo) consulting East Fishkill for services.

http://www.recordonline.com/apps/pbcs.dll/article?AID=/20090723/BIZ/907239960/-1/BIZ2101

The timing fits pretty well with the Alliances 32 nm CMOS production lines being ready for pre-production.

 

[-tkf-]

The PS2 knowledge and investment didn't stop them from throwing cell1 at the devs.

Thats my point, Microsoft did the same.

 

[Lazy8s]

Processing the logic for gameplay and orchestrating the many threads of a modern game application has more in common with running a PC OS than any streaming, ops crunching workload, leaving PS3 games under CELL more CPU limited, whether in design ambition or simply bottleneck, than they would've been with a competent, cost-comparable PC CPU.

Delegating graphics and physics to a GPU is more efficient, both in terms of processing and die area usage, than to CELL.

CELL was never competitve in any category of processing unit. HPC would be the closest fit, but I think Clearspeed's architecture offered better possibilities, not to mention other designs which could be targeted at that sector.

Sony themselves were never able to justify adapting CELL as an embedded solution for any other of their many lines of consumer electronics. They use PowerVR from their high-end Bravias for video processing down to their Ericsson mobile phones for graphics.

 

[Billy Idol]

IBM is pulling out of Cell development

http://www.playstationuniversity.com/ibm-cancels-cell-processor-development-1295/

I am really not surprised, at least from a HPC point of view...

 

[Panajev2001a]

No, OTSS is still a joint-venture, and Sony still owns 40% of the NTC...

It would be fun to see LRB produced there ... ok ok, no concrete info... but I think PS4 will be x86 + LRB...

single shared pool RAM set-up + dual or quad core Sandy Bridge based CPU + LRB based GPU... .

 

[Arwin]

I think Cell in the PS3 was a great success, at least at a technical level. It really contributed a lot. The RSX on the other hand lacked innovation and was more or less out-dated by the time the PS3 came out, and not quite up to the competition with the much more daring and innovative ATI card in the 360.

What's becoming clear though is that in this generation more than in any other, PC development has been extremely influential on console game development. I don't think we've ever seen this many games developed on PC and ported to the consoles. And this has always, every single time, been detrimental to the PS3's performance in multi-platform games - it was too different to be able to easily port a game from PC.

I could very well see that it would be too hard to find any kind of SDK solution to overcome this problem, which would result into Sony looking seriously into finding a hardware setup that is more in tune with PC developments. I think that goes double now that peripherals and software are becoming more important versus hardware.

I'm not saying this is a guarantee though mind you.

I do think there's a strong chance that the Cell will come in the PS4 in some shape or other, maybe in a spursengine type form, both for backward compatibility and because it should still be relatively easy to find a good use for it. And they may of course still go with a Cell centric design, but I'm assuming they'll have been working closely with one of the graphics vendors to create a well integrated design using whichever technology suits best.

 

[liolio]

I think Cell in the PS3 was a great success, at least at a technical level. It really contributed a lot. The RSX on the other hand lacked innovation and was more or less out-dated by the time the PS3 came out, and not quite up to the competition with the much more daring and innovative ATI card in the 360.

I don't agree as it fail to find its place in most markets it aimed at. Toshiba had to revamp it to be in a situation there can ship something based on it. IBM couldn't ship it as an autonomic part. SPU are a really good, low power, damned fast.

What's becoming clear though is that in this generation more than in any other, PC development has been extremely influential on console game development. I don't think we've ever seen this many games developed on PC and ported to the consoles. And this has always, every single time, been detrimental to the PS3's performance in multi-platform games - it was too different to be able to easily port a game from PC.

it's not true for all developers for instance Dice, and Cell excuse is a bit old how do you know some problem are not GPU related?

I could very well see that it would be too hard to find any kind of SDK solution to overcome this problem, which would result into Sony looking seriously into finding a hardware setup that is more in tune with PC developments. I think that goes double now that peripherals and software are becoming more important versus hardware.

Basically there may be no solution because Sony made weird choices early on in regard to the balance between CPU and GPU. I'm unsure I get what you mean by PC development, if it's SMP with coherent memory model well that goes way beyond PC development.

I'm not saying this is a guarantee though mind you.

I do think there's a strong chance that the Cell will come in the PS4 in some shape or other, maybe in a spursengine type form, both for backward compatibility and because it should still be relatively easy to find a good use for it. And they may of course still go with a Cell centric design, but I'm assuming they'll have been working closely with one of the graphics vendors to create a well integrated design using whichever technology suits best.

I hope not Panajev wants a single chip, Repi too, Nao too, etc. for SPU to be up to the task in graphics they should have to go through such a rework that they would be no longer have anything to do with SPUs, starting for scratch or using existing parts seems like a better idea.

 

[Arwin]

I don't agree as it fail to find its place in most markets it aimed at. Toshiba had to revamp it to be in a situation there can ship something based on it. IBM couldn't ship it as an autonomic part. SPU are a really good, low power, damned fast.

I said though that the Cell was a success in PS3.

it's not true for all developers for instance Dice, and Cell excuse is a bit old how do you know some problem are not GPU related?

Of course, some problems are GPU related. But in many cases, the PS3 simply has been treated as a single PPE with a split memory pool that means the GPU RAM maxes out at 256MB, versus the 360 with three PPEs with a unified memory pool that maxes out at close to 500 of GPU RAM. I'm exaggerating a little of course, but only a little. In most cases in the first years the SPEs were ignored completely. Things are changing now, in that some of the more tech-savvy PC developers are doing more with Cell, and are also more interested in doing so because this is not that different from the direction that PC development is moving into with the DirectX11 setup.

I hope not Panajev wants a single chip, Repi too, Nao too, etc. for SPU to be up to the task in graphics they should have to go through such a rework that they would be no longer have anything to do with SPUs, starting for scratch or using existing parts seems like a better idea.

I meant as a co-processing chip on a much lower level, similar to how the PS1's CPU was used as a controller chip in the PS2. A four to six SPE SpursEngine type chip (i.e. the version that doesn't have a PPE) might be really cheap at this point, and could still really help keeping backward compatibility.

 

[-tkf-]

Processing the logic for gameplay and orchestrating the many threads of a modern game application has more in common with running a PC OS than any streaming, ops crunching workload, leaving PS3 games under CELL more CPU limited, whether in design ambition or simply bottleneck, than they would've been with a competent, cost-comparable PC CPU.

Delegating graphics and physics to a GPU is more efficient, both in terms of processing and die area usage, than to CELL.

CELL was never competitve in any category of processing unit. HPC would be the closest fit, but I think Clearspeed's architecture offered better possibilities, not to mention other designs which could be targeted at that sector.

Sony themselves were never able to justify adapting CELL as an embedded solution for any other of their many lines of consumer electronics. They use PowerVR from their high-end Bravias for video processing down to their Ericsson mobile phones for graphics.

Uh?

All over the place?, this is about consoles not TV´s

I would say that the Cell made the PS3 what it is today, without it and a common CPU instead it would have been a disaster. I think that Uncharted 2 and Killzone 2, the best looking games from this generation, is all the evidence that is needed.

 

[Shifty Geezer]

I don't agree as it fail to find its place in most markets it aimed at. Toshiba had to revamp it to be in a situation there can ship something based on it. IBM couldn't ship it as an autonomic part. SPU are a really good, low power, damned fast.

From a technical POV, I think Cell very successful. What orher processor has been able to offer the price/performance/power consumption in realworld task over the past few years? And what was on the table as a viable platform in 2000/2001 when Sony were looking for a next-gen performance CPU? The real reason Cell has died was STI failed to get other parties on board. If they had cultivated a development community, the value of Cell would have reached its potential that we were talking about in the early days. However, they didn't encourage developers, which didn't cultivate the software knowhow, which made Cell too difficult so it was avoided. If they had actually got behind Cell with full software support, it'd be a different story now.

You can't just supply hardware and expect it to make it on its own. PS2 got lucky. Software and support is 2/3rds the experience.

 

[Gubbi]

I said though that the Cell was a success in PS3.

I wonder if history will be so kind to CELL. With Uncharted 2 we finally have a title that utilizes CELL to its full potential, - four years into this generation !!

Cheers

 

[liolio]

I said though that the Cell was a success in PS3.

It's disputable. It made sense whereas for IBM and Toshiba it's a bit more iffy Toshiba goes with SPURS and the respin took them some time, IBM did a DP version and the thing is still coupled with opterons.
The dilemma is that a four cores CPU that would have been kind of "an early compute oriented only larrabee" (In Order, 256bits wide SIMD units, supporting float and integer unlike wmx128, compiler friendly API for SIMD, supporting predication, scatter gather, two hardware threads, offering a 1/2 ratio for SP/DP, hardware fix to LHS, top class prefetch ability, backed by huge bandwidth( Xdram), and super fast FlexIO for multi chips use more than two chips and able to ensure memory coherency) would have done righ for IBM and Sony in the ps3, but it would have end useless for Toshiba and used Sony planned for it in other devices. On the other hand for these tasks Sony and Toshiba may have develop a VPU less specialized than actual GPU, scalable (at chip level) that may have also work as the PS3 GPU.
Too diverging goals imho.

Of course, some problems are GPU related. But in many cases, the PS3 simply has been treated as a single PPE with a split memory pool that means the GPU RAM maxes out at 256MB, versus the 360 with three PPEs with a unified memory pool that maxes out at close to 500 of GPU RAM.
I'm exaggerating a little of course, but only a little. In most cases in the first years the SPEs were ignored completely. Things are changing now, in that some of the more tech-savvy PC developers are doing more with Cell, and are also more interested in doing so because this is not that different from the direction that PC development is moving into with the DirectX11 setup.

As you said things have changed. In regard to devs well I would not say only "more tech savy" they deal with time line and multiple systems, I would prefer to say it took them time which was to be expected. More the Cell enforced a task based system that was not really needed elsewhere (pc, Wii, 360) at the time. Dice is for example happy as it pushed them in the good direction but they are really focus on what they are doing I'm unsure a lot of studio have this chance.

I meant as a co-processing chip on a much lower level, similar to how the PS1's CPU was used as a controller chip in the PS2. A four to six SPE SpursEngine type chip (i.e. the version that doesn't have a PPE) might be really cheap at this point, and could still really help keeping backward compatibility.

Well our system will last quiet sometime it seems, they'll end cheap I'm not sure BC will be worse the cost (even if tiny), neither the ps3 or the 360 are backward compatible and I could see editors move to re-edition instead only for some big much wanted games.

 

[liolio]

CELL was never competitve in any category of processing unit. HPC would be the closest fit, but I think Clearspeed's architecture offered better possibilities, not to mention other designs which could be targeted at that sector.

Sony themselves were never able to justify adapting CELL as an embedded solution for any other of their many lines of consumer electronics. They use PowerVR from their high-end Bravias for video processing down to their Ericsson mobile phones for graphics.

It's doing really great at video/audio processing but it's over-sized for as an embedded solution in in consumer electronics.