Predict: The Next Generation Console Tech

[V3]

To not be a dead end, it needs an architecture roadmap that's being actively developed. Cell would need IBM/STI to be developing new Cells, and if there's no market for these processors, there's no point developing the chips. Cell was developed for a larger market than just PS3. Now that's apparent that it was only PS3 that provided a market, retrospectively we can say the cost wasn't worth it. If Cell2 is only going to feature in PS4, the cost of developing it likely won't be worth it. Now it may be Sony can just double up a current 1:8 Cell and that'd suffice, without needing much development, but it wouldn't provide the ease of development of a more rounded processor (think Cell with better OOO Power core or whatever).

Actually Cell scales to multiple chips, so Sony can put how many Cell they can afford into a single die without needing much development. Creating an improved SPE version 2 on the other hand is sort of a dead end without breaking backward compatibility. Is it wise to put as many Cell into PS4 as they can afford, probably not.

Without Cell in PS4, BC will be difficult. I doubt Sony want to include another BC hardware solution into PS4. But beside Cell what other good candidates for the PS4 CPU ?

I reckon the full 8 SPE Cell at higher clock speed paired with next gen GPU or two should be enough for PS4. Devs will get result (better graphic) faster with that sort of configuration compare to multi Cell with so so GPU.

The fact that Toshiba tried Cell in high-end TVs but has dropped it in favour of custom ASICs pretty much seals Cell's demise IMO.[/QUOTE]

 

[hoho]

Biggest problem with Cell2 is that scaling it up directly would give it tens of SPEs and it's not exactly simple to map regular algorithms to so many separate computing units. With only a handful of them as in PS3 it's not that bad but having to coordinate 20-30 cores (with 2-4 SMT threads in each?) will be tad bit complicated to say the least even if it would provide tons of theoretical raw throughput.

 

[MrSpiggott]

If Sony really do want to reduce their RnD spend then what are the chances that the current Cell, with perhaps the 8th SPU returning due to better yields, being used in PS4?
Current games seem to be performing quite well with the majority of Cell resources being spent helping on graphics. Is a two or three fold increase in performance needed if the GPU is more capable?
Spend what ever resources they have on RAM and a much better GPU and perhaps launch at a lower price.
How disadvantaged would such a console be performance wise?

 

[liolio]

Cell 2 won't happen if IBM and Toshiba see have no interest in the product.
I believe neither are interested in the product, Toshiba moved to something as V3 stated, IBM is treaten in perf by Intel in the super high end server CPU market (last Intel 10 cores part is impressive). I'm not sure a Cell2 could qualify as a counter measure aginst thins threat hence I'm not sure they will spend their money on sharing the R&D effort.

 

[RobertR1]

If Sony really do want to reduce their RnD spend then what are the chances that the current Cell, with perhaps the 8th SPU returning due to better yields, being used in PS4?
Current games seem to be performing quite well with the majority of Cell resources being spent helping on graphics. Is a two or three fold increase in performance needed if the GPU is more capable?
Spend what ever resources they have on RAM and a much better GPU and perhaps launch at a lower price.
How disadvantaged would such a console be performance wise?

That depends on why the competition is doing, I'd say.

It would seem that Sony does not want to be late to the party. They will want to make sure they have a new console ready incase MS makes the jump in 2013. Wii2 will likely get one year of the market to itself.

I hope neither Sony or MS take the low tech road and really go after the performance crown, again.

 

[liolio]

That depends on why the competition is doing, I'd say.

It would seem that Sony does not want to be late to the party. They will want to make sure they have a new console ready incase MS makes the jump in 2013. Wii2 will likely get one year of the market to itself.

I hope neither Sony or MS take the low tech road and really go after the performance crown, again.

I agree but it depends on the crown you're after. If it's about to beat the Wii2 you can bet on it if it's about beating or matching a 2013/2014 PC gamer rig (even a single GPU one)... I'm not confident ar all or a manufacturer will make to take really risky and radical design choices.

 

[AmFreak]

Cell 2 won't happen if IBM and Toshiba see have no interest in the product.

IBM or Toshiba doesn't need to have interest in the product. All it needs is Sony giving IBM enough money.

 

[function]

IBM or Toshiba doesn't need to have interest in the product. All it needs is Sony giving IBM enough money.

Are there any scenarios in which this would be a good use of Sony's R&D budget though?

 

[liolio]

IBM or Toshiba doesn't need to have interest in the product. All it needs is Sony giving IBM enough money.

Doing a proper Cell 2 not just tweak it would const a bunch of money which Sony doesn't seem to have. Cell is the result of the efforts of three companies (including IBM because neither Sony or toshiba alone could have pulled it imho (actually I wonder if IBM alone could have done better while not trying to please everybody) and thus add an healthy development budget.
Yes all it needs is budget/money but it's a pointless argument imho.

 

[AmFreak]

Doing a proper Cell 2 not just tweak it would const a bunch of money which Sony doesn't seem to have. Cell is the result of the efforts of three companies (including IBM because neither Sony or toshiba alone could have pulled it imho (actually I wonder if IBM alone could have done better while not trying to please everybody) and thus add an healthy development budget.
Yes all it needs is budget/money but it's a pointless argument imho.

The question now is, what is a Cell 2?
Surely they won't design a new CPU from the ground up, but some redesign like add more SPU's / PPU's, wider the SPU's, add more local store, etc ... could be a possibillity.

 

[Rangers]

Supposedly in this page, Sony says they were talking about NGP by next gen platform not PS4?

http://translate.google.com/transla...tp://www.j-cast.com/2011/05/27096783.html?p=2

 

[liolio]

The question now is, what is a Cell 2?
Surely they won't design a new CPU from the ground up, but some redesign like add more SPU's / PPU's, wider the SPU's, add more local store, etc ... could be a possibillity.

What I would call a Cell2? A "lesser" larrabee (in regard to how SPU compared to orignal larrabee core). As nAo described it, an instruction cache, an implementation of multithreading and wider SIMD.

I agree that Sony could ask IBM to do some changes, adding a better CPU should be OK, touching to the SPUs I'm not sure. They were pretty state of the art parts for me, transistor density looked pretty hight, power and thermal characteristic were outstanding in regard to the clock speed these things were running at. I'm not sure it would be easy to heavily modify them (assuming for cheap) while not messing up with what made them impressive, they were tiny, cool and fast.
May some experts could shime in on the matter?

 

[eastmen]

I don't think you'd see anything more than a cell with more units and even then it wont be a doubling , perhaps a 1x12 or a 2x12 . I bet the same is going to happen with MS , they will just do a 6 core waternoose cpu with perhaps 8MB of cache instead of the 1MB they have now.


I see both MS and Sony going for the biggest gpus they can with the cpu falling by the way side. I think MS proved with the 360 that the better gpu will win out

 

[hoho]

Another "problem" with the Cell is it fills roughly the same gap that current modern GPUs do: a relatively flexible massively parallel vector unit. There aren't really that many things that a Cell-like architecture could do significantly better than a decent GPU. Back before 2005 it made sense to make an architecture like that as GPU flexibility was rather bad but things are completely different today and will be even more in a year or two when next PS will come out.

 

[borowki]

Creating an improved SPE version 2 on the other hand is sort of a dead end without breaking backward compatibility. Is it wise to put as many Cell into PS4 as they can afford, probably not.

Didn't IBM already make some major improvements to the SPE in the PowerXCell-8i?

 

[AlNom]

That's just double precision support IIRC.

 

[mrcorbo]

Another "problem" with the Cell is it fills roughly the same gap that current modern GPUs do: a relatively flexible massively parallel vector unit. There aren't really that many things that a Cell-like architecture could do significantly better than a decent GPU. Back before 2005 it made sense to make an architecture like that as GPU flexibility was rather bad but things are completely different today and will be even more in a year or two when next PS will come out.

I agree. The rise of GPGPU took a lot of the wind out of Cell's sails.

If Sony opts to move away from Cell for PS4, would it be reasonable to keep it around as a co-processor like the original PS1 processor was in the PS2 and allow for backward compatibility? It could be used to run the OS and provide things like DVR functions, background downloading, etc. so that the main processors can be fully dedicated to running games. Too expensive?

 

[AlNom]

Well, they'd have to re-engineer for 28nm and below too. 8 SPEs + EIB should be tiny, so keeping them around shouldn't be too bad. It might be problematic for layout anyway, as 3dilettante mentioned, even if they were to focus more on main cores.

 

[hoho]

Well, they'd have to re-engineer for 28nm and below too. 8 SPEs + EIB should be tiny, so keeping them around shouldn't be too bad. It might be problematic for layout anyway, as 3dilettante mentioned, even if they were to focus more on main cores.

Going by wikipedia numbers 6x SPE's should take 6x21M transistors or around 126M in total (IIRC games weren't really given access to more than 6 SPEs, out of 8 one was disabled and one was reserved to OS). Add in EIB for some more and you'll probably get somewhere around 140-150M transistors not including PPE (I assume the "real" CPU could emulate it good enough).

If they manage to add it to the new CPU that takes somewhere around 1B transistors in total then it wouldn't probably be out of the question but it would have to be used for something else as well when it's not used for emulating PS3 stuff. ~15% of your transistor budget just sitting there doing nothing wouldn't be the best idea.

 

[AlNom]

it wouldn't probably be out of the question but it would have to be used for something else as well when it's not used for emulating PS3 stuff. ~15% of your transistor budget just sitting there doing nothing wouldn't be the best idea.

Do what first party devs do: MLAA and Tile-Based Deferred Lighting!

Well, I presume that devs could continue to use the SPEs as they do now, physics and audio and AI. I don't think they'd need too many more main cores either - one full fledged core for background OS tasks. edit: three more for... other. They could possibly get away with 6 SPEs since #7 was for the OS and #8 was lasered out.