so PS3 will have ~534 million transistors + other questions

the Playstation3 is going to have roughly 534 million transistors between Cell and RSX.
that is about as many transistors expected for the CPU alone, back when PS3 was supposed to be getting an Emotion Engine 3.


the 218 Gflops, does this come from the 7 SPEs alone, or is it from the 7 SPEs plus the PPE with its VMX unit ?

in the recent gamespot article, it was said that Playstation3 might improve.

might they activate the 8th SPE, crank up the core clockspeed, use the 2nd revision of Cell which has 250M transistors, or all of the above?
(if anything)


I don't see Sony/SCEI going to an all-new type of architecture with PS4. as some of us have said before, the Cell technology will probably be drasticly scaled up for PS4, use more cores on a smaller process
(32nm, 22nm etc). I don't expect clock frequency to grow very drastically for PS4. maybe 7-10 GHz. (if even that) which would be the smallest increase generation over generation.

PS1 CPU: 33 MHz to PS2 CPU: 297 MHz = 9x increase

PS2 CPU: 297 MHz to PS3 CPU: 3200 MHz / 3.2 GHz = 10.7x increase

PS3 CPU: 3200 MHz / 3.2 GHz to PS4 CPU: 10 GHz (max) = 3.1x increase.

I realize that the PS4 CPU may not even reach anywhere near 10 GHz, but i'm just tossing that number out. so with the probable loss of the standard 9x to 10x increase in clockspeed, where is PS4 CPU going to gain performance, to even reach the same increase over PS3 CPU that PS3 CPU is going to be over PS2 CPU? more cores? yes. and more effiecient cores? hopefully. more agressive process technology compared to PS3 CPU's conservative 90nm for 2006? hopefully. larger die size compared to PS3 CPU? hopefully. more than one die? hopefully. :)

since I am not in the technology industry, I really dont have a clue as to everything that Im talking about. just going by the gen over gen advances that I've observed.



anyway back to what I was saying...

another 35x increase in flops performance over PS3 CPU would give us about 7.6 TFLOPs which sounds pretty good. that would meet and exceed the original goal of 1000 times PS2 performance, assuming Sony originally meant 1000x PS2's CPU flops performance.


then there's the PS4 GPU. lets look at Nvidia's roadmap circa 2004 at GeForce 6800s launch:

1n71vj.jpg


1n86so.jpg


1n96yd.jpg



I *don't* think this is an actual roadmap of exactly where Nvidia will be, as it is just a scaling of the GeForce FX 5900 process into the year 2014, but I suppose we can use this as a rough basis.

by 2012 when PS4 should probably be due out, I think we'll have at least a 2 billion transistor Nvidia GPU with 30 to 50 billion vertice/sec performance, shader model 5.0+ operating at 1.5 GHz or faster.

ok I've just lost my train of thought. going to post this now and edit in anything else later 8)
 
Well i won't comment on possible speed bumps on PS3 final specs, but regarding PS4:

I really hope that the monumental investments Sony/Toshiba/IBM have made and are making will be a bit longer lasting than 5-6 years.

I really hope, for them, that they won't have to spend billions upon billions to build a completely new architecture every 5 or 6 years.

It seems the Cell project is designed to last a long time (just adding more Cells or just modifying the existing designs) so maybe PS4 won't cost as much as PS3 to design, in terms of R&D. So Sony might have spent an impossible amount on the PS3 architecture, but they might just get it all back since they won't have to do as much work for PS4...

This is all my own speculation...
 
Re: so PS3 will have ~534 million transistors + other questi

Megadrive1988 said:
the 218 Gflops, does this come from the 7 SPEs alone, or is it from the 7 SPEs plus the PPE with its VMX unit ?

218 is the total rating, PPE+VMX included.

Megadrive1988 said:
in the recent gamespot article, it was said that Playstation3 might improve.

It's possible, but unlikely IMO. Of course RSX is still apparently under development, so one would hope it's under constant improvement ;) They didn't announce too many details to provide us a guide on what they could improve on, however.

Megadrive1988 said:
might they activate the 8th SPE, crank up the core clockspeed, use the 2nd revision of Cell which has 250M transistors, or all of the above?
(if anything)

I doubt the first two, but I'd say it's likely they're using the 2nd revision (?)

Interesting nvidia slides! Methinks they aren't a serious projection for the future..I highly doubt we'll see 100Ghz GPUs in 10 years ;)
 
london-boy said:
Well i won't comment on possible speed bumps on PS3 final specs, but regarding PS4:

I really hope that the monumental investments Sony/Toshiba/IBM have made and are making will be a bit longer lasting than 5-6 years.

I really hope, for them, that they won't have to spend billions upon billions to build a completely new architecture every 5 or 6 years.

It seems the Cell project is designed to last a long time (just adding more Cells or just modifying the existing designs) so maybe PS4 won't cost as much as PS3 to design, in terms of R&D. So Sony might have spent an impossible amount on the PS3 architecture, but they might just get it all back since they won't have to do as much work for PS4...

This is all my own speculation...
makes sense :)

I also believe that instead of having to design a completely new CPU architecture, as they have been doing ..... I believe PS4 will likely use something like 8 4th generation CELL processors built on at least a 45 nm (but hopefully a 32 nm) process...
 
I personally believe the production DD2 die had the PPE beefed up specifically because they knew they were going to lose an SPE in the PS3 implementation. The server units i would imagine they get the full 8 SPE function version since they're not exactly volume parts.
 
Wunderchu said:
london-boy said:
Well i won't comment on possible speed bumps on PS3 final specs, but regarding PS4:

I really hope that the monumental investments Sony/Toshiba/IBM have made and are making will be a bit longer lasting than 5-6 years.

I really hope, for them, that they won't have to spend billions upon billions to build a completely new architecture every 5 or 6 years.

It seems the Cell project is designed to last a long time (just adding more Cells or just modifying the existing designs) so maybe PS4 won't cost as much as PS3 to design, in terms of R&D. So Sony might have spent an impossible amount on the PS3 architecture, but they might just get it all back since they won't have to do as much work for PS4...

This is all my own speculation...
makes sense :)

I also believe that instead of having to design a completely new CPU architecture, as they have been doing ..... I believe PS4 will likely use something like 8 4th generation CELL precessors built on at least a 45 nm (but hopefully a 32 nm) process...

That was my feeling too. Just put in the fastest Cell configuration available at the time of release and slap a new NVIDIA GPU. Some RAM, salt, pepper, and off u go.
 
london-boy said:
Well i won't comment on possible speed bumps on PS3 final specs, but regarding PS4:

I really hope that the monumental investments Sony/Toshiba/IBM have made and are making will be a bit longer lasting than 5-6 years.

I really hope, for them, that they won't have to spend billions upon billions to build a completely new architecture every 5 or 6 years.

It seems the Cell project is designed to last a long time (just adding more Cells or just modifying the existing designs) so maybe PS4 won't cost as much as PS3 to design, in terms of R&D. So Sony might have spent an impossible amount on the PS3 architecture, but they might just get it all back since they won't have to do as much work for PS4...

This is all my own speculation...


pretty good speculation, in my opinion
 
Re: so PS3 will have ~534 million transistors + other questi

Megadrive1988 said:
might they activate the 8th SPE, crank up the core clockspeed, use the 2nd revision of Cell which has 250M transistors, or all of the above? (if anything)
DD2 is what PS3 is supposed to have as Suzuoki said that DD1 was unsatisfying in power consumption and they could iron out it just before the end of 2004 then the tapeout of DD2 followed. PPE was improved to be 2-issue and got more complicated, probably it's one of the reason the clockspeed of Cell in PS3 is at 3.2Ghz.
 
Re: so PS3 will have ~534 million transistors + other questi

one said:
DD2 is what PS3 is supposed to have as Suzuoki said that DD1 was unsatisfying in power consumption and they could iron out it just before the end of 2004 then the tapeout of DD2 followed. PPE was improved to be 2-issue and got more complicated, probably it's one of the reason the clockspeed of Cell in PS3 is at 3.2Ghz.
ahh ... makes sense, as you state, as one of the possible reasons as to why PS3 's implementation of CELL is clocked surprisingly low..
 
Re: so PS3 will have ~534 million transistors + other questi

Wunderchu said:
one said:
DD2 is what PS3 is supposed to have as Suzuoki said that DD1 was unsatisfying in power consumption and they could iron out it just before the end of 2004 then the tapeout of DD2 followed. PPE was improved to be 2-issue and got more complicated, probably it's one of the reason the clockspeed of Cell in PS3 is at 3.2Ghz.
ahh ... makes sense, as you state, as one of the possible explanations as to why PS3 's implementation of CELL is clocked surprisingly low..

:oops: 2GHz would be surprisingly low...

3.2GHz - NOT final - is pretty cool to me..
 
Re: so PS3 will have ~534 million transistors + other questi

london-boy said:
:oops: 2GHz would be surprisingly low...

3.2GHz - NOT final - is pretty cool to me..
hehehe .. you have a point.. 3.2 GHz is not a clock speed to be ashamed of :)
 
Cool thread Megadrive - I always like prediction topics!

I think that Cell probably will form the basis of PS4 at this point, simply because indeed if little else, it is extremely scalable. There's big upside to that from an R&D perspective, but of course the downside is that there's no new 'revolutionary' architecture for us to speculate on for the next 5+ years. 8)

On the Cell in PS3, I wouldn't be surprised if Sony changes the specs to 8 SPE's if in between now and the beginning of their manufacturing run they are able to improve their yields substantially. Likewise I don't think a move from 3.2 GHz up to 3.6 GHz is out of the question for them, depending on heat dissipation. I don't think either of the above are likley, but I do think they are both a possibility - Sony kind of likes to have that 'last minute' spec increase with their systems. All that aside I'm expecting revision 2 Cells in PS3.
 
Megadrive1988 said:
might they activate the 8th SPE, crank up the core clockspeed, use the 2nd revision of Cell which has 250M transistors, or all of the above?

IIRC, the EE was shown at ISSCC 1999 on a 250nm process and later the PS2 debuted on a 250nm EE and 180nm GS. So unless they do something similar for the RSX and drop it to 65nm for PS3's release, I doubt we'll see any revisions.

Also the CELL DD1 was a prototype for ISSCC 2005 and CELL DD2 with 250 mil trannies should make it into PS3. But around ~ 21 mil trannies of the 8th SPE would be de-activated.
 
Re: so PS3 will have ~534 million transistors + other questi

one said:
Megadrive1988 said:
might they activate the 8th SPE, crank up the core clockspeed, use the 2nd revision of Cell which has 250M transistors, or all of the above? (if anything)
DD2 is what PS3 is supposed to have as Suzuoki said that DD1 was unsatisfying in power consumption and they could iron out it just before the end of 2004 then the tapeout of DD2 followed. PPE was improved to be 2-issue and got more complicated, probably it's one of the reason the clockspeed of Cell in PS3 is at 3.2Ghz.


good, I hope you are right, One.

although I was assuming that PS3 as of now, was going to use the original Cell, not the DD2 revision, because in Sony's PS3 slides it showed the original 234M transistor Cell not the new 250M transistor DD2 revision. but obviously it makes complete sense to use the DD2 revision.
 
Jaws said:
So unless they do something similar for the RSX and drop it to 65nm for PS3's release, I doubt we'll see any revisions.
Well EE WAS clocked higher in PS2 then originally announced at ISSCC, and there was a bunch of small revisions and bugfixes done to it between May and September (don't know the exact time when it was locked down, but it wasn't until the DTL10000 came out that we got the final spec chips.
 
Fafalada said:
Jaws said:
So unless they do something similar for the RSX and drop it to 65nm for PS3's release, I doubt we'll see any revisions.
Well EE WAS clocked higher in PS2 then originally announced at ISSCC, and there was a bunch of small revisions and bugfixes done to it between May and September (don't know the exact time when it was locked down, but it wasn't until the DTL10000 came out that we got the final spec chips.

Yeah, the EE went from something like ~ 250 MHz to 300 MHz and it's transistor count went form 10.5 mil to around 13 mil with bug fixes, IIRC.

We've already had our CELL DD2 revision with a transistor increase but unfortunately, a downgrade from ISSCC with a clock to 3.2 GHz, and a diasabled SPE! So I'm hoping for an upgrade instead of a downgrade with the RSX! Well, 65nm seems a bit much... :(
 
Being able to stick more Cell chips in for future gen also will really make backwards compatibility very simple.
 
Back
Top