How large will Cell be ?

[Megadrive1988]

what is the largest possible size the PS3 version of Cell
(the main CPU Cell) could be, given that they will be fabbed on .065 micron / 65 nm and quickly get moved down to .045 micron / 45 nm ?

 

[Paul]

Around 250-270mm squared.

Huge simply, and also will create alot of heat.

 

[cthellis42]

It depends what resolution you scan it it, and what kind of printer you use.

 

[DeadmeatGA]

...

280 mm2
Dual CELL core
166~200 million transistors
2~3 Ghz
256~384 GFLOPS/s

 

[Vince]

Re: ...

280 mm2
Dual CELL core
166~200 million transistors
2~3 Ghz
256~384 GFLOPS/s

Well good to see, on a lithography note, Sony will be doing with a 65nm process what Intel and AMD can do on a 130nm one. Usually it's the other way around with the dedicated IC pushing much harder, but oh well. Your opinion is just as valid and acceptable to me as anyone elses, including mine. We'll just see what's up in 2005.

Also, I bet PSP's SoC MPU comes close to the 100-150M tranistsor level. Which is why, for a SoC like this, using transistor count as a metric is fallicious. Area should be of concern IMHO.

 

[Panajev2001a]

200 MTransistors on a 65 nm process ???

Come on that is barely 2x of what ATI does in 150 nm with considerably smaller chip than 280 mm^2.

R300, in 150 nm, is ~200 mm^2.

Add extra 80 mm^2 and shrink from 150 nm to 65 nm and you end up with more than 200 Mransistors.

I would say something between 500-800 MTransistors is possible, but the Transistor metric is not as valuable when you pack MBs of SRAM and e-DRAM whose density is quite high and would lead to inflated Transistors numbers.

 

[DeadmeatGA]

...

200 MTransistors on a 65 nm process ???

That's what happens when you try to cram 72 FMACs, 72 MACs, two PPC cores, and a bunch of other logic circuit onto single die. Remember that the density of logic gate is 1/8th of SRAM gates...

 

[Vince]

Re: ...

That's what happens when you try to cram 72 FMACs, 72 MACs, two PPC cores, and a bunch of other logic circuit onto single die. Remember that the density of logic gate is 1/8th of SRAM gates...

The R300 core also has over 100M near 'pure' logic tranisistors on a 150nm process with a die size of 14.8*14.8mm. So please, just by posting a bunch of [wrong] nomenclature and numbers, doesn't mean you know a damn thing.

 

[Panajev2001a]

Like the R300 is not composed of a lot of logic gates...

Just shrinking the R300 using 65 nm technology and allowing the die to grow to 280 mm^2 you could do better than 200 MTransistors.

Also you are forgetting the nice SRAM SCE and Toshiba are putting inside: 128 KB of ultra dense ( some of the smallest if not the smallest SRAM and e-DRAM cells ) SRAM and then some ultra-dense e-DRAM.

Also, I thought you read IBM's patent too, but each APU does not have 4 FMACs and 4 MACs: given the approach they are taking they cna probably make 4 mixed FP/FX Units which share more logic than what 4 FP + 4 FX Units would share.

 

[Josiah]

for what it's worth, the 90nm "PS2 on a chip" is 53.5 million transistors and 86mm2 in size.

 

[DeadmeatGA]

...

for what it's worth, the 90nm "PS2 on a chip" is 53.5 million transistors and 86mm2 in size.

And the EE portion of that chip eats up 43 mm2 to lay foundation for only 13 million "logic" transistors. A textbook example of low logic gate density that Sony fans would not accept.

 

[V3]

what is the largest possible size the PS3 version of Cell
(the main CPU Cell) could be,

420 mm2, I don't think it will be larger than that.

 

[Panajev2001a]

Re: ...

for what it's worth, the 90nm "PS2 on a chip" is 53.5 million transistors and 86mm2 in size.

And the EE portion of that chip eats up 43 mm2 to lay foundation for only 13 million "logic" transistors. A textbook example of low logic gate density that Sony fans would not accept.

86 mm^2 / 2, right Deadmeat ?

Good ol' Deadmeat...

Edit: now I know how you derived your Transistor count for the CELL CPU of the PlayStation 3, division.. multiplication, multiply by 2 and add some padding. You always give SCE so much love and trust, don't you ?

 

[London Geezer]

silly me, and i thought Mathematics wasn't an OPINION.

i mean, isn't the whole point of Mathematics the fact that one problem always has one and only one solution?

if so, how come so many people here have different "opinions" on aspects that can be described by mathematics?

i mean, i'm not against one or the other, since i have no idea whatsoever how big and how many transistors will be in a chip that (1) doesnt exist and (2) will only come out in 2 years.

all i'm saying is that 2+2 (always) = 4 (and only 4).

sorry for being a geek...

 

[Dio]

What's the math on the power supply and heat dissipation of this?

 

[London Geezer]

What's the math on the power supply and heat dissipation of this?

Now THAT is what concerns me the most. This big fat chip will need some strong cooling. Possibly quiet too.

 

[Fafalada]

water cooled
86 m3
535 diodes
13 flop-minute

And it will run dual layered rewritable DVD+AOD discs!

 

[London Geezer]

water cooled
86 m3
535 diodes
13 flop-minute

And it will run dual layered rewritable DVD+AOD discs!

anyway, i was thinking about this "only one solution" thing, and it just dawned on me...

we all have different expectations and "numbers" because... NO ONE KNOWS A BLOODY THING ABOUT THE PS3 VERSION OF CELL...

that's why i really don't see the point in arguing and factioning NOW that no-one can come up with FACTS....
i mean, let's speculate and "exchange estimates"... but arguing? Mah...

 

[zidane1strife]

all i'm saying is that 2+2 (always) = 4 (and only 4).

Is it?

 

[London Geezer]

all i'm saying is that 2+2 (always) = 4 (and only 4).

Is it?

*gets the calculator out*


YES!!!


serious, see my point? u can only get so many transistors of a certain size in a certain area. one can't have different numbers. or at least not as different as 200 million (which would be pathetic by the way, Deadmeat, since we have chips on .15 and .13 TODAY with that many transistors) and 800 million (figure i am not sure about to be honest)...

really, i don't espect PS3 Cell to have 800M transistors (but if it does, GOOD), but also expecting 200M on .065 in 2 years time on an area that big is just "being blind"