Support NeoGAF

[jimbo]

Does anybody still remember articles like this?

http://www.macworld.co.uk/news/index.cfm?RSS&NewsID=10794

Because as of now all that looks more and more like hot air. It seems to me that by the time this chip is actually on the market it's really going to fit in quite nicely with Moore's law. Right or wrong....is it breaking it...or is it really still withing moore's law?

 

[Shompola]

I thoiught the amount of transistors in the cell chip goes hand in hand with the so called moores law.

 

[Elios83]

I don't know if it breaks it but as far as floating point performance is concerned it will smoke every other CPU for a looong time.

 

[akascream]

It will redifine moore's law, while taking it to the next level and blowing the roof off moore's law.

 

[sol5377]

Cell breaking Moore's Law...bull or true?

Truly Bull.

or not

 

[Tenacious-V]

I'm pretty sure that by the time it comes out, it will be about right with Moore's Law.

 

[border]

Moore's Law is supposed to be that every 12-18 months, the amount of transistors you can fit onto a wafer doubles.

Somehow it has been twisted and mutated into the perception that performance doubles every 18 months (or that clockspeed doubles every 18 months).

 

[jimbo]

"I don't know if it breaks it but as far as floating point performance is concerned it will smoke every other CPU for a looong time."

Yes but do you think it's going to more than double the floating point performance of a fpp-intensive chip that would have come out two years before the CELL, whenever the cell actually launches?

 

[gofreak]

Moore's Law is supposed to be that every 12-18 months, the amount of transistors you can fit onto a wafer doubles.

Somehow it has been twisted and mutated into the perception that performance doubles every 18 months (or that clockspeed doubles every 18 months).

Yeah, exactly. Anyone talking about Cell or any chip breaking Moore's law is probably thinking in terms of computational performance in a particular area, which isn't what Moore's Law is about.

 

[jimbo]

"Moore's Law is supposed to be that every 12-18 months, the amount of transistors you can fit onto a wafer doubles.

Somehow it has been twisted and mutated into the perception that performance doubles every 18 months (or that clockspeed doubles every 18 months)."

Yes I realize this, but Sony themselves are the ones that first used the catchphrase as power rather than transistors...so I'm going by what they said.

""We are working for the third-generation (PlayStation) with this very aggressive and crazy goal," he said. "Moore's Law is too slow for us. We can't wait 20 years to achieve a 1,000-fold increase in PlayStation performance." "

 

[Elios83]

"I don't know if it breaks it but as far as floating point performance is concerned it will smoke every other CPU for a looong time."

Yes but do you think it's going to more than double the floating point performance of a fpp-intensive chip that would have come out two years before the CELL, whenever the cell actually launches?

I'm not sure I've understood your question but are you asking if Cell will be more powerful in the floating point calculations two years from now or whenever it launches?
If that's your question,yes I can't see any other processor topping it in fp until there will be 8 cores general purpose CPUs.

 

[Mihail]

Only Sony hype can break Moore's Law. They are quite good at it.

 

[cyberheater]

I'm not sure I've understood your question but are you asking if Cell will be more powerful in the floating point calculations two years from now or whenever it launches?
If that's your question,yes I can't see any other processor topping it in fp until there will be 8 cores general purpose CPUs.

The cell doesn't have 8 core general purpose cpu.

 

[Elios83]

The cell doesn't have 8 core general purpose cpu.

And so?
I'm only talking about floating point calculations.

 

[jimbo]

"I'm not sure I've understood your question but are you asking if Cell will be more powerful in the floating point calculations two years from now or whenever it launches?"

No I am asking if you compare cell, whenever it actually launces, to a chip that's also fpp-heavy that came out 2 years before that time....if it will be more than double in performance....better yet... do youbelieve Cell not to be doubled in two years after it comes out?

 

[kaching]

Yes I realize this, but Sony themselves are the ones that first used the catchphrase as power rather than transistors...so I'm going by what they said.

""We are working for the third-generation (PlayStation) with this very aggressive and crazy goal," he said. "Moore's Law is too slow for us. We can't wait 20 years to achieve a 1,000-fold increase in PlayStation performance." "

They're not saying that Moore's Law equals power rather than transistors, they're full acknowledging that Moore's Law dictates a certain increase in transistors over a given period in time which in turn will translate into a certain increase in power. But strictly following the path that Moore's Law provides to greater performance doesn't net them the kind of performance gains they want to attain in a given period of time. That's what was being explained in this quote. So they began to look outside of the performance gains that can be netted merely by increase in transistor count.

 

[akascream]

They're not saying that Moore's Law equals power rather than transistors, they're full acknowledging that Moore's Law dictates a certain increase in transistors over a given period in time which in turn will translate into a certain increase in power.

So if that can be inferred from Sony's statements, why can't the same apply to statements about power in this thread?

 

[Elios83]

"I'm not sure I've understood your question but are you asking if Cell will be more powerful in the floating point calculations two years from now or whenever it launches?"

No I am asking if you compare cell, whenever it actually launces, to a chip that's also fpp-heavy that came out 2 years before that time....if it will be more than double in performance....better yet... do youbelieve Cell not to be doubled in two years after it comes out?

I think it won't even be reached in a 2 years time in fpp if we're looking at Intel and AMD's roadmaps.Of course if someone else in the industry designs a CPU specifically meant to be a floating point moster then yes,Cell is not made with magic so it will be beaten.
But at the moment just considering the general purpose CPUs from Intel&AMD it will be a long way just to match Cell.

 

[arhra]

Moore's Law is supposed to be that every 12-18 months, the amount of transistors you can fit onto a wafer doubles.

Somehow it has been twisted and mutated into the perception that performance doubles every 18 months (or that clockspeed doubles every 18 months).

Actually, it's a little bit more complex than that, but that idea was one of the central tenets of the original paper.

And yeah, Cell isn't all that special in terms of transistor count for the process node it'll be manufactured on. It's on the high-end of the integration curve, for sure, but it's not some insane outlier, there's lots of other stuff at around the same level of integration (the xenon cpu, high-end GPUs, etc). It gets the impressive performance numbers by being incredibly specialised, and making a lot of sacrifices (in a similar way to how GPUs are incredibly specialised for graphics functions, and can do an insane amount of work per-cycle, as long as you want to do the things they're good at, only not quite as extreme).

 

[kaching]

aka, I'm not sure I understand what you're asking.

 

[acidviper]

Moore's Law isn't that just an observation, thus making it a misnomer. Its not like the Law of Gravity where it has been proved ad infinitum..

 

[rainer516]

Gravity is theory bub, It's a law only as far as Newtonian physics is concerned. And in the same way Moore's law is simply conjecture, not a physical rule that all reality adheres to. Plus it's also great framework for tech company product release schedules.