CELL will be there in PS3!

[PC-Engine]

2005 -> Cell with few cores < 1 TFLOPS
2007 -> Cell with more cores = 1 TFLOPS

 

[zidane1strife]

2005 -> Cell with few cores < 1 TFLOPS
2007 -> Cell with more cores = 1 TFLOPS

Well, that might be... but I think this is more likely...

2005-> Cell that exceeds 1Tflops... likely 3-6 Tflops perf(but I hope it's 13 Tflops.)

2007-> Cell that exceeds 20Tflops...

 

[Panajev2001a]

At what expense will it come for Sony to deliver the performance of roughly 1000 times over the PS2? If we assume that the chip will make for 1 TFLOP and a high performance rasterizer, what are the costs Sony will be looking at?

My guess is that it wont be much... and if they deliver what they've promised the market will GROW... The profits will be huge since the market will LIKELY GROW... we could be talking more than half a billion consoles(combined aka all systems) sold by the end of the next cycle...

Anyway at .1micron they probably could've gotten a 1Tflop peak chip, but at 65nm they will most likely have the space to squash most bottlenecks and achieve sustained perf. in excess of 1Tflop...

I do not think that the theoretical max would have been 1 TFLOPS using 100 nm technology ( I do not think it could have hit ~4 GHz ) and even if it did ( multi-chip ? ) it would have never had the e-DRAM, the wide busses, all the SRAM based Local Storage to support such FP power and you would not see even a small fraction of that in real-world scenarios...

Also the chip would have been way to massive...

It is still going to be a BIG chip using 65 nm ( they are also moving to 300 mm wafers to cut costs ) as they seem not to be able to wait for the moment in which they can shrink it to 45 nm...

Even with 90 nm a lot would have had to be butchered and would have killed Cell's performance... I cannot imagine how a 100 nm Broadband Engine would have been like...

Is it an accident that Cell plans have been pushed forward once Sony, IBM and Toshiba agreed to revise the plans to include up to 45-50 nm technology and Sony and Toshiba started the R&D on and finished the development of their 65 nm process ?

If 100 nm were still the target manufacturing process, you would not see half the optimism they are showing at this moment...

 

[Panajev2001a]

mid-late 2005, PlayStation 3 = ~1TFLOPS

2007, Cell implementations > 1 TFLOPS ( ~2-4 TFLOPS depending on the market )...

 

[zidane1strife]

I do not think that the theoretical max would have been 1 TFLOPS using 100 nm technology ( I do not think it could have hit ~4 GHz ) and even if it did ( multi-chip ? ) it would have never had the e-DRAM, the wide busses, all the SRAM based Local Storage to support such FP power and you would not see even a small fraction of that in real-world scenarios...

If 100 nm were still the target manufacturing process, you would not see half the optimism they are showing at this moment...

Well, I too think the cell would not have been possible on .1micron, but when they originally announced the EE3 as a 500M transistor .1micron processor, years ago before such a project was canned, I'm sure the engineers had concepts that would have viably hit peak 1Tflop even if it never achieved such numbers in the real world.

The fabled manufacturing process below .1micron and near the edge of 50nm will surely allow them to create far bigger processors... and allow them to fullfill their goals...

 

[Panajev2001a]

Well, I too think the cell would not have been possible on .1micron, but when they originally announced the EE3 as a 500M transistor .1micron processor, years ago before such a project was canned, I'm sure the engineers had concepts that would have viably hit peak 1Tflop even if it never achieved such numbers in the real world.

Maybe, but the logic designers probably slapped them upside the head and they researched better solutions ( hence the partnership with IBM and Toshiba on Cell ) and better manufacturing technologies...

 

[jvd]

Who is to say in 2007 they will have 1tflop. They may still be less than 1tflop. Since all that has flown around was .65nm 4ghz 1tflop cell. So even if it comes out in 2007 they will still have only .65nm 4ghz 1 tflop cell . Mabye with some tweaks they can hit 5 or 6 ghz but how much will that boost the tflop rating 1.5tflops mabye 2tflops at the most ? If even that . Ithink people are getting thier heads way to high . Then if in 2007 sony only releases a 1tflop machine everyone that was saying oh more than 1tflop 3-6 tflops blah blah blah will then go and say well sony has allways said 1tflop so the delivered what is promised . That is what happened with the first ps2. That is happening with this ps . Sony says a few words and then the fans take it to the extremes .

 

[Panajev2001a]

I think they can release a 1 TFLOPS class system in mid-late 2005... Japanese launch...

 

[jvd]

I think they can release a 1 TFLOPS class system in mid-late 2005... Japanese launch...

On .9 ? They havent' started building thier plants for .65nm and many articles have stated they will start next fiscal year . Which if i'm not mistaken would be febuary 2004.

 

[Paul]

jvd, you talk like Tflops are nothing. Intel doesn't even have a 10GFLOP chip out yet, and they won't have close to a 1tflop chip out in 2010, let alone 2007.

 

[jvd]

jvd, you talk like Tflops are nothing. Intel doesn't even have a 10GFLOP chip out yet, and they won't have close to a 1tflop chip out in 2010, let alone 2007.

Paul. It is not i that talks like tflops aren't important. Its the people throwing around the multi tflops like they are childs play. Although in 2007 i believe 1tflop will not be a big deal . It will be alot but many other chips will have caught up or will be very close to it . Also the geforce fx is spec at 250tflops . We are talking 4 years after the fx. You can bet the gpu of the xbox 2 (if released only shortly before the ps3) will be close to 1tflop. We both know that sony puts more importance in the main cpu than gpu and ms vise versa .

 

[Panajev2001a]

I think they can release a 1 TFLOPS class system in mid-late 2005... Japanese launch...

On .9 ? They havent' started building thier plants for .65nm and many articles have stated they will start next fiscal year . Which if i'm not mistaken would be febuary 2004.

jvd, Sony and Toshiba already announced that they expect to start 65 nm production in March-April 2004 ( Toshiba's guys were confident about the March 2004 date )... this was December 2002 when they announced they completed their 65 nm manufacturing process...

I think they can get enough units for a March-May ( maybe even September ) Japanese launch...

 

[Panajev2001a]

jvd, you talk like Tflops are nothing. Intel doesn't even have a 10GFLOP chip out yet, and they won't have close to a 1tflop chip out in 2010, let alone 2007.

Paul. It is not i that talks like tflops aren't important. Its the people throwing around the multi tflops like they are childs play. Although in 2007 i believe 1tflop will not be a big deal . It will be alot but many other chips will have caught up or will be very close to it . Also the geforce fx is spec at 250tflops . We are talking 4 years after the fx. You can bet the gpu of the xbox 2 (if released only shortly before the ps3) will be close to 1tflop. We both know that sony puts more importance in the main cpu than gpu and ms vise versa .

GFX is rated at 250 nVFLOPS which by what I understand is the number of GFLOPS you would need to do similar real-time rendering with the same speed as the GFX using full FP math...

 

[Panajev2001a]

Remember that excepth the new 90 nm EE+GS chip the PlayStation 2 production has been moving to China and other countries and the fabs they will leave "behind" will probably be upgraded to 65 nm technology shortening the time it would take for Sony to start producing 65 nm chips...

 

[Paul]

Paul. It is not i that talks like tflops aren't important. Its the people throwing around the multi tflops like they are childs play. Although in 2007 i believe 1tflop will not be a big deal . It will be alot but many other chips will have caught up or will be very close to it . Also the geforce fx is spec at 250tflops . We are talking 4 years after the fx. You can bet the gpu of the xbox 2 (if released only shortly before the ps3) will be close to 1tflop. We both know that sony puts more importance in the main cpu than gpu and ms vise versa .

One person has an opinion that ps3 will hit multi teraflop, why not direct everything your saying to him instead of "us people". Because only one person says this, and his opinion is valid as far as i'm concerned. But IMO PS3 will not hit multi teraflop performance.

1tflop not a big deal? By 2010 intel won't have a 1tflop cpu. Also, you can't bring GPU's into this, because a GPU is different from a CPU and you know this. Also, as seen from the patent, the "visualizer" has Cell attributes, meaning it will pack a heftly flop rating. 1tflop CPU + 256gflops rasterizer vs Xbox2's lets say 800gflops GPU and 20gflop CPU.

All estimation of course, about the Xbox2.[/img]

 

[jvd]

Paul. It is not i that talks like tflops aren't important. Its the people throwing around the multi tflops like they are childs play. Although in 2007 i believe 1tflop will not be a big deal . It will be alot but many other chips will have caught up or will be very close to it . Also the geforce fx is spec at 250tflops . We are talking 4 years after the fx. You can bet the gpu of the xbox 2 (if released only shortly before the ps3) will be close to 1tflop. We both know that sony puts more importance in the main cpu than gpu and ms vise versa .

One person has an opinion that ps3 will hit multi teraflop, why not direct everything your saying to him instead of "us people". Because only one person says this, and his opinion is valid as far as i'm concerned. But IMO PS3 will not hit multi teraflop performance.

1tflop not a big deal? By 2010 intel won't have a 1tflop cpu. Also, you can't bring GPU's into this, because a GPU is different from a CPU and you know this. Also, as seen from the patent, the "visualizer" has Cell attributes, meaning it will pack a heftly flop rating. 1tflop CPU + 256gflops rasterizer vs Xbox2's lets say 800gflops GPU and 20gflop CPU.

All estimation of course, about the Xbox2.[/img]

See thats the problem. U wont let me bring gpu's into this. Yet for ms the gpu is more much more important than the cpu. Its the gpu of the xbox 2 that will be transforming the polygons. Its the gpu that will provide the fillrate , effects and 90% of the other things . The xbox 2 cpu will be doing physics and a.i I believe that say a 30gflop cpu ( Dunno good base though i think for 2007) would be fast enough for ai and physics. Then ms having a one tflop gpu will be able to display the same graphics as the ps3. Of course i think its the gamecube 2 that blows both of these systems away.

 

[Phil]

GFX is rated at 250 nVFLOPS which by what I understand is the number of GFLOPS you would need to do similar real-time rendering with the same speed as the GFX using full FP math...

Panajev, are those nVFLOPS in anyway comparable to "real" FLOPS specs? Are those messured numbers Nvidia is publishing or raw estimates?

 

[Phil]

i I believe that say a 30gflop cpu ( Dunno good base though i think for 2007) would be fast enough for ai and physics. Then ms having a one tflop gpu will be able to display the same graphics as the ps3. Of course i think its the gamecube 2 that blows both of these systems away.

I'm not sure how much floating point performance would be enough to compete in terms of AI and physics. I hope to see a drastic improvement in especially physics next generation - I am speaking insane amounts of physics here... frankly, I have no idea how much FLOPs you can burn doing the appropriate math, but I would think quite a bit...?

 

[Panajev2001a]

nVFLOPS should also considering texture filtering and similar things which use dedicated silicon to be performed... basically this is akin to say you need over 250 GFLOPS to render in Software what the GeForce FX can do...

 

[Panajev2001a]

A.I. should remain Integer heavvy and the APUs have each 4 FP Units and 4 Integer Units: 32 GFLOPS and 32 GOPS per APU...


In total, with the theretical maximum of 1 TFLOPS we would have a theoretical maximum of 1 TOPS as well