bbot said:
Actually, let's speculate : with 0.09 micron tech, 16 cores/chip + edram at 1 GHz.
Each core could do one "vector operation", that is 3 additions and 4 multiplys in one cycle. So, we have 16 * 7 * 1 GHz = 112 gigaflops.
Eight chips would get 8 * 112 gigaflops = 896 gigaflops. There is no doubt at all, that you couldn't build a system like this. Only question really is amount of edram you could put there. For example 32 MB needs something like 256 million transistors.
Of course this is not 5 teraflops, but it's still A LOT of processing power.
For example IBM has had for some time "Power" chips that have 170 million transistors with 1 GHz clock and that's with 0.18 micron tech... I think they have 2 "Power" cores and 8 MB of SRAM. (Which is different from edram I'm talking above).
marconelly!
Veteran
Why would they want to have already a year old performance in something that will be released three years from now? That doesn't make any sense.
V3 said:
I was referring to time based physics, not to doing per frame because obviously the movement of the player character can't be faster than the frame rate. Otherwise you'd have warp effects.
There's also no problem to precalculate the trajectory of projectiles based on time but you can't predict player actions / movements. That was what I was getting at.
marconelly!
Veteran
Yes... and it's basically several PS2s working in parallel. They made it last year. Time goes on, and their goals seems to be much higher than several times more power than PS2.
You don't actually perdict player actions, in real time system, it practically does it by reflex
You can't predict player's input, but you can predict the player's avatar's actions and movements, base on previous input and such. That's where statistic and probability should be put to good used.
But generally for such thing as cloth simulation you don't really need that kind of thing. Those prediction is probably more important for online games and such.
zidane1strife
Banned
I don't see what's so unrealistic about 5Tflops... if they could increase the EE perf by adding more transistors and increasing in speed while keeping the same trans/speed/perf ratio... they'd only have to.... 10M trans EE at 300Mhz.... 10Mtransx100 and 300Mhzx10... with 5yrs of engineering time i'm sure they'd be able to get around the problems scaling such architecture would cause....
I don't believe they'd take chances with cell if they thought it wouldn't be capable of delivering 1000perf... i believe the current cell design that's been shown at all this presentations is but a precursor to the one that'd be used in 2005 devices...
PS..... I still remember when ps2 was touted as 5M poly with DC level effects in many rumors/magazines/sites... at the end it was far more than that...
marconelly!
Veteran
No... I think they will made something more appropriate, according to a technology available at the time. Sheesh... do you really think anyone is so stupid to release something *five* years later that is just several times more powerful than the previous attempt? I think you are vastly underestimating things here.
marconelly! said:
IMO PS3 will have similar fillrate and bandwidth to that of the GSCube. CPU power might be more but I doubt it since the design will have to be locked down early for obvious reasons. In theory you can use the latest and greatest technology available, but in pratice you have to consider cost, lock down time, practicality, etc.
zidane1strife said:
How do you figure that? Other than blind faith that is ... disregarding the costs of the final product, 100x exceeds Moore's law by an order of magnitude, even with SOI such a design would almost certainly be using many 100's of Watts if it is just a scaled version of the EE.
Why not? Even if they manage something like 200 GFLOPS they will still be by a large margin the fastest programmable device around ...
Just out of curiosity, let's say that MS and Nintendo's next console use the same type of CPU like the one in Xbox and GC and assuming that both companies will respect the usual 5 year lifespan for most gaming consoles, meaning GC2 and Xbox2 would come out in 2006, what kind of GFLOPS figures should we expect from those consoles?
I keep hearing those GFLOPS numbers that may be possible in the PS3 (200 G, 1T, 5T???) I'm just curious to see what the competition will have. I'm well aware that those consoles will probably have a more conservative architecture and that their GPUs will most likely, again, have way more features then the PS3's. But I'm just talking about GFLOPS here. So, any idea?
I keep hearing those GFLOPS numbers that may be possible in the PS3 (200 G, 1T, 5T???) I'm just curious to see what the competition will have. I'm well aware that those consoles will probably have a more conservative architecture and that their GPUs will most likely, again, have way more features then the PS3's. But I'm just talking about GFLOPS here. So, any idea?
Nintendo and MS aren't dim - I'm sure they read Kutigari's speeches the same as we do, and they know what Sony claims to be developing as well as we do.
So their systems will probably meet or exceed the PS3 in terms of useful power. Note that multi-proc systems like the PS2 or PS3 are typically extremely inefficient in their use of processing power -- it's likely that a uni-processor system with significantly less power will be able to match the PS3 in real-world games.
...assuming you can find a decent uni-processor system. MS doesn't have any problem with that, since Intel and AMD make very fast chips. But Nintendo's going to have to either switch to X86 or follow Sony in using multi-proc.
(I still expect Nintendo to exit the hardware business next generation, but that's a different issue.)
So their systems will probably meet or exceed the PS3 in terms of useful power. Note that multi-proc systems like the PS2 or PS3 are typically extremely inefficient in their use of processing power -- it's likely that a uni-processor system with significantly less power will be able to match the PS3 in real-world games.
...assuming you can find a decent uni-processor system. MS doesn't have any problem with that, since Intel and AMD make very fast chips. But Nintendo's going to have to either switch to X86 or follow Sony in using multi-proc.
(I still expect Nintendo to exit the hardware business next generation, but that's a different issue.)
I disagree. If Sony used a 'Uniprocessor' instead of the 'distributed' method of dual VU's, they'd never have attained the sheer power that they did in the EE for that timeframe. At that time, the nearest Intel processor was like a P2 300-400mhz. Even with the developer 'difficulties' (which seem to be due more to Sony's incompetance) the EE kills the uniprocessor appraoch.
Uniprocessors are not scaling their preformance anywhere near what they used to. While you can turn to stuff like Slipstreaming and other departures, the Cellular architecture will probobly elipse the Uniprocessor in dynamic media processing for it's integrated memory that can support many parrallel executions and high on chip bandwith. IBM did a study on ths and the results were very poor scaling for a unicrocessor.
I agree, with all due respect, unless they tag along with IBM again, I don't see how their going to hang with MS and Sony. Both of these two are making the consoles much more than just gaming - and have the associated infastructure, software, and hardware talent to make it work.
Either Nintendo goes with another lower powered console simlar to Gamecube (comparable to XBox) and focuses instead on developers and new, innovative gaming, or their'll exit. I'm thinking the later - but you never know.
Like to hear more opinions... Ben?
marconelly!
Veteran
Nintendo makes a lot more money selling their games on their consoles than what they would make selling on other consoles. For them, it's a simple logic, and there's no point changing it. Besides, they can always use available components when the time comes, just like MS is doing, and build new console. It's not like they were ever losing money on hardware.
