further indication that Xbox 360 has 512 MB
- Thread starter Megadrive1988
- Start date
pahcman said:
Bah. Space simulators run on 20 yrs old tech. PS3 needs super hardware exactly *because* it's a game console.
Besides, I doubt Sony would have invested all that money in cell if they weren't not going for the performance crown. They'd be better off using existing cpus if they were going for second-best.
If PS3 is going have only 4 working SPEs I will buy the XB2 instead, unless that too shouldn't live up to the rumours.
Titanio said:It's better to moderate expectations now
Oh and of course..this is all my speculation....ahem..
I just REALLY wouldn't expect 8.
I would be surprised if it did not have 8... I mean, the entire order and balance of this forum hinges on PC3 CELL >= 8 SPEs.
Maybe the CELL is being cut in half for a PPU??
nAo said:
Good point nAo. Same goes with the clock speed I guess... if a 3.5GHz CELL gets a 80% yeild and a 4.2GHz gets 50% and 4.8GHz gets 30%, on a large chip like the CELL you have to be kinda price conscious. But if PS3 CELL does not debut at 90nm, 65nm is right around the corner so maybe they will be pretty aggressive knowing a process shrink will help in yeilds and cost. I am sure you know more about what is going on than me though
Speaking of 65nm, what do people feel the likelihood is of PS3 shipping with 65nm Cells? I had just assumed a 90nm launch, but with Intel already set to produce, and AMD sampling successfully, I would imagine Sony couldn't be too far behind, given their aggressive timetables to begin with for 65nm implementation.
Sandwich said:
If it'll make it easier to understand, try substituting optimal for ideal. For a given problem, some solutions will be better than others. If that problem changes, some previously ideal (or optimal, or great, or good, or whatever word helps clarify the point) solutions may become less ideal (or optimal, or great, or good).
Cell was a lock for the PS3 long before the GPU was selected. Speculation based on a few sources suggests that a very different GPU was originally intended to be paired with Cell. If a hugely important part of the system changed, this may have had an impact on system balance based on the roles that other components were originally inteded to play.
I was simply asking an experienced developer for any opinions he may or may not have on this matter. I genuinely thought I was asking a relatively simple question (even if the answer is hard to get due to limited availability of knowledge and NDAs).
Sandwich said:
There's more to making a successful console than obtaining "the performance crown". Sony didn't just invest in Cell so they could have a fast PS3 CPU btw.
Sandwich said:
"Second best" by which criteria? Suitability for a role can't always be measured in gigaflops. Besides, who's to say that using an "off the shelf" CPU you could even acheive "second best"?
jimpo said:What the hell is going on here....everybody is boasting their own guesses and hunches like they would know for sure. You dumb-asses!
But seriously - all this speculation presented as real facts starts to get a bit frustrating.
Whoa, takin' life a little too seriously, eh Jimpo?
Besides, where is anyone presenting any of their speculation as fact? It's open speculation and educated guesses and no one is making it out to be anything more than it is.
Everything's been spelled out across various posts in this thread now..
Sony doesn't need a 8 SPE chip to maintain a significant advantage on the CPU side.
2 SPEs @ 3Ghz = 78Gflops
4 SPEs @ 3Ghz = 126Gflops
6 SPEs @ 3Ghz = 174Gflops
7 SPEs @ 3Ghz = 198Gflops ("peak")
2 SPEs @ 3.5Ghz = 91Gflops
4 SPEs @ 3.5Ghz = 147Gflops
6 SPEs @ 3.5Ghz = 203Gflops
7 SPEs @ 3.5Ghz = 231Gflops ("peak")
2 SPEs @ 4Ghz = 104Gflops
4 SPEs @ 4Ghz = 168Gflops
6 SPEs @ 4Ghz = 232Gflops
7 SPEs @ 4Ghz = 264Gflops ("peak")
Take your pick...(I've included a range to indicate possiblities from the worst case to the best case in terms of clockspeed)
If you take it that Xenon's CPU is packing 10 flops per cycle per core, and the leaked spec (3 cores @ 3Ghz) is for real, it should weigh in at around 90Gflops. So assuming that's true PS3's CPU is likely to weigh in at at least around 2x that, and probably more, in terms even of just useable power. Not enough?
Sony doesn't need a 8 SPE chip to maintain a significant advantage on the CPU side.
2 SPEs @ 3Ghz = 78Gflops
4 SPEs @ 3Ghz = 126Gflops
6 SPEs @ 3Ghz = 174Gflops
7 SPEs @ 3Ghz = 198Gflops ("peak")
2 SPEs @ 3.5Ghz = 91Gflops
4 SPEs @ 3.5Ghz = 147Gflops
6 SPEs @ 3.5Ghz = 203Gflops
7 SPEs @ 3.5Ghz = 231Gflops ("peak")
2 SPEs @ 4Ghz = 104Gflops
4 SPEs @ 4Ghz = 168Gflops
6 SPEs @ 4Ghz = 232Gflops
7 SPEs @ 4Ghz = 264Gflops ("peak")
Take your pick...(I've included a range to indicate possiblities from the worst case to the best case in terms of clockspeed)
If you take it that Xenon's CPU is packing 10 flops per cycle per core, and the leaked spec (3 cores @ 3Ghz) is for real, it should weigh in at around 90Gflops. So assuming that's true PS3's CPU is likely to weigh in at at least around 2x that, and probably more, in terms even of just useable power. Not enough?
BOOMEXPLODE
Regular
Remember for the longest time many expected a 65nm Broadband Engine in PS3 with one billion transistors and performance of one teraflop! Most of us are just guessing, and the ones that don't need to guess can't talk. My guess: the Cell shown at the ISSCC will be in PS3, because it seems to already be in production, and what else could they be using it in? We know atleast that they went to the trouble of designing and taping out the chip, so it's going to be used in something. It could be intended for servers/workstations, but that's not exactly a mass market application, hardly the place to begin recouping the expense of R&D.
BOOMEXPLODE said:
Well I agree with you in principle, but it was after ISSCC that Sony came out and said they were going to try and get that same chip inside of PS3 but were having difficulties with yields. It was after that that everyone started speculating on perhaps a lower SPE-count Cell. I myself am putting things at 50/25/25 percentage chances, respectively, for 8, 6, and 4 SPE configuration Cells.
Link?
Little protozoaes whispering sweet nothings to whoever that is isn't exactly an admission of problems fabbing from STI. That's what we call unsubstantiated rumour. From a game site.
BOOMEXPLODE
Regular
Still I can't believe they would go back to the drawing board and design a new 4 or 6 SPE version. That's non-trivial work that would require more time and money. If anything I can imagine them having to back off on the clockspeed to maybe 3GHz if they're having fab problems.
That kotaku guy is talking about an IBM fab which has no relation to the PS3, that's enough for me to conclude he has no idea or is simply joking by a hoax he made up (more likely)xbdestroya said:Brad Grenz said:
LOL, wow you sound pretty self-important. Well, if you have better leads stating that Cell fabbing is going a-ok at 8 SPE's, please, don't hesitate to lay it on me.
Megadrive1988
Veteran
BOOMEXPLODE said:
I guess I am a total fool then.
I was REALLY hoping for a monster PS3 CPU. beyond the oft-speculated / reported 32 APU / 1 TFLOP Broadband Engine.
I was hoping for the CPU described in Mercury News. 72 processors in total. 8 PowerPC cores (then PUs, now PPEs) and 64 APUs (now SPEs).
to most people, 1 teraflop was an awesome performance target for PS3 to achieve. to me, it was like a downgrade.
(from 1000 times the power of PS2.. where at first glance, you'd need 6.2 teraflops).
so you might imagine how I feel about a CPU with less than 8 SPEs. I feel like, "ahh Sony why don't you just dump the Cell because a very powerful conventional CPU does not seem that much weaker than Cell now".
then I start to think rationally again (lol) and realize that, well north of 100 GFLOPs is still pretty damn powerful.
BOOMEXPLODE said:
Well I'm not saying they would go back to the drawing board either, I'm just saying like some others have, that they could 'kill' dead SPE's on chip to increase yields. As a random example, let's say 50% of all Cell chips fabbed with 8 SPE's on the die have 8 functional SPE units. Now, if out of that same fab, 80% of those same Cell's have 6 functional SPE's per Cell, it would make a strong economic case for deciding to use a 6 SPE Cell as the configuration for the PS3.
And honestly as far as redesigns go anyway, I'm sure if they wanted to just go with 6 SPE or 4 SPE dies in the first place, it couldn't be too hard or expensive to implement; I mean, that sort of ease of scalability is supposed to be one of Cell's strong points to begin with, right?
Similar threads
