DeanoC do you know when you get the final PS3 Harware?
- Thread starter sklaar
- Start date
And that's the point I was talking about. Somebody has to build it, pay for it. I doubt somebody will build a 100PFLOPS Single Precision machine in the next 15 years. Building an expensive 600K processor machine for questionable uses within the next 15 years? There's no reason to.
What if all those Cells were connected over a network 
How fast does the connection need to be to make distributed computing really feasible, and would that be feasible sooner than building a single box? I think Sony said previously you need 100Mbit connections at the very least to start doing that..
I'd agree with Bill that on the AI side, software will be a much bigger problem than hardware, but having massive computational power may help speed up some approaches.
How fast does the connection need to be to make distributed computing really feasible, and would that be feasible sooner than building a single box? I think Sony said previously you need 100Mbit connections at the very least to start doing that..
I'd agree with Bill that on the AI side, software will be a much bigger problem than hardware, but having massive computational power may help speed up some approaches.
Last edited by a moderator:
PC-Engine said:
Exactly, and when you're talking of the sheer volume that's going to be produced by 2010 and that you only need a little over 0.6% of the total production volume allocated for just PlayStation3 to accomplish your immense goal of 100PFlop/sec... the ability to field such a system suddenly becomes alot more feasible that it previously was for entities which invest is such supercomputing systems due to the economies of scale introduced by the PlayStation Platform, wouldn't you say?
And that's all I'm saying. Heck, for you to disagree is asinine.
PC-Engine said:
Argh, get off this single-precision argument. When you're talking about using Cell in this facility, any basis for it's use is relying on it's economies of scale to lower the per-unit cost into a region in which it can use it's sheer preformance to surpass the even lower cost that a commodity x86 system has. When you're talking about 100PFlop/sec, even if you need to use another 10 clocks to achieve IEEE754 DP complience you're still talking 10PFlop/sec -- an enormous amount of computing power.
This argument bores me and is inflating the thread with more useless posts.
Last edited by a moderator:
Depends on the problems you're trying to solve and how fast you need to solve them. Also SP isn't going to be very useful for solving real problems. I don't think any scientific organization is going to rely on SP gaming consoles for real supercomputing problems even if 100Mb/s network connections were the norm.
Who said anything about using commodity x86 systems? Regardless you're assuming Intel, AMD, etc are not going to be boosting DP performance in future processors beyond the current norm. These commodity chips will give CELL a run for its money. What's the highest commodity DP performing chip we have today? How much does it cost to produce? How does CELL compare in cost and performance?
I can't manage 1 FLOP. Heck, my integer performance probably has a theoretical peak of 5 operations per second on a good day. My old ZX81 was smarter than me...
Are our brain single precision or double precision?
/sarcasm
Would it be safe to assume that the day we start forgetting what we have in our hard drive (cause it's so big and full of crap) it's the day we find out wha the "capacity" of the brain is? Cause if that's the case, then i'm stuck at less than 160GB. Keep finding out new stuff on my hard drive i didn't know it was there every day!
How did this thread become a *YAWN* PS3-Matrix-Petaflops-Singleprecision-Sony's-crap thread?
/sarcasm
Would it be safe to assume that the day we start forgetting what we have in our hard drive (cause it's so big and full of crap) it's the day we find out wha the "capacity" of the brain is? Cause if that's the case, then i'm stuck at less than 160GB. Keep finding out new stuff on my hard drive i didn't know it was there every day!
How did this thread become a *YAWN* PS3-Matrix-Petaflops-Singleprecision-Sony's-crap thread?
PC-Engine said:
What is the PS3 Cell's DP capability? Multiply that by...?
Consider also that in 10 years there may be 100m PS4s out there, in 15 years 100m PS5s and so on. And they may all be "backwards compatible" too (i.e. a PS4 Cell could share work with a PS3 cell, perhaps, and so on).
Of course, that's assuming the network infrastructure was there. Could a million PS4s be connected by 100Mbit connections in 10 years? A million PS3s would be ~200 SP Petaflops (how much DP? 20 Petaflops?), what would a million PS4s be..? Or could a million PS5s be connected in 15 years?
Of course, this is only looking at the hardware side. Obviously on the software side, coding such distributed applications would surely be very challenging and require advances too.
SPEs aren't IEEE compliant are they?
The lack of IEEE would be significant for real scientific computing.
Edit:
Never mind, the docs say you can hack in IEEE double precision compliance by detecting a non-compliant result and retrying it with a slower code path. Icky. I wonder how much performance that costs you? Single precision doesn't even support more than truncation as a rounding mode, so IEEE is probably out for single precision completely.
The lack of IEEE would be significant for real scientific computing.
Edit:
Never mind, the docs say you can hack in IEEE double precision compliance by detecting a non-compliant result and retrying it with a slower code path. Icky. I wonder how much performance that costs you? Single precision doesn't even support more than truncation as a rounding mode, so IEEE is probably out for single precision completely.
Last edited by a moderator:
aaronspink
Veteran
Titanio said:
Picking this up here...
The PS3/CELL lacks a significant quantity of feature required to put 600k of them together and get any real usefull work done. The Blue Gene design from IBM dedicated over half its die to communication needs and even at that it just puts a small dent in the problem.
In 15 years time, we're looking at realistically 5-6 petaflops surrounded by a reaonable communication infrastructure. Unless there is a significant breakthrough in parallel programming, the useful performance achievable will be in the range of 1 petaflop. For a lot of the truely hard problems, it will be less.
Aaron Spink
speaking for myself inc.
Answering the original question...
Yes, yes I do... well.. actually... maybe, actually there a small chance of nuclear armegeddon so I better say No, no I don't, just to be safe.
Hopefully that clears everything up and doesn't break any NDAs...
Yes, yes I do... well.. actually... maybe, actually there a small chance of nuclear armegeddon so I better say No, no I don't, just to be safe.
Hopefully that clears everything up and doesn't break any NDAs...
DeanoC said:
Very clear!! We're all full of clearness now!
I can't believe something as benign (and useless) as when you think you might be getting final dev kits would be under NDA. That, is quite frustrating...
Can you answer this then? (seems neutral enough - and this goes out to all the PS3 devs!)
Will final dev kits be received within the rough time frame as Sony's previous *public* announcement indicated?
Can you answer this then? (seems neutral enough - and this goes out to all the PS3 devs!)
Will final dev kits be received within the rough time frame as Sony's previous *public* announcement indicated?
I think the NDA is basically that you can't talk about ANYTHING. That saves technicalities being argued on the event of a disclosure of information. As stated before, they can only talk a bit on information that has been given publically. If Sony haven't said it already, even if it's something benign like the colour of the box or the smell of the packaging, any dev talking such things will be tart-and-feathered, keel-hauled, hung, drawn and quartered, and given a good ticking off.
The best we've got is final kits are supposed to appear in December. I'm guessing the only confirmation we'll get on this is industry press like GI.biz. Don't expect any of our inside agents to actually provide any useful material!
The best we've got is final kits are supposed to appear in December. I'm guessing the only confirmation we'll get on this is industry press like GI.biz. Don't expect any of our inside agents to actually provide any useful material!
The arrival of final dev kits would seem to have implications about a console's launch timing. And this is not such a benign thing, is it?xbdestroya said:I can't believe something as benign (and useless) as when you think you might be getting final dev kits would be under NDA. That, is quite frustrating...
Can you answer this then? (seems neutral enough - and this goes out to all the PS3 devs!)
Will final dev kits be received within the rough time frame as Sony's previous *public* announcement indicated?
.Sis
Similar threads
