What is biting the bullet? Biting the bullet is recognising that CPUs can't keep getting hotter and using more power indefinitely, and finding a new way. That new way is increasing-parallel computing using small, extensible and fast chips like the 440 and the Crusoe.
No, that would be the old way. Neither the 440 or the Crusoe are new.
Now if you were talking about something like a chip capable of reversable computations, I'd say that would be something new. But 4 440's on an MCM?
No, that would be the old way. Neither the 440 or the Crusoe are new.
Now if you were talking about something like a chip capable of reversable computations, I'd say that would be something new. But 4 440's on an MCM?
So Cell is just the old way?
That's an honest question. Because what this all is is the adoption of the architecture that IBM has quite forthrightly said is the future, and IBM is building it around the 440 family, literally.
No, it's not an entirely new paradigm, like reversible ternary logic, but as a mainstream architecture it's radical enough to be called new, and feasible enough not to be a pipe dream.
At least as far as Apple's concerned (i.e., in terms of personal computer architecture), the old way involves a single CPU that does as much of the heavy lifting as possible, and Apple's already moving away from that.
Tomb, this is just beginning of mac's cell based future.
Well, it's nice that you think so, but the 440 has nothing, repeat, nothing to do with the cell architecture that IBM, Sony and Toshiba are developing. Which by the way, would be really new if they succeed. (And they may not, there is always a certain amount of risk in these kinds of joint ventures.)
Quote:
they can implement MPI early, and mac developer will also kno that future desktop will also have it
You apparently don't know what MPI is.
Quote:
each APU has two FPU and one VMX. I think it would be obvious to you with the 11.2GFlop DP performance total.
Sorry, missed that. So each 440 will have a 440 FPU2 and a VMX unit? Heh. So much for your low power implementation.
I still have problems with this whole concept. No, sorry, I don't think so. Your calculation is based on a 700 Mhz clock which is at the upper limit for the 440GX. It's doubtful that the short pipelines in the 440 will clock that high, not for any kinds of realistic yields anyway.
That's an honest question. Because what this all is is the adoption of the architecture that IBM has quite forthrightly said is the future, and IBM is building it around the 440 family, literally.
I'm sorry? Where does this come from? I haven't heard that IBM is basing the cell architecture on their 440 line of chips.
Sorry, missed that. So each 440 will have a 440 FPU2 and a VMX unit? Heh. So much for your low power implementation. .
The transistor count would still be FAR less than, for example, the 970.
Quote:
Originally posted by Tomb of the Unknown
I still have problems with this whole concept. No, sorry, I don't think so. Your calculation is based on a 700 Mhz clock which is at the upper limit for the 440GX. It's doubtful that the short pipelines in the 440 will clock that high, not for any kinds of realistic yields anyway.
The 440 won't scale because it has a short pipeline? WTF? The G4+ has seven stages. The G3 has 4 and it goes to 1GHz or so. Not only is the pipeline not that big a deal, but IBM's chip fabs will surely help it scale.
"The supercomputer has 1,024 cell processors on 512 chips"
"IBM's PowerPC microprocessors will serve as basic building blocks. In contrast to microprocessors for personal computers, however, IBM believes that future machines will be powered by many processors that work together like bees in a hive."
I'm sorry? Where does this come from? I haven't heard that IBM is basing the cell architecture on their 440 line of chips.
The articles on Blue Gene/L, actually, have mentioned that it's part of their Cell project. Perhaps they're wrong?
I'm curious about what Cell is based on, if not the 400 series, since that's not at all a bad place to start given the nature of the project, and given IBM's fondness for reusing cores (c.f. the POWER series), and also given the 440's flexibility and extensibility. The whole point, as I understand it, is to use lots of low-power chips in a cluster arrangement. The claims I have seen made about Cell utterly discount the 970 or any of its kin...
"The supercomputer has 1,024 cell processors on 512 chips"
"IBM's PowerPC microprocessors will serve as basic building blocks. In contrast to microprocessors for personal computers, however, IBM believes that future machines will be powered by many processors that work together like bees in a hive."
I repeat. Where has IBM said they are basing the cell architecture on the PPC 440?
I repeat. Where has IBM said they are basing the cell architecture on the PPC 440?
Where have they said what they are basing it on? I'm more interested in positive information than claims that absence of evidence is evidence of absence.
The whole point, as I understand it, is to use lots of low-power chips in a cluster arrangement. The claims I have seen made about Cell utterly discount the 970 or any of its kin...
Absolutely. But it's also based on improved chip to chip (or core to core, depending on how you want to look at it) communications. I think cell will be very different from the 440. It may not even be PPC.
The idea behind cell is a kind of computational mesh, a fabric of cores if you will. The 440 is just another "core on a bus" SoC design.
And yes, 7 stages is short, depending on other features. The 440GX is only rated to 650 Mhz at the top end. IBM may well be able to ship a 700MHz version, since their roadmap calls for it to scale up to 800MHz, but then, the roadmap has the 750GX at up to 1.1 GHz and I don't see any 1.1GHz iBooks. (Yet.)
But again, that's the top end. Where do you go from there, even if you can get there in the first place?
Where have they said what they are basing it on? I'm more interested in positive information than claims that absence of evidence is evidence of absence.
Sony and Toshiba and IBM (STI) announced that they had teamed up to design an architecture for what is termed a system-on-a-chip (SoC) design. Code-named Cell, chips based on the architecture will be able to use ultra high-speed broadband connectivity to interoperate with one another as one complete system, similar to the way neural cells interoperate over the brain's network.
So, uh, we have SoC chips connected in a high-speed fabric to operate as a complete system. This is different from the technology under discussion in what way? And this rules the 440 (or a similar technology) out how?
The page is long on marketing and short on detail, but:
Quote:
IBM has an unmatched history and capability of building custom chips and believes the one-size-fits-all model of the PC does not apply in the embedded space; embedded applications will require a flexible architecture, like Cell.
Cell also brings together, for the first time, many leading-edge IBM chip technologies and circuit designs developed for its servers.
Again, this looks a lot like what we're talking about: The 440 is the basis for any number of custom chips; it's an embedded CPU, which Cell is explicitly using here; it's a flexible, extensible architecture; and not only is it bringing in leading edge technologies and designs, this thread adds more: L2 as a prefetch cache for L3 and the MCM.
Sorry, but after reading that, the 440 looks even more likely as a candidate, or at least as a core from which IBM will develop Cell.
Comments
they can implement MPI early, and mac developer will also kno that future desktop will also have it
PowerPC 970 is like 60 watts. good luck putting that in laptop
each APU has two FPU and one VMX. I think it would be obvious to you with the 11.2GFlop DP performance total.
Originally posted by Barto
You are dead wrong, my friend.
I don't think so.
What is biting the bullet? Biting the bullet is recognising that CPUs can't keep getting hotter and using more power indefinitely, and finding a new way. That new way is increasing-parallel computing using small, extensible and fast chips like the 440 and the Crusoe.
No, that would be the old way. Neither the 440 or the Crusoe are new.
Now if you were talking about something like a chip capable of reversable computations, I'd say that would be something new. But 4 440's on an MCM?
It is to laugh.
the new way is using many low power core
Originally posted by Tomb of the Unknown
No, that would be the old way. Neither the 440 or the Crusoe are new.
Now if you were talking about something like a chip capable of reversable computations, I'd say that would be something new. But 4 440's on an MCM?
So Cell is just the old way?
That's an honest question. Because what this all is is the adoption of the architecture that IBM has quite forthrightly said is the future, and IBM is building it around the 440 family, literally.
No, it's not an entirely new paradigm, like reversible ternary logic, but as a mainstream architecture it's radical enough to be called new, and feasible enough not to be a pipe dream.
At least as far as Apple's concerned (i.e., in terms of personal computer architecture), the old way involves a single CPU that does as much of the heavy lifting as possible, and Apple's already moving away from that.
Originally posted by Nr9
Tomb, this is just beginning of mac's cell based future.
Well, it's nice that you think so, but the 440 has nothing, repeat, nothing to do with the cell architecture that IBM, Sony and Toshiba are developing. Which by the way, would be really new if they succeed. (And they may not, there is always a certain amount of risk in these kinds of joint ventures.)
they can implement MPI early, and mac developer will also kno that future desktop will also have it
You apparently don't know what MPI is.
each APU has two FPU and one VMX. I think it would be obvious to you with the 11.2GFlop DP performance total.
Sorry, missed that. So each 440 will have a 440 FPU2 and a VMX unit? Heh. So much for your low power implementation.
I still have problems with this whole concept. No, sorry, I don't think so. Your calculation is based on a 700 Mhz clock which is at the upper limit for the 440GX. It's doubtful that the short pipelines in the 440 will clock that high, not for any kinds of realistic yields anyway.
Originally posted by Amorph
So Cell is just the old way?
That's an honest question. Because what this all is is the adoption of the architecture that IBM has quite forthrightly said is the future, and IBM is building it around the 440 family, literally.
I'm sorry? Where does this come from? I haven't heard that IBM is basing the cell architecture on their 440 line of chips.
Originally posted by Tomb of the Unknown
Sorry, missed that. So each 440 will have a 440 FPU2 and a VMX unit? Heh. So much for your low power implementation.
The transistor count would still be FAR less than, for example, the 970.
Originally posted by Tomb of the Unknown
I still have problems with this whole concept. No, sorry, I don't think so. Your calculation is based on a 700 Mhz clock which is at the upper limit for the 440GX. It's doubtful that the short pipelines in the 440 will clock that high, not for any kinds of realistic yields anyway.
The 440 won't scale because it has a short pipeline? WTF? The G4+ has seven stages. The G3 has 4 and it goes to 1GHz or so. Not only is the pipeline not that big a deal, but IBM's chip fabs will surely help it scale.
Barto
"The supercomputer has 1,024 cell processors on 512 chips"
"IBM's PowerPC microprocessors will serve as basic building blocks. In contrast to microprocessors for personal computers, however, IBM believes that future machines will be powered by many processors that work together like bees in a hive."
I know what MPI is. you will see. wait for next year.
440 FPU2 and VMX is not that high power.
you can have all your problems you want with this concept, but try not to turn it into a battlefront-esque discussion.
you will see next year.
Originally posted by Tomb of the Unknown
I'm sorry? Where does this come from? I haven't heard that IBM is basing the cell architecture on their 440 line of chips.
The articles on Blue Gene/L, actually, have mentioned that it's part of their Cell project. Perhaps they're wrong?
I'm curious about what Cell is based on, if not the 400 series, since that's not at all a bad place to start given the nature of the project, and given IBM's fondness for reusing cores (c.f. the POWER series), and also given the 440's flexibility and extensibility. The whole point, as I understand it, is to use lots of low-power chips in a cluster arrangement. The claims I have seen made about Cell utterly discount the 970 or any of its kin...
Originally posted by Barto
Link to Blue Gene/L.
"The supercomputer has 1,024 cell processors on 512 chips"
"IBM's PowerPC microprocessors will serve as basic building blocks. In contrast to microprocessors for personal computers, however, IBM believes that future machines will be powered by many processors that work together like bees in a hive."
I repeat. Where has IBM said they are basing the cell architecture on the PPC 440?
Originally posted by Tomb of the Unknown
I repeat. Where has IBM said they are basing the cell architecture on the PPC 440?
Where have they said what they are basing it on? I'm more interested in positive information than claims that absence of evidence is evidence of absence.
Originally posted by Amorph
The whole point, as I understand it, is to use lots of low-power chips in a cluster arrangement. The claims I have seen made about Cell utterly discount the 970 or any of its kin...
Absolutely. But it's also based on improved chip to chip (or core to core, depending on how you want to look at it) communications. I think cell will be very different from the 440. It may not even be PPC.
The idea behind cell is a kind of computational mesh, a fabric of cores if you will. The 440 is just another "core on a bus" SoC design.
And yes, 7 stages is short, depending on other features. The 440GX is only rated to 650 Mhz at the top end. IBM may well be able to ship a 700MHz version, since their roadmap calls for it to scale up to 800MHz, but then, the roadmap has the 750GX at up to 1.1 GHz and I don't see any 1.1GHz iBooks. (Yet.)
But again, that's the top end. Where do you go from there, even if you can get there in the first place?
Originally posted by Amorph
Where have they said what they are basing it on? I'm more interested in positive information than claims that absence of evidence is evidence of absence.
http://www-1.ibm.com/mediumbusiness/...rging/sti.html
Originally posted by Nr9
the Blue Gene is a demonstration of the viability of cell technlogy using 440 cores.
No, it most certainly is not. It is a demonstration of how to build a supercompter from embedded CPUs.
Cell computing is a concept. It is beginning to surface in prototypes of the Blue Gene using 440 cores.
Originally posted by Tomb of the Unknown
http://www-1.ibm.com/mediumbusiness/...rging/sti.html
From that:
Sony and Toshiba and IBM (STI) announced that they had teamed up to design an architecture for what is termed a system-on-a-chip (SoC) design. Code-named Cell, chips based on the architecture will be able to use ultra high-speed broadband connectivity to interoperate with one another as one complete system, similar to the way neural cells interoperate over the brain's network.
So, uh, we have SoC chips connected in a high-speed fabric to operate as a complete system. This is different from the technology under discussion in what way? And this rules the 440 (or a similar technology) out how?
The page is long on marketing and short on detail, but:
IBM has an unmatched history and capability of building custom chips and believes the one-size-fits-all model of the PC does not apply in the embedded space; embedded applications will require a flexible architecture, like Cell.
Cell also brings together, for the first time, many leading-edge IBM chip technologies and circuit designs developed for its servers.
Again, this looks a lot like what we're talking about: The 440 is the basis for any number of custom chips; it's an embedded CPU, which Cell is explicitly using here; it's a flexible, extensible architecture; and not only is it bringing in leading edge technologies and designs, this thread adds more: L2 as a prefetch cache for L3 and the MCM.
Sorry, but after reading that, the 440 looks even more likely as a candidate, or at least as a core from which IBM will develop Cell.
Originally posted by Nr9
Cell computing is a concept. It is beginning to surface in prototypes of the Blue Gene using 440 cores.
No, cellular computing is a concept. One that has been around for some time.
You are conflating cellular computing with the cell architecture that STI are researching, that's all.