It is yet too early to tell, but Cell does not seem a suitable general-purpose-CPU to me. The Power5 core (or cores?) seem to be running at 1Ghz, whereas the APUs reach 4Ghz. The quoted Gflops can only be reached under very special circumstances - when code is suitable and optimized for it.
Basically, a lot of code would have to be rewritten from the ground as software cells - nothing developers like.
Furthermore, there is the price issue: Cell as presented occupies 221mm^2 - compared to the 65mm^2 a 970fx takes, this is quite a large and expensive chip. Sony and or IBM will have to swallow huge losses if they use it for a gaming console.
If Apple is going to use it, then in a different form: dual-core Power/PPC core with higher frequency and less APUs as a CoreImage/CoreVideo accellerated CPU.
The one thing that I'm not completely sure on with cell is also the one thing that I think is getting totally blown out of proportion by news writers, and that is the use of: "Multiple simultaneous operating systems including Linux." I do not believe they are referring to full scale Operating systems, as in Windows, or the Mac OS type. I think this is in reference to a scaled function centric OS. At least for the time being. I don't think you'll be seeing a processor like this running Mac OS, and windows at the same time any time soon. Even though I believe these are more function focused OS's being (very lightly) described (and I think it's being done intentionally to help sell company stock) they are still being what I am perceiving as emulated. I may be wrong on this point too, but the SPE cores can support multiple operating systems and programming models through the use of virtualization technologies. Isn't the term virtualization technologies just a glorified use of the term emulation?
Even if this were hardware emulation only running these operating systems and programming models, I think they are very scaled function focused routines.
Anyone else's thought's are, requested, and encouraged.
The PS2 already runs multiple operating systems, just not simultaneously...
1) the PlayStation2 OS (obviously a "scaled function centric" OS...)
2) whatever distro of Linux that Sony sells with it's Linux kit... (there is even a Linux cluster of PS2s out there somewhere, the URL eludes me at this time...)
It is yet too early to tell, but Cell does not seem a suitable general-purpose-CPU to me. The Power5 core (or cores?) seem to be running at 1Ghz, whereas the APUs reach 4Ghz. The quoted Gflops can only be reached under very special circumstances - when code is suitable and optimized for it.
Basically, a lot of code would have to be rewritten from the ground as software cells - nothing developers like.
Furthermore, there is the price issue: Cell as presented occupies 221mm^2 - compared to the 65mm^2 a 970fx takes, this is quite a large and expensive chip. Sony and or IBM will have to swallow huge losses if they use it for a gaming console.
If Apple is going to use it, then in a different form: dual-core Power/PPC core with higher frequency and less APUs as a CoreImage/CoreVideo accellerated CPU.
Sony is planning on fabbing these themselves. If they amortize the R&D directly as a PS3 expense for the chips they plan to put into the PS3 then it is an overall widget lifecycle cost comparison, not a per chip gain/loss, and over 5-7 years they come out with an overall big win.
It is yet too early to tell, but Cell does not seem a suitable general-purpose-CPU to me. The Power5 core (or cores?) seem to be running at 1Ghz, whereas the APUs reach 4Ghz. The quoted Gflops can only be reached under very special circumstances - when code is suitable and optimized for it.
It is not a Power5 core -- there might be some design elements in common, but it is a considerably different processor. And it is running at the same speed as the SPEs (i.e. 4+ GHz).
The quoted GFLOPS of any processor can only be reached under certain circumstances. It also happens that these circumstances are quite like the circumstances that allow the G4/G5 VMX units to get their peak performance. And Pentium4/Opteron peaks as well, for that matter. This is an important qualifier, but it is not unique to the Cell. The important point is that the payoff for arranging code this way on the Cell is tremendous.
Looks like IBM is finally going to use a low-k dielectric for the 90 nm fab. Combined with dual-liner strained silicon, a svelt Powerbook G5 will be very possible.
For Cell, it's not exactly a panacea. Depending on what the PPE architecture is like, it could be a worse performer at scalar ops than the G4 is. Also, 221 sq mm die running in excess of 2 GHz. It must be a concidence than no power numbers are bandied about for the entire chip.
100+ Watts for a 3 GHz chip? Can't wait to see the box that Sony will ship the thing in.
Apparently not quite that bad...
Quote:
Cell will probably consume around 30 watts of power, similar to the Emotion Engine processor in the PlayStation 2 console, said Peter Glaskowsky, a technical analyst with The Envisioneering Group. This is also similar to the power consumption of Intel's Pentium M processor.
It would seem that Cell technology could be used in the future, but not any time soon. The conclusion in part 2 of this Cell introduction has this to say about Cell and Apple:
Quote:
Finally, before signing off, I should clarify my earlier remarks to the effect that I don't think that Apple will use this CPU. I originally based this assessment on the fact that I knew that the SPUs would not use VMX/Altivec. However, the PPC core does have a VMX unit. Nonetheless, I expect this VMX to be very simple, and roughly comparable to the Altivec unit of the first G4. Everything on this processor is stripped down to the bare minimum, so don't expect a ton of VMX performance out of it, and definitely not anything comparable to the G5. Furthermore, any Altivec code written for the new G4 or G5 would have to be completely reoptimized due to inorder nature of the PPC core's issue.
So the short answer is, Apple's use of this chip is within the realm of concievability, but it's extremely unlikely in the short- and medium-term. Apple is just too heavily invested in Altivec, and this processor is going to be a relative weakling in that department. Sure, it'll pack a major SIMD punch, but that will not be a double-precision Alitvec-type punch.
It is exciting technology, but for now probably aimed at consumer electronics and I suspect one of the first areas will be to boost mobile phone/pda abilities... But that like most other discussion about Cell is pure chit-chat!
I seem to remember reading an online article showing IBM's recent patents and one of those related to a compiler that could optimize code for, I think it mentioned, multiple processes or threads or something. I'll look again, but this could be interesting times ahead.
This has nothing to do with Apple. There are several compelling reasons why Apple uses GCC. Most people don't know that the modified version of GCC Apple uses has various optimizations for ObjC and ObjC++, not to mention Apple will contribute ObjC++ to GCC once Tiger is out.
Color me impressed if it is indeed some 30 Watts at the clock rates (4+ GHz) talked about and for a 90 nm chip. But overall it doesn't make sense for a 234 million transistor chip clocked at 4+ GHz to consume 30 Watts in "typical" usage. I pretty much refuse to believe it unless it can be shown why it is so low.
This is the 90 nm Cell we're talking about or the 65 nm?
Color me impressed if it is indeed some 30 Watts at the clock rates (4+ GHz) talked about and for a 90 nm chip. But overall it doesn't make sense for a 234 million transistor chip clocked at 4+ GHz to consume 30 Watts in "typical" usage. I pretty much refuse to believe it unless it can be shown why it is so low.
This is the 90 nm Cell we're talking about or the 65 nm?
I presumed the final 65nm version.
Don't forget that this chip WILL have to ship in a box that can probably only be (at its biggest) about the same size as the original PS2. Console buyers do make judgements based on size (The Japanese Xbox controllers spring to mind).
[Edit: And if they are putting them in HDTVs anytime soon, they cannot run that hot]
This has nothing to do with Apple. There are several compelling reasons why Apple uses GCC. Most people don't know that the modified version of GCC Apple uses has various optimizations for ObjC and ObjC++, not to mention Apple will contribute ObjC++ to GCC once Tiger is out.
Thank you for the response, I still can't find the links to the article I read.
Another nagging thought keeps entering my brain. A few months ago some one on these boards was touting Apple's next generation lap top chip to be some agglomeration of 440 style chips using low low power. In looking at information on Cell, the SPE's(I think this is the right term) are actually somewhat independent specialized cpu's??? Is it even remotely possible that this person's posts about multiple 440's on die for a low power lap top chip had any validity whatsoever, or at least were spawned from the Cell??
With a lot of people including me dismissing his posts as anything from dubious to out and out BS, wouldn't it be a real kick in the pants if it was Cell he was describing and it does end up in an Apple laptop in some form.
Please forgive me if I'm totally wrong as I really have no clue about this technology.
Yes but what kind of workstation are we talking about? It seems to me that you should be able to kill on scientific computing, and calculations on one of these like no other, but that could still make it a very limited machine in short term. I don't think a full strength GUI OS like Mac OS, or Windows would run under it. Not to mention all the apps that go with them.
I found the thread, it was spawned by Nr9, called "PowerBook G5".
I found a response by Amorph interesting.
Quote:
Originally posted by Amorph 11-16-2003 11:00 PM _ _ _ _ _ _
_
OK, let's step back and look at this. If it helps, don't think of it as a PowerBook, just an engineering exercize.
By itself, the 440 series is nothing to write home about. 4 of them, however, with four AltiVec units and FPUs, offer some real potential. They're very low power. The purpose of the MCM design is to get really high throughput between the chips in the module, so bandwidth and latency within the MCMs should be very high and very low, respectively. This is critical for clustering / massive SMP.
Now, consider the independent source that offered that IBM and Apple were developing a portable CPU from the 300 series, not the 900 series. What's the 300 series? That's a good question, but if IBM's naming conventions hold, the chips in its family resemble the 400 series more than they resemble the 900 series.
Consider also that IBM is currently building a supercomputer out of PowerPC 440s that will utterly crush the current #1.
Consider Cell, the IBM project to build a platform on massively parallel solutions for the first time (massive parallelism is nothing new, but each implementation has been a custom job with custom code, not really a platform). The above-mentioned supercomputer will be made of 128 nodes, each with 1,024 CPUs.
Now, we're talking about 4 cores, 4 AltiVec units, previously unheard-of bandwidth and latency between CPUs (on MCMs) - all requiring maybe 8 watts? They can easily grow into a 64-bit variant, since the PowerPC spec is natively 64 bit.
This is a highly unconventional platform for a personal computer, but I feel compelled to point something out that I haven't in a while: The only remaining champion of the traditional personal computer architecture is Intel. Everyone else is taking advantage of technologies like HyperTransport that allow for workstation-like motherboard designs at PC prices. These architectures are built around lots of bandwidth, which deëmphasizes the need for one single powerful CPU. OS X can already do MP. Apple is already working on software for clustering. You can run this setup as a small cluster with wonderful gobs of bandwidth and negligible latency, and it will run very well indeed.
This, to me, is a far more interesting rumor than the 970 was - the 970 was exciting, and more than welcome, but it's fundamentally conventional: Big, fast, hot. This sort of 440-based board is one of the mammals scurrying around under the feet of the mighty dinosaurs.
As to whether the PowerBook is ready for this, that's an interesting question. Multiple low-power processors are better at doing multiple tasks at once than they are at doing one big task (unless of course that big task is split up into lots of little ones behind the scenes) - c.f. the dual-processor iPod. So, if you can imagine something that will need to run some fairly compute-intensive services regularly while maintaining a responsive interface and doing light-to-medium work, that would do nicely. (OS X is already moving this way, using the GPU to run QE, and that's only the beginning of what they could do.) The main obstacle here is the programming paradigm implied by this architecture, which most current software (and most current languages) are ill-suited to. This might have the raw theoretical power to replace a single fast CPU, but it might be a while before software that can exploit it is widespread.
reading through some of the thread, there are differences between what Nr9 was saying, but there are striking similarities.
another quote
Quote:
Originally posted by Nr9 11-17-2003 12:23 AM _ _ _ _ _
_
this computer doesnt need redesign of core
the 440 is modular
it has an auxiliary core port
this auxiliary core is bigger than the current 440 FPU and it has altivec
the whole motherboard is based on SOC
it is similar in design to Blue Gene Compute Card except it has reduced L3 cache and SO-DIMM slot for up to 2GB DDR
"the 440 is modular" confused with the SPE??
"this core is bigger than the current 440 FPU and it has altivec" the PPE?
Yes but what kind of workstation are we talking about? It seems to me that you should be able to kill on scientific computing, and calculations on one of these like no other, but that could still make it a very limited machine in short term. I don't think a full strength GUI OS like Mac OS, or Windows would run under it. Not to mention all the apps that go with them.
This PPC440 speculation is just nonsense. IBM has sold off those assets and this new core is an unrelated design which, if based on anything, is based on POWER5 technology. Superficial dispatch details aside, the execution units, lower power design, SMT, cache heirarchy, and long pipelines are all very similar to POWER5 not 440. And the SPEs in the Cell are completely new and not at all Power or PowerPC... yet they are far closer to the new Power core in nature than to the 440.
Quote:
Originally posted by onlooker
Yes but what kind of workstation are we talking about? It seems to me that you should be able to kill on scientific computing, and calculations on one of these like no other, but that could still make it a very limited machine in short term. I don't think a full strength GUI OS like Mac OS, or Windows would run under it. Not to mention all the apps that go with them.
Considering that its Power core is plenty fast enough for most non-media oriented tasks (at 4 GHz and >10x the memory bandwidth, it'll beat any existing G4 and probably many existing G5s), and on any scientific/media problems the Cell will kill anything out there, it ought to make a fine workstation. Any complaints about Quartz2D being slow ought to be completely gone if they ship a Cell-based Mac.
How this fits in with the clock rate "wall" that Nr9 was putting forward is interesting. I've held that Nr9 was too literal in stating that there is a hard limit -- what makes more sense is that we've hit a point where a given design can't be easily scaled but new designs can win by specializing. This Cell chip is an example of that: by going with heavily pipelined, modular but simple core designs they've managed to jack up the clock rate. IBM's 970 and Intel's Pentium4 are most likely limited by their complex decode and OoOE circuits, not to mention Intel's double rate ALUs. The Cell design throws away everything that is complex and stretches out the pipelines to the max. The result is a chip hugely optimized for bandwidth and throughput, but which doesn't significantly (if at all) improve on the performance of general code. Even at 4 GHz I doubt the Cell's Power core will best a 2 - 2.5 GHz 970FX on "normal" code. Look at what it and its SPEs do with properly optimized vector code, however, and it'll beat pretty much anything around today (or on the x86 / PPC roadmaps)... by a significant margin.
Comments
Basically, a lot of code would have to be rewritten from the ground as software cells - nothing developers like.
Furthermore, there is the price issue: Cell as presented occupies 221mm^2 - compared to the 65mm^2 a 970fx takes, this is quite a large and expensive chip. Sony and or IBM will have to swallow huge losses if they use it for a gaming console.
If Apple is going to use it, then in a different form: dual-core Power/PPC core with higher frequency and less APUs as a CoreImage/CoreVideo accellerated CPU.
Originally posted by onlooker
The one thing that I'm not completely sure on with cell is also the one thing that I think is getting totally blown out of proportion by news writers, and that is the use of: "Multiple simultaneous operating systems including Linux." I do not believe they are referring to full scale Operating systems, as in Windows, or the Mac OS type. I think this is in reference to a scaled function centric OS. At least for the time being. I don't think you'll be seeing a processor like this running Mac OS, and windows at the same time any time soon. Even though I believe these are more function focused OS's being (very lightly) described (and I think it's being done intentionally to help sell company stock) they are still being what I am perceiving as emulated. I may be wrong on this point too, but the SPE cores can support multiple operating systems and programming models through the use of virtualization technologies. Isn't the term virtualization technologies just a glorified use of the term emulation?
Even if this were hardware emulation only running these operating systems and programming models, I think they are very scaled function focused routines.
Anyone else's thought's are, requested, and encouraged.
The PS2 already runs multiple operating systems, just not simultaneously...
1) the PlayStation2 OS (obviously a "scaled function centric" OS...)
2) whatever distro of Linux that Sony sells with it's Linux kit... (there is even a Linux cluster of PS2s out there somewhere, the URL eludes me at this time...)
Originally posted by THT
. . . Also, 221 sq mm die running in excess of 2 GHz. It must be a concidence than no power numbers are bandied about for the entire chip.
When the Cell is made with a 65nm process, that die should be about 115 square mm.
Originally posted by Smircle
It is yet too early to tell, but Cell does not seem a suitable general-purpose-CPU to me. The Power5 core (or cores?) seem to be running at 1Ghz, whereas the APUs reach 4Ghz. The quoted Gflops can only be reached under very special circumstances - when code is suitable and optimized for it.
Basically, a lot of code would have to be rewritten from the ground as software cells - nothing developers like.
Furthermore, there is the price issue: Cell as presented occupies 221mm^2 - compared to the 65mm^2 a 970fx takes, this is quite a large and expensive chip. Sony and or IBM will have to swallow huge losses if they use it for a gaming console.
If Apple is going to use it, then in a different form: dual-core Power/PPC core with higher frequency and less APUs as a CoreImage/CoreVideo accellerated CPU.
Sony is planning on fabbing these themselves. If they amortize the R&D directly as a PS3 expense for the chips they plan to put into the PS3 then it is an overall widget lifecycle cost comparison, not a per chip gain/loss, and over 5-7 years they come out with an overall big win.
Originally posted by Smircle
It is yet too early to tell, but Cell does not seem a suitable general-purpose-CPU to me. The Power5 core (or cores?) seem to be running at 1Ghz, whereas the APUs reach 4Ghz. The quoted Gflops can only be reached under very special circumstances - when code is suitable and optimized for it.
It is not a Power5 core -- there might be some design elements in common, but it is a considerably different processor. And it is running at the same speed as the SPEs (i.e. 4+ GHz).
The quoted GFLOPS of any processor can only be reached under certain circumstances. It also happens that these circumstances are quite like the circumstances that allow the G4/G5 VMX units to get their peak performance. And Pentium4/Opteron peaks as well, for that matter. This is an important qualifier, but it is not unique to the Cell. The important point is that the payoff for arranging code this way on the Cell is tremendous.
Originally posted by Programmer
. The important point is that the payoff for arranging code this way on the Cell is tremendous.
That's why I believe that the Cell powermac is not for 2005. Taking advantage of the cell architecture is not a cakewalk.
I think that we will a dual ppc 970 first, and then the G6 will be a cell chip, but a different one thant the PS chip.
Originally posted by THT
Looks like IBM is finally going to use a low-k dielectric for the 90 nm fab. Combined with dual-liner strained silicon, a svelt Powerbook G5 will be very possible.
For Cell, it's not exactly a panacea. Depending on what the PPE architecture is like, it could be a worse performer at scalar ops than the G4 is. Also, 221 sq mm die running in excess of 2 GHz. It must be a concidence than no power numbers are bandied about for the entire chip.
100+ Watts for a 3 GHz chip? Can't wait to see the box that Sony will ship the thing in.
Apparently not quite that bad...
Cell will probably consume around 30 watts of power, similar to the Emotion Engine processor in the PlayStation 2 console, said Peter Glaskowsky, a technical analyst with The Envisioneering Group. This is also similar to the power consumption of Intel's Pentium M processor.
(From MacWorld UK.)
It would seem that Cell technology could be used in the future, but not any time soon. The conclusion in part 2 of this Cell introduction has this to say about Cell and Apple:
Finally, before signing off, I should clarify my earlier remarks to the effect that I don't think that Apple will use this CPU. I originally based this assessment on the fact that I knew that the SPUs would not use VMX/Altivec. However, the PPC core does have a VMX unit. Nonetheless, I expect this VMX to be very simple, and roughly comparable to the Altivec unit of the first G4. Everything on this processor is stripped down to the bare minimum, so don't expect a ton of VMX performance out of it, and definitely not anything comparable to the G5. Furthermore, any Altivec code written for the new G4 or G5 would have to be completely reoptimized due to inorder nature of the PPC core's issue.
So the short answer is, Apple's use of this chip is within the realm of concievability, but it's extremely unlikely in the short- and medium-term. Apple is just too heavily invested in Altivec, and this processor is going to be a relative weakling in that department. Sure, it'll pack a major SIMD punch, but that will not be a double-precision Alitvec-type punch.
It is exciting technology, but for now probably aimed at consumer electronics and I suspect one of the first areas will be to boost mobile phone/pda abilities... But that like most other discussion about Cell is pure chit-chat!
Originally posted by rickag
I seem to remember reading an online article showing IBM's recent patents and one of those related to a compiler that could optimize code for, I think it mentioned, multiple processes or threads or something. I'll look again, but this could be interesting times ahead.
This has nothing to do with Apple. There are several compelling reasons why Apple uses GCC. Most people don't know that the modified version of GCC Apple uses has various optimizations for ObjC and ObjC++, not to mention Apple will contribute ObjC++ to GCC once Tiger is out.
Originally posted by DaveLee
Apparently not quite that bad...
Color me impressed if it is indeed some 30 Watts at the clock rates (4+ GHz) talked about and for a 90 nm chip. But overall it doesn't make sense for a 234 million transistor chip clocked at 4+ GHz to consume 30 Watts in "typical" usage. I pretty much refuse to believe it unless it can be shown why it is so low.
This is the 90 nm Cell we're talking about or the 65 nm?
Originally posted by THT
Color me impressed if it is indeed some 30 Watts at the clock rates (4+ GHz) talked about and for a 90 nm chip. But overall it doesn't make sense for a 234 million transistor chip clocked at 4+ GHz to consume 30 Watts in "typical" usage. I pretty much refuse to believe it unless it can be shown why it is so low.
This is the 90 nm Cell we're talking about or the 65 nm?
I presumed the final 65nm version.
Don't forget that this chip WILL have to ship in a box that can probably only be (at its biggest) about the same size as the original PS2. Console buyers do make judgements based on size (The Japanese Xbox controllers spring to mind).
[Edit: And if they are putting them in HDTVs anytime soon, they cannot run that hot]
Originally posted by Smircle
It is yet too early to tell, but Cell does not seem a suitable general-purpose-CPU to me.
It's general purpose enough for IBM to plan a release of a Cell-based workstation later this year, however.
http://www.macsimumnews.com/index.php/archive/2690/
Originally posted by mdriftmeyer
This has nothing to do with Apple. There are several compelling reasons why Apple uses GCC. Most people don't know that the modified version of GCC Apple uses has various optimizations for ObjC and ObjC++, not to mention Apple will contribute ObjC++ to GCC once Tiger is out.
Thank you for the response, I still can't find the links to the article I read.
Another nagging thought keeps entering my brain. A few months ago some one on these boards was touting Apple's next generation lap top chip to be some agglomeration of 440 style chips using low low power. In looking at information on Cell, the SPE's(I think this is the right term) are actually somewhat independent specialized cpu's??? Is it even remotely possible that this person's posts about multiple 440's on die for a low power lap top chip had any validity whatsoever, or at least were spawned from the Cell??
With a lot of people including me dismissing his posts as anything from dubious to out and out BS, wouldn't it be a real kick in the pants if it was Cell he was describing and it does end up in an Apple laptop in some form.
Please forgive me if I'm totally wrong as I really have no clue about this technology.
Originally posted by Dave J
It's general purpose enough for IBM to plan a release of a Cell-based workstation later this year, however.
http://www.macsimumnews.com/index.php/archive/2690/
Yes but what kind of workstation are we talking about? It seems to me that you should be able to kill on scientific computing, and calculations on one of these like no other, but that could still make it a very limited machine in short term. I don't think a full strength GUI OS like Mac OS, or Windows would run under it. Not to mention all the apps that go with them.
Arstechnica has a first look
More in next days...
Originally posted by Smircle
It is yet too early to tell, but Cell does not seem a suitable general-purpose-CPU to me.
It's general purpose enough for IBM to plan a release of a Cell-based workstation later this year, however.
http://www.macsimumnews.com/index.php/archive/2690/
I found a response by Amorph interesting.
Originally posted by Amorph 11-16-2003 11:00 PM _ _ _ _ _ _
_
OK, let's step back and look at this. If it helps, don't think of it as a PowerBook, just an engineering exercize.
By itself, the 440 series is nothing to write home about. 4 of them, however, with four AltiVec units and FPUs, offer some real potential. They're very low power. The purpose of the MCM design is to get really high throughput between the chips in the module, so bandwidth and latency within the MCMs should be very high and very low, respectively. This is critical for clustering / massive SMP.
Now, consider the independent source that offered that IBM and Apple were developing a portable CPU from the 300 series, not the 900 series. What's the 300 series? That's a good question, but if IBM's naming conventions hold, the chips in its family resemble the 400 series more than they resemble the 900 series.
Consider also that IBM is currently building a supercomputer out of PowerPC 440s that will utterly crush the current #1.
Consider Cell, the IBM project to build a platform on massively parallel solutions for the first time (massive parallelism is nothing new, but each implementation has been a custom job with custom code, not really a platform). The above-mentioned supercomputer will be made of 128 nodes, each with 1,024 CPUs.
Now, we're talking about 4 cores, 4 AltiVec units, previously unheard-of bandwidth and latency between CPUs (on MCMs) - all requiring maybe 8 watts? They can easily grow into a 64-bit variant, since the PowerPC spec is natively 64 bit.
This is a highly unconventional platform for a personal computer, but I feel compelled to point something out that I haven't in a while: The only remaining champion of the traditional personal computer architecture is Intel. Everyone else is taking advantage of technologies like HyperTransport that allow for workstation-like motherboard designs at PC prices. These architectures are built around lots of bandwidth, which deëmphasizes the need for one single powerful CPU. OS X can already do MP. Apple is already working on software for clustering. You can run this setup as a small cluster with wonderful gobs of bandwidth and negligible latency, and it will run very well indeed.
This, to me, is a far more interesting rumor than the 970 was - the 970 was exciting, and more than welcome, but it's fundamentally conventional: Big, fast, hot. This sort of 440-based board is one of the mammals scurrying around under the feet of the mighty dinosaurs.
As to whether the PowerBook is ready for this, that's an interesting question. Multiple low-power processors are better at doing multiple tasks at once than they are at doing one big task (unless of course that big task is split up into lots of little ones behind the scenes) - c.f. the dual-processor iPod. So, if you can imagine something that will need to run some fairly compute-intensive services regularly while maintaining a responsive interface and doing light-to-medium work, that would do nicely. (OS X is already moving this way, using the GPU to run QE, and that's only the beginning of what they could do.) The main obstacle here is the programming paradigm implied by this architecture, which most current software (and most current languages) are ill-suited to. This might have the raw theoretical power to replace a single fast CPU, but it might be a while before software that can exploit it is widespread.
reading through some of the thread, there are differences between what Nr9 was saying, but there are striking similarities.
another quote
Originally posted by Nr9 11-17-2003 12:23 AM _ _ _ _ _
_
this computer doesnt need redesign of core
the 440 is modular
it has an auxiliary core port
this auxiliary core is bigger than the current 440 FPU and it has altivec
the whole motherboard is based on SOC
it is similar in design to Blue Gene Compute Card except it has reduced L3 cache and SO-DIMM slot for up to 2GB DDR
"the 440 is modular" confused with the SPE??
"this core is bigger than the current 440 FPU and it has altivec" the PPE?
Originally posted by Dave J
It's general purpose enough for IBM to plan a release of a Cell-based workstation later this year, however.
http://www.macsimumnews.com/index.php/archive/2690/
Yes but what kind of workstation are we talking about? It seems to me that you should be able to kill on scientific computing, and calculations on one of these like no other, but that could still make it a very limited machine in short term. I don't think a full strength GUI OS like Mac OS, or Windows would run under it. Not to mention all the apps that go with them.
Originally posted by onlooker
Yes but what kind of workstation are we talking about? It seems to me that you should be able to kill on scientific computing, and calculations on one of these like no other, but that could still make it a very limited machine in short term. I don't think a full strength GUI OS like Mac OS, or Windows would run under it. Not to mention all the apps that go with them.
Considering that its Power core is plenty fast enough for most non-media oriented tasks (at 4 GHz and >10x the memory bandwidth, it'll beat any existing G4 and probably many existing G5s), and on any scientific/media problems the Cell will kill anything out there, it ought to make a fine workstation. Any complaints about Quartz2D being slow ought to be completely gone if they ship a Cell-based Mac.
How this fits in with the clock rate "wall" that Nr9 was putting forward is interesting. I've held that Nr9 was too literal in stating that there is a hard limit -- what makes more sense is that we've hit a point where a given design can't be easily scaled but new designs can win by specializing. This Cell chip is an example of that: by going with heavily pipelined, modular but simple core designs they've managed to jack up the clock rate. IBM's 970 and Intel's Pentium4 are most likely limited by their complex decode and OoOE circuits, not to mention Intel's double rate ALUs. The Cell design throws away everything that is complex and stretches out the pipelines to the max. The result is a chip hugely optimized for bandwidth and throughput, but which doesn't significantly (if at all) improve on the performance of general code. Even at 4 GHz I doubt the Cell's Power core will best a 2 - 2.5 GHz 970FX on "normal" code. Look at what it and its SPEs do with properly optimized vector code, however, and it'll beat pretty much anything around today (or on the x86 / PPC roadmaps)... by a significant margin.