Well it's time I put my 2¢ into this thing. I fully believe Nr9 is wrong. All we have hit is a decent size wall that takes a little longer to climb over, it's happened before and will happen again. We are however starting to see the end of huge Ghz increases, I still think we can make it to at least 4 Ghz (referring it IBM) but more is not impossible. It's going to require some creative thinking and will probably take a few years to get there but it will eventually happen. In the mean time to keep things going up they will keep sticking cores on them and shipping them out. I'm really glad this is happening though it's finally going to get programmers to realize the benefits of multi threaded apps, Apple has been there a long time but the Windoze side of the industry needs to catch up in my opinion. (No offense to all the programmers in the audience, I hope to join you in a few years). Hopefully a replacement for transistors will be into production before we hit the wall of all walls, whether it's quantum computers or something else.
Finally the thread is talking about hardware again.
Nr9 is not my concern. I could care less if he worked at IBM, or Burger King. Either way what he said was preposterous. Nobody with an innovative mind could agree with a word of his pessimism. Everything evolves eventually.
Although I must say I don't think that 4GHz is anywhere near the immediate agenda. 3.2GHz is overly optimistic at this juncture IMO. Maybe by WWDC 2005, or What used to be MWNY 2005 we will something close to 3.2, but I'm not counting on it.
I do expect to see Dual Core not as an immediate, or next update (although I'd love to see it) but probably in the following update.
The G6 may be where we can actually see some serious improvements.
Nobody with an innovative mind could agree with a word of his pessimism. Everything evolves eventually.
The G6 may be where we can actually see some serious improvements.
Agreed ....
But don't you think the g6 is pretty far off . I mean there might not even be a g6... who knows. There might "probably" won't be a pentium 5... instead it is called what?
The Product is currently implmented the old way due to time constraints. Of course, now it's done it would probably have been faster and easier to build it from scratch, using our nice XML derived structures. Maybe for v2.
How much of a gain is the lack of cache in Cell's vector units? I suppose that SIMD often comes in streams, so data is just "passing through".
I'm not defending Nr9, but it's possible and quite believable. I can imagine the situation like this:
I'm neither defending nor supporting Nr9, I just think he is either misunderstanding something or being mislead.
Quote:
When IBM got 2.5GHz chips and realized that they're too hot (they may still be acceptably hot for IBM, but Steve said he's not buying chips any hotter), this instantly became a limitation #1. IBM engineers then had 2 big headaches: low yields and high power dissipation.
IBM has a problem but I see it as pushing to 90nm to fast and with too little innovation. IBM has pretty much demonstrated that they aren't ready to be a merchant supplier of processor chips. As far as the 970 and the 970FX neither chip could be considered an outstanding performer thermally.
With the current FX problems, people seem to forget there was a lot of SMOKE & MIRRORS with respect to the thermal performance of the 970. The question is simply this, is the thermal profile of the 970 industry leading with respect to computational performance. I would have to say that it is not so IBM clearly has room to grow in.
Quote:
They worked on these and managed to improve yields somehow. However, better yields are money today while better power consumption is money tomorrow. They tried and tried and tried and yes, they have 3+GHz chips right now, which are either too few for a customer like Apple, or too hot for showmen like Steve, or just too unreliable in the long term. IBM can tweak the process more, of course, to squeeze a couple of MHz without sacrificing the far-from-excellent yields, but...
IBM can tweak the process, that is pretty much assumed. To make the gains they really need though they need to change to a revised 90nm process.
Quote:
The dual-core version got so much closer during this struggle that it's probably more economically effective to launch PPC970MP now, than to delay it until they perfect the process so that you can stick a 3.5GHz PPC970FX into a pocket PC.
The MP is an interesting thought. Unless IBM can improve the thermal profile though the chip will be running extremely hot for a desktop PC. Think about possibly 140 watts or more. I suspect that the MP will be built on an improved 90nm process with that process eventually migrating back to the single core chips.
Quote:
This does not mean we'll never see a PowerPC running at more than 2.5GHz. This only means that we may see a dual-core before the painful transition to 90nm culminates in higher clock rates.
As long as people realize that dual core is a solution to another problem we should be OK.
Quote:
This only means that IBM engineers may find it easier and quicker to launch a dual-core 970 than to mass-produce any PPC at, say, 3GHz with their current technology.
The MP will come onto the market because it has to. That is there will be a very rapid migration to multiprocessing in the computing world due to one very important reality. That reality is that it is now possible to place more than one core on a die and remain economical. As I see it dual cores really have nothing to do with the current thermal and Hz limits. Multiprocessing is simply a way to make good use of the die space available to todays designers.
Quote:
And this does not mean IBM suddenly stops maturing the process in favour of dual-core designs, because this same process (if I'm not totally off) will help them make faster both single and dual cores until they switch to 65nm. And if they do switch to 65nm in the near future is an 'if'.
What do you think?
I don't really think there is a big delta between what I suspect will happen and what you suspect will happen. The MP can be seen as introducing 90nm rev2 and much better thermal profiles. I expect this technology to be quickly applied to a laptop processor as it will allow acceptable battery lifetimes.
I'm neither defending nor supporting Nr9, I just think he is either misunderstanding something or being mislead.
IBM has a problem but I see it as pushing to 90nm to fast and with too little innovation. IBM has pretty much demonstrated that they aren't ready to be a merchant supplier of processor chips. As far as the 970 and the 970FX neither chip could be considered an outstanding performer thermally.
With the current FX problems, people seem to forget there was a lot of SMOKE & MIRRORS with respect to the thermal performance of the 970. The question is simply this, is the thermal profile of the 970 industry leading with respect to computational performance. I would have to say that it is not so IBM clearly has room to grow in.
IBM can tweak the process, that is pretty much assumed. To make the gains they really need though they need to change to a revised 90nm process.
The MP is an interesting thought. Unless IBM can improve the thermal profile though the chip will be running extremely hot for a desktop PC. Think about possibly 140 watts or more. I suspect that the MP will be built on an improved 90nm process with that process eventually migrating back to the single core chips.
As long as people realize that dual core is a solution to another problem we should be OK.
The MP will come onto the market because it has to. That is there will be a very rapid migration to multiprocessing in the computing world due to one very important reality. That reality is that it is now possible to place more than one core on a die and remain economical. As I see it dual cores really have nothing to do with the current thermal and Hz limits. Multiprocessing is simply a way to make good use of the die space available to todays designers.
I don't really think there is a big delta between what I suspect will happen and what you suspect will happen. The MP can be seen as introducing 90nm rev2 and much better thermal profiles. I expect this technology to be quickly applied to a laptop processor as it will allow acceptable battery lifetimes.
I think you are misunderstanding something. dual cores have everything to do with current thermal and hz limits. this is the whole subject of dr. meyerson's keynote at the fall processor forum this week. your guesses at what will hapen with the chip industry are wrong. why don't you try to design a faster chip. im tellin u, its impossible.
Originally posted by Nr9 I think you are misunderstanding something. dual cores have everything to do with current thermal and hz limits. this is the whole subject of dr. meyerson's keynote at the fall processor forum this week. your guesses at what will hapen with the chip industry are wrong. why don't you try to design a faster chip. im tellin u, its impossible.
Look Nr9 your statement that it is impossible is simply to absolute to be respected. Either give it up or stop posting as your are wearing a little thin on credibility.
If you have half the experience in the industry you claim to have, you would realize that the current thermal limits are the result of the process that the chips are built on. To indicate that it is impossible to get better thermal performance is to imply that current process technology is static. Everyone here knows that is not the case.
Finally; why would I have to design a chip myself when many products are being delivered that have gotten considerable benefit from the move to 90nm? Should I just ignore the Hz increase TI, Fujitsu and others have gotten with their transistion to 90nm? The point is the shrink to 90nm has lead to much faster processors from vendors other than IBM.
You are also rather mistaken with respect to dual core technology. The industry would have gone this route long before if it where economical to do so. The issue was that for a long time there was plenty of room to get significnat payoff from modest growth in the complexity of a single core. Cores today have matured to the point where getting that payoff is much more difficult, you get better results from simply installing another core on the chip. After the transition to dual core the industry will still have to up clock frequency to continue to ramp performance at a given feature size. Either that or the addition of special function units to the cores will be implemented.
Many intelligent men have made mistakes with respect to their respective knowledge bases. Hopefully not all of the PPC industry will latch on to these negative thoughts. If this ends up being the case, we may very well be entering into another period of darkness with respect to the rest of the industry. It is like the Motorola problem all over again! Even Motorola was able to over come their 500MHz limit and road blocks after that. If IBM wants to throw in the towel then I suspect they will end up loosing much respect within the industry.
Look Nr9 your statement that it is impossible is simply to absolute to be respected. Either give it up or stop posting as your are wearing a little thin on credibility.
If you have half the experience in the industry you claim to have, you would realize that the current thermal limits are the result of the process that the chips are built on. To indicate that it is impossible to get better thermal performance is to imply that current process technology is static. Everyone here knows that is not the case.
Finally; why would I have to design a chip myself when many products are being delivered that have gotten considerable benefit from the move to 90nm? Should I just ignore the Hz increase TI, Fujitsu and others have gotten with their transistion to 90nm? The point is the shrink to 90nm has lead to much faster processors from vendors other than IBM.
You are also rather mistaken with respect to dual core technology. The industry would have gone this route long before if it where economical to do so. The issue was that for a long time there was plenty of room to get significnat payoff from modest growth in the complexity of a single core. Cores today have matured to the point where getting that payoff is much more difficult, you get better results from simply installing another core on the chip. After the transition to dual core the industry will still have to up clock frequency to continue to ramp performance at a given feature size. Either that or the addition of special function units to the cores will be implemented.
Many intelligent men have made mistakes with respect to their respective knowledge bases. Hopefully not all of the PPC industry will latch on to these negative thoughts. If this ends up being the case, we may very well be entering into another period of darkness with respect to the rest of the industry. It is like the Motorola problem all over again! Even Motorola was able to over come their 500MHz limit and road blocks after that. If IBM wants to throw in the towel then I suspect they will end up loosing much respect within the industry.
thanks
Dave
you're dumb. not only you ignore what i have to say, but you also ignore what dr. meyerson has to say. you dont even know anything about the current state of industry and you say im mistaken. when you get to 90nm, your gate oxide is about 6 atoms thick, and quality is very hard to control at 6 atoms thick. no matter what you are going to get a lot of leakge. process scaling is at a dead end, everyone in the industry knows thats the case.
It would be rather shortsighted for anybody to take the views of anyone man as absolute knowledge. At least the views of any man walking earth today. It is not that anyone here considers the "brain trust that is AI" to be a force to be admired, it is simply that there is more to innovation than intelligence of anyone person.
Up until a few days ago it was accepted that diamond was the hardest material known to man. Now it is known that we can fabricate even harder materials. Should the researchers who accomplished this have given up on their quest because some other respected person thought it was impossible?
Mr Meyerson may very well believe what he is saying, but that does not mean that his opinions need to be guidance for every other researcher out there. The whole problem with this discusion is that there is evidence today that other manufactures are doing much better with respect to 90nm and scaling frequency. It does bring into question just what is the motivation behind the declared wall -- is it an admission on IBM's part that they simply don't have the technology yet to do better?
The way people have been talking in this thread you would expect that we will never see a low power 970 for use in a portable. Does everybody here think that IBM gave up on low power devices also?
Thanks
Dave
Quote:
Originally posted by murk
Surely you aren't suggesting Bernard Meyerson is smarter than the brain trust that is AI!
The way people have been talking in this thread you would expect that we will never see a low power 970 for use in a portable. Does everybody here think that IBM gave up on low power devices also?
Who said that? I certainly haven't. Power dissapation at high clock rates is not the same thing as at low clock rates. The problems at 90nm seem to be the high clock rate parts, while the power profiles of low clock rates seem to be getting good results and have the potential to improve as the processes are refined. Its not clear that those results will map to significant high frequency improvements.
you're dumb. not only you ignore what i have to say, but you also ignore what dr. meyerson has to say. you dont even know anything about the current state of industry and you say im mistaken.
Who is ingnoring you? If that was the case I wouldn't have responded at all. What I'm saying is that the current state of the industry as you describe it, does not represent the industry as a whole.
Quote:
when you get to 90nm, your gate oxide is about 6 atoms thick, and quality is very hard to control at 6 atoms thick. no matter what you are going to get a lot of leakge. process scaling is at a dead end, everyone in the industry knows thats the case.
I really don't know what to say here, if this was the case then why is so much research effort being put into 90nm and why even bother with 65nm? It is a given that those issues will have to be controlled at 65nm or you won't have a process to begin with. Or atleast a process that results in a thermally manageable device.
The big question for you though, is why have some manufactures successfully scaled product? If it wasn't for this I'd give much more credibility to your statments.
you're dumb. not only you ignore what i have to say, but you also ignore what dr. meyerson has to say. you dont even know anything about the current state of industry and you say im mistaken. when you get to 90nm, your gate oxide is about 6 atoms thick, and quality is very hard to control at 6 atoms thick. no matter what you are going to get a lot of leakge. process scaling is at a dead end, everyone in the industry knows thats the case.
You haven't said anything like that at all Programmer nor would I expect you to. What I was trying to get at is that some people posting here seem to believe that their is no room for improvement at 90nm and that process scaling is dead.
First, I don't believe that anybody has given up on development at 90nm. That is with respect to anything related to the feature size be in manufacturablity, static power, dynamic power or any other parameter.
With respect to clock rates, as one scaling measurement, it does appear that some manufactures are doing fine with this. The issue seems to be IBM related more than anything. Remember that the 970, on 130nm, started out as a very hot processor for its performance point when introduced. The 90nm transisiton did lower the chips power point but it is still relatively hot, it is not surprising at all that only a modest %Hz increase was obtained due to thermal problems.
As far as the potential for improvement, that is exactly what I was getting at. I expect that IBM is working on low power devices through process improvments or changes to new proecess. The difference is that I also expect IBM to be doing the same thing with respect to ultimate clock rate. I just have a hard time accepting the position, of some on this board, that scaling (anyway you measure it) is dead at 90nm or that people are no longer developing technology for this feature size. As far as mapping improvements to anyone parameter, what is to preven multiple improvements targeted at different parameters?
It is just a little to early for me to accept that 2.5GHz is that best that can be had with the 970. Maybe that is the best that can be had with IBM's current 90nm process but that doesn't imply that there isn't the possibility for process improvement. I'm not sure anybody knows how far the current 970 can clock given the elimination of thermal problems. At some point the design of the 970 itself will limit clockrate but I don't think that is an issue at all now.
Dave
Quote:
Originally posted by Programmer
Who said that? I certainly haven't. Power dissapation at high clock rates is not the same thing as at low clock rates. The problems at 90nm seem to be the high clock rate parts, while the power profiles of low clock rates seem to be getting good results and have the potential to improve as the processes are refined. Its not clear that those results will map to significant high frequency improvements.
Who is ingnoring you? If that was the case I wouldn't have responded at all. What I'm saying is that the current state of the industry as you describe it, does not represent the industry as a whole.
I really don't know what to say here, if this was the case then why is so much research effort being put into 90nm and why even bother with 65nm? It is a given that those issues will have to be controlled at 65nm or you won't have a process to begin with. Or atleast a process that results in a thermally manageable device.
The big question for you though, is why have some manufactures successfully scaled product? If it wasn't for this I'd give much more credibility to your statments.
Dave
it does represent the industry as a whole.
because they want to have a less shitty 90nm process than a shitty one. manufacturer that successfully scale product aren't as high frequency or high power density
because they want to have a less shitty 90nm process than a shitty one. manufacturer that successfully scale product aren't as high frequency or high power density
scaling is dead. its a fact.
What professional talks like this when explaining something technical? So what are you going to do when IBM scales the 2.5 up? Can we ban your ip from this site?
the growth area is in laptops, intel, ibm moto are hitting a wall with speed , heat and battery life. how to max battery life. which setup would max battery life. a slower dual core or a more efficient single gore. my .9 ibook is fast enough for internet and word processing, i want more battery life on any laptop. battery tech is also at a wall, so which way to the dream machine.
What professional talks like this when explaining something technical? So what are you going to do when IBM scales the 2.5 up? Can we ban your ip from this site?
Comments
Nr9 is not my concern. I could care less if he worked at IBM, or Burger King. Either way what he said was preposterous. Nobody with an innovative mind could agree with a word of his pessimism. Everything evolves eventually.
Although I must say I don't think that 4GHz is anywhere near the immediate agenda. 3.2GHz is overly optimistic at this juncture IMO. Maybe by WWDC 2005, or What used to be MWNY 2005 we will something close to 3.2, but I'm not counting on it.
I do expect to see Dual Core not as an immediate, or next update (although I'd love to see it) but probably in the following update.
The G6 may be where we can actually see some serious improvements.
Originally posted by onlooker
Nobody with an innovative mind could agree with a word of his pessimism. Everything evolves eventually.
The G6 may be where we can actually see some serious improvements.
Agreed ....
But don't you think the g6 is pretty far off
How much of a gain is the lack of cache in Cell's vector units? I suppose that SIMD often comes in streams, so data is just "passing through".
Originally posted by Programmer
Ugh. Are you ignoring me, or just not getting it?
I'm just very literal, and the claim made in the original post in this thread (even the implied claim) is incorrect in my opinion.
Originally posted by costique
I'm not defending Nr9, but it's possible and quite believable. I can imagine the situation like this:
I'm neither defending nor supporting Nr9, I just think he is either misunderstanding something or being mislead.
When IBM got 2.5GHz chips and realized that they're too hot (they may still be acceptably hot for IBM, but Steve said he's not buying chips any hotter), this instantly became a limitation #1. IBM engineers then had 2 big headaches: low yields and high power dissipation.
IBM has a problem but I see it as pushing to 90nm to fast and with too little innovation. IBM has pretty much demonstrated that they aren't ready to be a merchant supplier of processor chips. As far as the 970 and the 970FX neither chip could be considered an outstanding performer thermally.
With the current FX problems, people seem to forget there was a lot of SMOKE & MIRRORS with respect to the thermal performance of the 970. The question is simply this, is the thermal profile of the 970 industry leading with respect to computational performance. I would have to say that it is not so IBM clearly has room to grow in.
They worked on these and managed to improve yields somehow. However, better yields are money today while better power consumption is money tomorrow. They tried and tried and tried and yes, they have 3+GHz chips right now, which are either too few for a customer like Apple, or too hot for showmen like Steve, or just too unreliable in the long term. IBM can tweak the process more, of course, to squeeze a couple of MHz without sacrificing the far-from-excellent yields, but...
IBM can tweak the process, that is pretty much assumed. To make the gains they really need though they need to change to a revised 90nm process.
The dual-core version got so much closer during this struggle that it's probably more economically effective to launch PPC970MP now, than to delay it until they perfect the process so that you can stick a 3.5GHz PPC970FX into a pocket PC.
The MP is an interesting thought. Unless IBM can improve the thermal profile though the chip will be running extremely hot for a desktop PC. Think about possibly 140 watts or more. I suspect that the MP will be built on an improved 90nm process with that process eventually migrating back to the single core chips.
This does not mean we'll never see a PowerPC running at more than 2.5GHz. This only means that we may see a dual-core before the painful transition to 90nm culminates in higher clock rates.
As long as people realize that dual core is a solution to another problem we should be OK.
This only means that IBM engineers may find it easier and quicker to launch a dual-core 970 than to mass-produce any PPC at, say, 3GHz with their current technology.
The MP will come onto the market because it has to. That is there will be a very rapid migration to multiprocessing in the computing world due to one very important reality. That reality is that it is now possible to place more than one core on a die and remain economical. As I see it dual cores really have nothing to do with the current thermal and Hz limits. Multiprocessing is simply a way to make good use of the die space available to todays designers.
And this does not mean IBM suddenly stops maturing the process in favour of dual-core designs, because this same process (if I'm not totally off) will help them make faster both single and dual cores until they switch to 65nm. And if they do switch to 65nm in the near future is an 'if'.
What do you think?
I don't really think there is a big delta between what I suspect will happen and what you suspect will happen. The MP can be seen as introducing 90nm rev2 and much better thermal profiles. I expect this technology to be quickly applied to a laptop processor as it will allow acceptable battery lifetimes.
Originally posted by wizard69
I'm neither defending nor supporting Nr9, I just think he is either misunderstanding something or being mislead.
IBM has a problem but I see it as pushing to 90nm to fast and with too little innovation. IBM has pretty much demonstrated that they aren't ready to be a merchant supplier of processor chips. As far as the 970 and the 970FX neither chip could be considered an outstanding performer thermally.
With the current FX problems, people seem to forget there was a lot of SMOKE & MIRRORS with respect to the thermal performance of the 970. The question is simply this, is the thermal profile of the 970 industry leading with respect to computational performance. I would have to say that it is not so IBM clearly has room to grow in.
IBM can tweak the process, that is pretty much assumed. To make the gains they really need though they need to change to a revised 90nm process.
The MP is an interesting thought. Unless IBM can improve the thermal profile though the chip will be running extremely hot for a desktop PC. Think about possibly 140 watts or more. I suspect that the MP will be built on an improved 90nm process with that process eventually migrating back to the single core chips.
As long as people realize that dual core is a solution to another problem we should be OK.
The MP will come onto the market because it has to. That is there will be a very rapid migration to multiprocessing in the computing world due to one very important reality. That reality is that it is now possible to place more than one core on a die and remain economical. As I see it dual cores really have nothing to do with the current thermal and Hz limits. Multiprocessing is simply a way to make good use of the die space available to todays designers.
I don't really think there is a big delta between what I suspect will happen and what you suspect will happen. The MP can be seen as introducing 90nm rev2 and much better thermal profiles. I expect this technology to be quickly applied to a laptop processor as it will allow acceptable battery lifetimes.
I think you are misunderstanding something. dual cores have everything to do with current thermal and hz limits. this is the whole subject of dr. meyerson's keynote at the fall processor forum this week. your guesses at what will hapen with the chip industry are wrong. why don't you try to design a faster chip. im tellin u, its impossible.
Originally posted by Nr9 I think you are misunderstanding something. dual cores have everything to do with current thermal and hz limits. this is the whole subject of dr. meyerson's keynote at the fall processor forum this week. your guesses at what will hapen with the chip industry are wrong. why don't you try to design a faster chip. im tellin u, its impossible.
Look Nr9 your statement that it is impossible is simply to absolute to be respected. Either give it up or stop posting as your are wearing a little thin on credibility.
If you have half the experience in the industry you claim to have, you would realize that the current thermal limits are the result of the process that the chips are built on. To indicate that it is impossible to get better thermal performance is to imply that current process technology is static. Everyone here knows that is not the case.
Finally; why would I have to design a chip myself when many products are being delivered that have gotten considerable benefit from the move to 90nm? Should I just ignore the Hz increase TI, Fujitsu and others have gotten with their transistion to 90nm? The point is the shrink to 90nm has lead to much faster processors from vendors other than IBM.
You are also rather mistaken with respect to dual core technology. The industry would have gone this route long before if it where economical to do so. The issue was that for a long time there was plenty of room to get significnat payoff from modest growth in the complexity of a single core. Cores today have matured to the point where getting that payoff is much more difficult, you get better results from simply installing another core on the chip. After the transition to dual core the industry will still have to up clock frequency to continue to ramp performance at a given feature size. Either that or the addition of special function units to the cores will be implemented.
Many intelligent men have made mistakes with respect to their respective knowledge bases. Hopefully not all of the PPC industry will latch on to these negative thoughts. If this ends up being the case, we may very well be entering into another period of darkness with respect to the rest of the industry. It is like the Motorola problem all over again! Even Motorola was able to over come their 500MHz limit and road blocks after that. If IBM wants to throw in the towel then I suspect they will end up loosing much respect within the industry.
thanks
Dave
Originally posted by wizard69
Look Nr9 your statement that it is impossible is simply to absolute to be respected. Either give it up or stop posting as your are wearing a little thin on credibility.
If you have half the experience in the industry you claim to have, you would realize that the current thermal limits are the result of the process that the chips are built on. To indicate that it is impossible to get better thermal performance is to imply that current process technology is static. Everyone here knows that is not the case.
Finally; why would I have to design a chip myself when many products are being delivered that have gotten considerable benefit from the move to 90nm? Should I just ignore the Hz increase TI, Fujitsu and others have gotten with their transistion to 90nm? The point is the shrink to 90nm has lead to much faster processors from vendors other than IBM.
You are also rather mistaken with respect to dual core technology. The industry would have gone this route long before if it where economical to do so. The issue was that for a long time there was plenty of room to get significnat payoff from modest growth in the complexity of a single core. Cores today have matured to the point where getting that payoff is much more difficult, you get better results from simply installing another core on the chip. After the transition to dual core the industry will still have to up clock frequency to continue to ramp performance at a given feature size. Either that or the addition of special function units to the cores will be implemented.
Many intelligent men have made mistakes with respect to their respective knowledge bases. Hopefully not all of the PPC industry will latch on to these negative thoughts. If this ends up being the case, we may very well be entering into another period of darkness with respect to the rest of the industry. It is like the Motorola problem all over again! Even Motorola was able to over come their 500MHz limit and road blocks after that. If IBM wants to throw in the towel then I suspect they will end up loosing much respect within the industry.
thanks
Dave
you're dumb. not only you ignore what i have to say, but you also ignore what dr. meyerson has to say. you dont even know anything about the current state of industry and you say im mistaken. when you get to 90nm, your gate oxide is about 6 atoms thick, and quality is very hard to control at 6 atoms thick. no matter what you are going to get a lot of leakge. process scaling is at a dead end, everyone in the industry knows thats the case.
Up until a few days ago it was accepted that diamond was the hardest material known to man. Now it is known that we can fabricate even harder materials. Should the researchers who accomplished this have given up on their quest because some other respected person thought it was impossible?
Mr Meyerson may very well believe what he is saying, but that does not mean that his opinions need to be guidance for every other researcher out there. The whole problem with this discusion is that there is evidence today that other manufactures are doing much better with respect to 90nm and scaling frequency. It does bring into question just what is the motivation behind the declared wall -- is it an admission on IBM's part that they simply don't have the technology yet to do better?
The way people have been talking in this thread you would expect that we will never see a low power 970 for use in a portable. Does everybody here think that IBM gave up on low power devices also?
Thanks
Dave
Originally posted by murk
Surely you aren't suggesting Bernard Meyerson is smarter than the brain trust that is AI!
Originally posted by wizard69
The way people have been talking in this thread you would expect that we will never see a low power 970 for use in a portable. Does everybody here think that IBM gave up on low power devices also?
Who said that? I certainly haven't. Power dissapation at high clock rates is not the same thing as at low clock rates. The problems at 90nm seem to be the high clock rate parts, while the power profiles of low clock rates seem to be getting good results and have the potential to improve as the processes are refined. Its not clear that those results will map to significant high frequency improvements.
Originally posted by Nr9
you're dumb. not only you ignore what i have to say, but you also ignore what dr. meyerson has to say. you dont even know anything about the current state of industry and you say im mistaken.
Who is ingnoring you? If that was the case I wouldn't have responded at all. What I'm saying is that the current state of the industry as you describe it, does not represent the industry as a whole.
when you get to 90nm, your gate oxide is about 6 atoms thick, and quality is very hard to control at 6 atoms thick. no matter what you are going to get a lot of leakge. process scaling is at a dead end, everyone in the industry knows thats the case.
I really don't know what to say here, if this was the case then why is so much research effort being put into 90nm and why even bother with 65nm? It is a given that those issues will have to be controlled at 65nm or you won't have a process to begin with. Or atleast a process that results in a thermally manageable device.
The big question for you though, is why have some manufactures successfully scaled product? If it wasn't for this I'd give much more credibility to your statments.
Dave
Originally posted by Nr9
you're dumb. not only you ignore what i have to say, but you also ignore what dr. meyerson has to say. you dont even know anything about the current state of industry and you say im mistaken. when you get to 90nm, your gate oxide is about 6 atoms thick, and quality is very hard to control at 6 atoms thick. no matter what you are going to get a lot of leakge. process scaling is at a dead end, everyone in the industry knows thats the case.
The convulsively attempt to obtain acceptance...
Nr9, the man who played bullshit-bingo
Where are the inside informations
First, I don't believe that anybody has given up on development at 90nm. That is with respect to anything related to the feature size be in manufacturablity, static power, dynamic power or any other parameter.
With respect to clock rates, as one scaling measurement, it does appear that some manufactures are doing fine with this. The issue seems to be IBM related more than anything. Remember that the 970, on 130nm, started out as a very hot processor for its performance point when introduced. The 90nm transisiton did lower the chips power point but it is still relatively hot, it is not surprising at all that only a modest %Hz increase was obtained due to thermal problems.
As far as the potential for improvement, that is exactly what I was getting at. I expect that IBM is working on low power devices through process improvments or changes to new proecess. The difference is that I also expect IBM to be doing the same thing with respect to ultimate clock rate. I just have a hard time accepting the position, of some on this board, that scaling (anyway you measure it) is dead at 90nm or that people are no longer developing technology for this feature size. As far as mapping improvements to anyone parameter, what is to preven multiple improvements targeted at different parameters?
It is just a little to early for me to accept that 2.5GHz is that best that can be had with the 970. Maybe that is the best that can be had with IBM's current 90nm process but that doesn't imply that there isn't the possibility for process improvement. I'm not sure anybody knows how far the current 970 can clock given the elimination of thermal problems. At some point the design of the 970 itself will limit clockrate but I don't think that is an issue at all now.
Dave
Originally posted by Programmer
Who said that? I certainly haven't. Power dissapation at high clock rates is not the same thing as at low clock rates. The problems at 90nm seem to be the high clock rate parts, while the power profiles of low clock rates seem to be getting good results and have the potential to improve as the processes are refined. Its not clear that those results will map to significant high frequency improvements.
Originally posted by wizard69
Who is ingnoring you? If that was the case I wouldn't have responded at all. What I'm saying is that the current state of the industry as you describe it, does not represent the industry as a whole.
I really don't know what to say here, if this was the case then why is so much research effort being put into 90nm and why even bother with 65nm? It is a given that those issues will have to be controlled at 65nm or you won't have a process to begin with. Or atleast a process that results in a thermally manageable device.
The big question for you though, is why have some manufactures successfully scaled product? If it wasn't for this I'd give much more credibility to your statments.
Dave
it does represent the industry as a whole.
because they want to have a less shitty 90nm process than a shitty one. manufacturer that successfully scale product aren't as high frequency or high power density
scaling is dead. its a fact.
Originally posted by Nr9
scaling is dead. its a fact.
This is simply not true.
Originally posted by Nr9
it does represent the industry as a whole.
because they want to have a less shitty 90nm process than a shitty one. manufacturer that successfully scale product aren't as high frequency or high power density
scaling is dead. its a fact.
What professional talks like this when explaining something technical? So what are you going to do when IBM scales the 2.5 up? Can we ban your ip from this site?
Originally posted by emig647
What professional talks like this when explaining something technical? So what are you going to do when IBM scales the 2.5 up? Can we ban your ip from this site?