First off it is IBM; their process & design is what is causing the problems. Now that doesn't mean that they aren't trying to repair what is broken but they do have a ways to go.
It is not a limitation of the CPU in any shape or form. It is a problem with the process and the design. If IBM can get a handle on that, we may have stable 970FX or what ever they will be calling them next. IBM needs to address both the dynamic and static power used by the current process or move to something else.
Yes everyone has problems, I would imagine that even Nr9 has problems, but the wall IBM has been talking about is a wall they laid brick by brick themselves. There is a huge amount of work going into 90nm, not all of that engineering work is in the hands of IBM & Intel. Freescale is taking their approach, which if I understand them and their market will be rather conservative. But outside the maintstream markets you have corporations such as TI doing leading edge research at these geometries. Pop your head up a bit and look around IBM may have been the first to 90nm but it is not on what could be considered an innovative process.
Will we one day hit a physical limitation? Sure that is bound to happen but it won't be at 90nm and is very unlikely to happen at 65 nm. IBM's number one goal should be to develop a much lower power transitor element and a process to build that transistor. It is pretty clear that if IBM doesn't keep on the ball here they will be eclipsed by others in the market place. That may not concern IBM much but it certainly should be a concern to Apple.
Dave
Quote:
Originally posted by Nr9
wizard69, its not just IBM
EVERYONE has power problems. its the physical limitation of CPUs
A dual core chip is just that, two cores on one piece of silicon. Putting two cores on one piece of silicon does not make it easier to cool as your heat density is the same for the cores themselves. Other parts of the chip may actually see an increase in thermal load for example the bus interface. As to the single core processors and the heat out as Hz increase, yeah that is a problem with current processes. But that has been conquered many times in the history of this technology, there is room at 90nm to do it again.
Dual core or not the manufactures have no choice but to increase the operating frequency of the processors. The performance demands are such for leading edge software, that every trick in the book will be required to meet those needs in the future. This means dual cores, extended instruction sets and much faster processors. It is up to IBM to meet the challanges of tommorrows markets. If they can ever get the process corrected, we could very well see beyond 3GHz by the end of 2005.
Oh by the way AMD has been introducing faster processors for some time now. Unfortunately they hitched up with IBM for 90nm, it will be interesting to see how that pans out as other 90nm processes mature.
Dave
Quote:
Originally posted by Nr9
it solves it because it is twice the area hence easier to cool. Also, you don't have to drive the voltage up(which you have to do if you want to up the frequency instead) power is proportional to voltage ^2 so that matters a lot
you are wrong about faster processors coming. ask ibm, intel and amd
First off not everyone believes this. There is not a physical limit at 90nm it is a process limitation and nothing more. IBM will either evolve their process or they will become the modern day Motorola. It is IBM's game to play and it appears to be a case of attemtping to change the rules in the middle of play with the claims of a wall being hit. Frankly sound like more of a marketing scam than anything.
The reality is that research and process development is going strong at these sizes. Frankly I do not know why IBM would have declared a wall being hit as it is a very poor reflection on their engineers and scientist. IBM has no choice they either have to be successful with the 970 and get the 90nm process under control or kiss their fabrication business good by. So I very much expect that we will see IBM introduce CPU's on a revised 90nm process.
Dave
Quote:
Originally posted by Nr9
thats what they thought, but everyone who went to 90nm realized that they have hit the physical limit.
Chipworks discovered that IBM does not use all the fabrication techniques they said they'd use on the 970FX. It seems to Chipworks that IBM is largely using their 130 nm technology just shrunk to 90 nm with less than great results. My conclusion is that IBM has not mastered the intended 90 nm fabrication and it's the reason why we see hotter and slower 970FX processors than we we were promised. When/if IBM get's their fab straight I see no reason why they can't deiver on their 3 GHz promise.
Chipworks discovered that IBM does not use all the fabrication techniques they said they'd use on the 970FX. It seems to Chipworks that IBM is largely using their 130 nm technology just shrunk to 90 nm with less than great results. My conclusion is that IBM has not mastered the intended 90 nm fabrication and it's the reason why we see hotter and slower 970FX processors than we we were promised. When/if IBM get's their fab straight I see no reason why they can't deiver on their 3 GHz promise.
Last I'd heard IBM had made some rather poor material selections when it came to the 90 nm process. Turned out the materials weren't anywhere near as effective as they'd believed they would be and correcting that has been what has driven their 90 nm process back by a lot.
it solves it because it is twice the area hence easier to cool. Also, you don't have to drive the voltage up(which you have to do if you want to up the frequency instead) power is proportional to voltage ^2 so that matters a lot
Yes, power is goes up by the square of the voltage, but that is what die shrinks are for. Also, as someone mentioned already, twice the area doesn't mean half the power density. It's still the same power density. Eventually, research into thermoelectric materials, innovative chip packaging and better cooling solutions can all help dissipate higher power processors.
Quote:
you are wrong about faster processors coming. ask ibm, intel and amd
I will predict Intel will produce a 4 GHz Prescott (not much of a leap though) by 3Q 2005, AMD will be produce a 2.8 GHz in 2.7 GHz Athlon 64/Opteron by 3Q 2005, and IBM will produce a 2.8 GHz 970fx or dual core by 2Q 2005. I don't think clock rates will be standing still for an entire year.
guys, i never said dual core is lower density with respect to a single core running at the same frequency.
i said it is lower heat density with respect to a single core that provides the same throughput, ie, one that runs at twice the clock.
dont put words in my mouth
Also, don't believe it if you dont want to. but its not just IBM, 90nm is simply the end of road. Where are intel's great results with 90nm? For all I know, the pentium 4 570 and 580 are CANCELLED. This is why they are going dual core too. Similarily with IBM, the 3GHz 970 is CANCELLED.
Apple may yet receive 3GHZ 970FX chips with PowerTune if they can find a way to keep them cool while also keeping the PowerMacs fairly quiet.
Some speculate that this is the intent of the liquid cooling system in the 2.5GHz G5s which appears to be overengineered for that processor. (That is, Apple expects it to reduce fan noise in higher-clocked systems to an acceptable level as it is not absolutely necessary to cool a 2.5Ghz at low dbs.)
Some speculate that this is the intent of the liquid cooling system in the 2.5GHz G5s which appears to be overengineered for that processor. (That is, Apple expects it to reduce fan noise in higher-clocked systems to an acceptable level as it is not absolutely necessary to cool a 2.5Ghz at low dbs.)
the liquid cooling is definitely required for the 2.5
the liquid cooling is definitely required for the 2.5
I heard that the reason it needed liquid cooling was not because of the amount of heat that needed to be removed but the amount of heat per surface area. The die shink concentrated all the heat in a small surface area, Traditional methods could not remove the heat fast enough?
Yes, power is goes up by the square of the voltage, but that is what die shrinks are for. Also, as someone mentioned already, twice the area doesn't mean half the power density. It's still the same power density. Eventually, research into thermoelectric materials, innovative chip packaging and better cooling solutions can all help dissipate higher power processors.
Those approaches are all well and good but licking the problem at the root is a smarter course of action. New processes and geometries need to be icorporated to lower static power drastically and to address dynamic power.
IBM has done very little with respect to this with the 90nm 970. It is also why we are hearing about delays with the mobile chips. IBM is working on lowering power but that does not mean they have a solution yet. In the end static heat from a processor is nothing but waste.
Quote:
I will predict Intel will produce a 4 GHz Prescott (not much of a leap though) by 3Q 2005, AMD will be produce a 2.8 GHz in 2.7 GHz Athlon 64/Opteron by 3Q 2005, and IBM will produce a 2.8 GHz 970fx or dual core by 2Q 2005. I don't think clock rates will be standing still for an entire year.
I don't think clock rates will be standing still either. The question is will IBM keep up with the pack? Clock rate is very important to scaling certain sorts of performance. Multi processing serves to in crease performance in other ways, the industry needs to incorporate both advances to keep moving forward. So yes some in the industry hit a speed bump, the smart players are treating it just as that and are recovering and moving forward. I'm not sure what to think of the people who think they hit a wall.
Keep in mind the POWER5 was by and large simple enhancements to the POWER4. There's not really much that's radical or new in it.
Um... except faster bus, on-chip mem controller, SSOI instead of SOI, 1mb l2 cache... yah there is a lot that is different... oh yah... both are ppc and 130nm
I heard that the reason it needed liquid cooling was not because of the amount of heat that needed to be removed but the amount of heat per surface area. The die shink concentrated all the heat in a small surface area, Traditional methods could not remove the heat fast enough?
This is true. Apple has this documented in the white papers.
the liquid cooling is definitely required for the 2.5
Not from my understanding, although it is required if you don't want excessive fan noise.
Now I do not know people at Apple, but I think that noise has become part of the elegance issue for them. They've got a monopoly on intelligent design. They've got an intelligent, stable OS that is Unix based. They're trying to create machines that will not frustrate people and will see to work naturally. They've obviously been focused on quiet machines since the G5 hit. I remember all the complaints people had about the G4 PowerMacs and noise.
The G5s are very quiet - far more so than the AMD and Intel dual processor competition that people purchase. The iMac is also dead quiet. The laptops are the quietest I've (not) heard. It's part of the refinement issue for Apple. They want to offer speed while also offering refinement. Not just raw power, but something more.
I haven't taken a look at the cooling system for the 2.5 G5 as of yet, but I think Apple could come up with an active system to cool 3GHZ G5s - and as wizard has pointed out, some of IBM's issues have also created problems. I honestly don't know all the details about what's going on at 90nm as I am not a chip guru nor do I know tons about chip manufacture (I know enough to get by). My contacts are purely based on my software experience working at IBM - I had to know people who designed IBM's POWER and PowerPC chips for references when I worked on OS/2 several years ago (oh, and how I miss the Workplace Shell). I do know that IBM engineers feel they are making good progress on issues they've had at 90nm and anticipated issues at 65nm. the good news is that the 90nm issues has helped them refine the next process.
Comments
It is not a limitation of the CPU in any shape or form. It is a problem with the process and the design. If IBM can get a handle on that, we may have stable 970FX or what ever they will be calling them next. IBM needs to address both the dynamic and static power used by the current process or move to something else.
Yes everyone has problems, I would imagine that even Nr9 has problems, but the wall IBM has been talking about is a wall they laid brick by brick themselves. There is a huge amount of work going into 90nm, not all of that engineering work is in the hands of IBM & Intel. Freescale is taking their approach, which if I understand them and their market will be rather conservative. But outside the maintstream markets you have corporations such as TI doing leading edge research at these geometries. Pop your head up a bit and look around IBM may have been the first to 90nm but it is not on what could be considered an innovative process.
Will we one day hit a physical limitation? Sure that is bound to happen but it won't be at 90nm and is very unlikely to happen at 65 nm. IBM's number one goal should be to develop a much lower power transitor element and a process to build that transistor. It is pretty clear that if IBM doesn't keep on the ball here they will be eclipsed by others in the market place. That may not concern IBM much but it certainly should be a concern to Apple.
Dave
Originally posted by Nr9
wizard69, its not just IBM
EVERYONE has power problems. its the physical limitation of CPUs
Dual core or not the manufactures have no choice but to increase the operating frequency of the processors. The performance demands are such for leading edge software, that every trick in the book will be required to meet those needs in the future. This means dual cores, extended instruction sets and much faster processors. It is up to IBM to meet the challanges of tommorrows markets. If they can ever get the process corrected, we could very well see beyond 3GHz by the end of 2005.
Oh by the way AMD has been introducing faster processors for some time now. Unfortunately they hitched up with IBM for 90nm, it will be interesting to see how that pans out as other 90nm processes mature.
Dave
Originally posted by Nr9
it solves it because it is twice the area hence easier to cool. Also, you don't have to drive the voltage up(which you have to do if you want to up the frequency instead) power is proportional to voltage ^2 so that matters a lot
you are wrong about faster processors coming. ask ibm, intel and amd
The reality is that research and process development is going strong at these sizes. Frankly I do not know why IBM would have declared a wall being hit as it is a very poor reflection on their engineers and scientist. IBM has no choice they either have to be successful with the 970 and get the 90nm process under control or kiss their fabrication business good by. So I very much expect that we will see IBM introduce CPU's on a revised 90nm process.
Dave
Originally posted by Nr9
thats what they thought, but everyone who went to 90nm realized that they have hit the physical limit.
Originally posted by Nr9
you seem to be making this up.
Kettle meet pot.
Originally posted by Programmer
Hannibal's latest post on Ars has me wondering if there will be no POWER5-lite, simply enhancements to the 970's existing design.
Keep in mind the POWER5 was by and large simple enhancements to the POWER4. There's not really much that's radical or new in it.
Originally posted by Henriok
Chipworks discovered that IBM does not use all the fabrication techniques they said they'd use on the 970FX. It seems to Chipworks that IBM is largely using their 130 nm technology just shrunk to 90 nm with less than great results. My conclusion is that IBM has not mastered the intended 90 nm fabrication and it's the reason why we see hotter and slower 970FX processors than we we were promised. When/if IBM get's their fab straight I see no reason why they can't deiver on their 3 GHz promise.
Last I'd heard IBM had made some rather poor material selections when it came to the 90 nm process. Turned out the materials weren't anywhere near as effective as they'd believed they would be and correcting that has been what has driven their 90 nm process back by a lot.
Originally posted by Nr9
it solves it because it is twice the area hence easier to cool. Also, you don't have to drive the voltage up(which you have to do if you want to up the frequency instead) power is proportional to voltage ^2 so that matters a lot
Yes, power is goes up by the square of the voltage, but that is what die shrinks are for. Also, as someone mentioned already, twice the area doesn't mean half the power density. It's still the same power density. Eventually, research into thermoelectric materials, innovative chip packaging and better cooling solutions can all help dissipate higher power processors.
you are wrong about faster processors coming. ask ibm, intel and amd
I will predict Intel will produce a 4 GHz Prescott (not much of a leap though) by 3Q 2005, AMD will be produce a 2.8 GHz in 2.7 GHz Athlon 64/Opteron by 3Q 2005, and IBM will produce a 2.8 GHz 970fx or dual core by 2Q 2005. I don't think clock rates will be standing still for an entire year.
Originally posted by Telomar
Keep in mind the POWER5 was by and large simple enhancements to the POWER4. There's not really much that's radical or new in it.
You really have no idea what you are talking about.
http://www.findarticles.com/p/articl...8/ai_ziff49585
i said it is lower heat density with respect to a single core that provides the same throughput, ie, one that runs at twice the clock.
dont put words in my mouth
Also, don't believe it if you dont want to. but its not just IBM, 90nm is simply the end of road. Where are intel's great results with 90nm? For all I know, the pentium 4 570 and 580 are CANCELLED. This is why they are going dual core too. Similarily with IBM, the 3GHz 970 is CANCELLED.
Originally posted by Mr. MacPhisto
Apple may yet receive 3GHZ 970FX chips with PowerTune if they can find a way to keep them cool while also keeping the PowerMacs fairly quiet.
Some speculate that this is the intent of the liquid cooling system in the 2.5GHz G5s which appears to be overengineered for that processor. (That is, Apple expects it to reduce fan noise in higher-clocked systems to an acceptable level as it is not absolutely necessary to cool a 2.5Ghz at low dbs.)
Originally posted by Tomb of the Unknown
Some speculate that this is the intent of the liquid cooling system in the 2.5GHz G5s which appears to be overengineered for that processor. (That is, Apple expects it to reduce fan noise in higher-clocked systems to an acceptable level as it is not absolutely necessary to cool a 2.5Ghz at low dbs.)
the liquid cooling is definitely required for the 2.5
Originally posted by Nr9
the liquid cooling is definitely required for the 2.5
I heard that the reason it needed liquid cooling was not because of the amount of heat that needed to be removed but the amount of heat per surface area. The die shink concentrated all the heat in a small surface area, Traditional methods could not remove the heat fast enough?
Originally posted by THT
Yes, power is goes up by the square of the voltage, but that is what die shrinks are for. Also, as someone mentioned already, twice the area doesn't mean half the power density. It's still the same power density. Eventually, research into thermoelectric materials, innovative chip packaging and better cooling solutions can all help dissipate higher power processors.
Those approaches are all well and good but licking the problem at the root is a smarter course of action. New processes and geometries need to be icorporated to lower static power drastically and to address dynamic power.
IBM has done very little with respect to this with the 90nm 970. It is also why we are hearing about delays with the mobile chips. IBM is working on lowering power but that does not mean they have a solution yet. In the end static heat from a processor is nothing but waste.
I will predict Intel will produce a 4 GHz Prescott (not much of a leap though) by 3Q 2005, AMD will be produce a 2.8 GHz in 2.7 GHz Athlon 64/Opteron by 3Q 2005, and IBM will produce a 2.8 GHz 970fx or dual core by 2Q 2005. I don't think clock rates will be standing still for an entire year.
I don't think clock rates will be standing still either. The question is will IBM keep up with the pack? Clock rate is very important to scaling certain sorts of performance. Multi processing serves to in crease performance in other ways, the industry needs to incorporate both advances to keep moving forward. So yes some in the industry hit a speed bump, the smart players are treating it just as that and are recovering and moving forward. I'm not sure what to think of the people who think they hit a wall.
Dave
Originally posted by Nr9
the liquid cooling is definitely required for the 2.5
Prove it.
Originally posted by Telomar
Keep in mind the POWER5 was by and large simple enhancements to the POWER4. There's not really much that's radical or new in it.
Um... except faster bus, on-chip mem controller, SSOI instead of SOI, 1mb l2 cache... yah there is a lot that is different... oh yah... both are ppc and 130nm
Originally posted by tadunne
I heard that the reason it needed liquid cooling was not because of the amount of heat that needed to be removed but the amount of heat per surface area. The die shink concentrated all the heat in a small surface area, Traditional methods could not remove the heat fast enough?
This is true. Apple has this documented in the white papers.
Originally posted by Nr9
the liquid cooling is definitely required for the 2.5
Not from my understanding, although it is required if you don't want excessive fan noise.
Now I do not know people at Apple, but I think that noise has become part of the elegance issue for them. They've got a monopoly on intelligent design. They've got an intelligent, stable OS that is Unix based. They're trying to create machines that will not frustrate people and will see to work naturally. They've obviously been focused on quiet machines since the G5 hit. I remember all the complaints people had about the G4 PowerMacs and noise.
The G5s are very quiet - far more so than the AMD and Intel dual processor competition that people purchase. The iMac is also dead quiet. The laptops are the quietest I've (not) heard. It's part of the refinement issue for Apple. They want to offer speed while also offering refinement. Not just raw power, but something more.
I haven't taken a look at the cooling system for the 2.5 G5 as of yet, but I think Apple could come up with an active system to cool 3GHZ G5s - and as wizard has pointed out, some of IBM's issues have also created problems. I honestly don't know all the details about what's going on at 90nm as I am not a chip guru nor do I know tons about chip manufacture (I know enough to get by). My contacts are purely based on my software experience working at IBM - I had to know people who designed IBM's POWER and PowerPC chips for references when I worked on OS/2 several years ago (oh, and how I miss the Workplace Shell). I do know that IBM engineers feel they are making good progress on issues they've had at 90nm and anticipated issues at 65nm. the good news is that the 90nm issues has helped them refine the next process.
Originally posted by Mr. MacPhisto
the good news is that the 90nm issues has helped them refine the next process.
That is comforting