. . . Also, if a manufacturing flaw kills part of a hyper-threaded processor then the whole processor is dead.
Single processor failures in a dual processor chip are common enough that IBM is shipping whole lines of machines based on them! Even if it is a fairly low percentage, this practice is only going to become more common as the number of cores grows. . .
</strong><hr></blockquote>
Do you think IBM will ignore hyper-threading, or is there a possibility that it is being worked on for the Power 5, as another post suggested? (I guess if they did do hyper-threading, it would be a reason to use it only with single core chips.) Or is it going to be a contest between single core, hyper-threaded chips, and multi core PPC chips?
The problem with a hyper-threaded core is that resources are shared so that the threads need to be in the same process, and those threads should be doing different things (i.e. integer/float, float/vector, integer/vector). Also, if a manufacturing flaw kills part of a hyper-threaded processor then the whole processor is dead.
Single processor failures in a dual processor chip are common enough that IBM is shipping whole lines of machines based on them! Even if it is a fairly low percentage, this practice is only going to become more common as the number of cores grows. If you have 16 cores and a flaw kills one (or two, or three, etc) of them but your design still allows the others to operate independently, then why not sell it as a less powerful chip? This is going to be as important in the future as selling different chips speeds is today.</strong><hr></blockquote>
Good point but I think that the reason there are so much single core POWER4 with defective 2nd cores is due to the newness of the process and processor. Once these kinks are worked out they should have a higher output of complete processors. Maybe IBM didn't want to dispose of otherwise good single core processors and used them in low end machines. When the POWER4 manufacturing process starts maturing to the point where 95% are whole dual core, it should become economically feasable to just put in the whole dual core POWER4.
<strong>Good point but I think that the reason there are so much single core POWER4 with defective 2nd cores is due to the newness of the process and processor. Once these kinks are worked out they should have a higher output of complete processors. Maybe IBM didn't want to dispose of otherwise good single core processors and used them in low end machines. When the POWER4 manufacturing process starts maturing to the point where 95% are whole dual core, it should become economically feasable to just put in the whole dual core POWER4.</strong><hr></blockquote>
I think yields in general are lower than you think -- I've heard numbers like 5-15% for the G4, and no higher than 50% even for well-tuned processors in full production. That IBM article on cellular architectures makes a big deal about being able to use chips which have one or more flaws which kill individual cores within the overall chip. That level of redundancy might be able to make yields 95% usable, albeit at a variety of core counts.
I can see the advantage of a cellular approach that would have some 16 mini cores or so. If one, two, or five cores die you would still have viable processor useful for another situation.
I think yields in general are lower than you think -- I've heard numbers like 5-15% for the G4, and no higher than 50% even for well-tuned processors in full production. . .
</strong><hr></blockquote>
It has been several years since I have been connected with IC manufacturing, but that sounds about right for processes a while back. There are a lot of chips on a waffer, and they are tested and marked before packaging. The larger the chip, the greater the chance for a defect. Smaller chips do have a yield advantage. Being able to disconnect bad cores would give some of that advantage to multicore processors. However, a defect anywhere else but in a core still kills the whole chip.
During the G4 fiasco, Apple began looking to IBM for its next generation processor. In the fall of 2000, IBM assembled its 970 development team at the request of Apple.</strong><hr></blockquote>
Forgive me, quotes such as this will explain why I don't take these rumors too seriously at all... IBM assembled it's team at the request of Apple... whatever...
Apple is in no position to be picky about speed right now. By the time the 970 is ready to go to market, Intel will be so far ahead of the game Apple won't have a chance. A better speculation would assume that Apple is trying to devise a new business design that would allow the release of OS X onto x86. Thanks to Apple's own "processor wars", consumers are wise to CPU performance. If Apple doesn't approach a standard platform to compete head-on with Microsoft it will show a blatant disregard for the needs of its customers. From a business perspective, x86 offers a lower-priced performance solution with a standard upgrade pricing ramp. Investors are happy, customers are happy--it's a win-win across the board. All that is needed at this point is a strategy to make it work--and I'm not sure Mr. Jobs is our man for the task.
because the upcoming G3 from IBM will have DDR-Ram support and Rapid-IO and will be multicore superscalar...? (and this all besides the fact that it'll sport a SIMD unit)
Not multicore - the G3 architecture was built for speed and simplicity, among other things by leaving out things that support multi-processing.
Similarly, the 970 is not planned to come in multicore versions - after all, it's not a Power4. The 990 family, on the other hand, is expected to include several multicore variations - but that depends entirely on what the market (read: IBM-produced unit sales) looks like by then. Three years is a long time.
Too many comments here are based on an Irish stew-perception of processor-development - "I want some of this, some of that - and then just a quick stir..."
Not multicore - the G3 architecture was built for speed and simplicity, among other things by leaving out things that support multi-processing.
Similarly, the 970 is not planned to come in multicore versions - after all, it's not a Power4. The 990 family, on the other hand, is expected to include several multicore variations - but that depends entirely on what the market (read: IBM-produced unit sales) looks like by then. Three years is a long time.
Too many comments here are based on an Irish stew-perception of processor-development - "I want some of this, some of that - and then just a quick stir..."
engpjp</strong><hr></blockquote>
ok i will modify a bit just to clear some things up. the "upcoming G3" will be a next generation chip from IBM. so it's not a G3 anymore.
--- also my statemens are not based on an Irish-stew-perception of processor-development but on the ibm-roadmap
--- further on i never said anything about multicore 970-ppcs ... although i believe further development of that cpu will result in a 970fc or whatever which could be multicored (ok if you want call it 990)...
If Wintel is trying to get out of the old legacy x86 for various reasons whey would Apple try to get to x86?
Yes Intel/AMD have pushed the old cranky x86 to speeds noboyd could have dreamed about. The Pentium replacement was planned to come out 1997 and it is sligtly deolayed but accordning to Intel that is were their future CPUs will be and they ought to knew.
Win XP is made for x86. Mac OS X with the Quarts engine etc is made for CPUs with the Velocity Engine etc. My guess is that even native OS X86 applications would suffer for a long time and then we have all the emaulation code.
Moving from the struggling 68040 to the 601 was a big leap. The 601 was a powerful CPU well on par with the pentiums at similar pentiums running UNIX. In the Mac however the emulation in OS and applications killed the performance completly and the pentium totaly killed those 601 computers.
Some months ago I booted a 7200/90 into the earliest OS 7 that came with the computer. Truly terrible way slower than 8.6!
A migration to x86will set up the Mac for such trouble agian. Not a good idea. In the future when the Macintosh has leaft all the pre OS X legacy behind and the Intel is past x86 it might be a good alternative but now it is not <img src="graemlins/oyvey.gif" border="0" alt="[No]" />
[quote] By the time the 970 is ready to go to market, Intel will be so far ahead of the game Apple won't have a chance. <hr></blockquote>
Well first of all it would be nice if you based your post on real information rathar than a prediction based on previous years advances in the speed of intel processors. Intel itself has projected 3.4Ghz by around the time the 970 would be out.
In addition, just because 1.8Ghz is the target speed for the first 970 chips does not mean that it won't actually debut at 2Ghz or higher once the assembly lines get going.
Thirdly the key word for the 970 is going to be "throughput" it's not just Mhz as evidenced by the preliminary benchmark projections in comparison to the current P4.
What x86 processor do you think is going to go in an iBook a Celeron? Oh please. You know there hasn't been a x86 laptop for a long time that hasn't had a crippled or half baked excuse for a processor in it for a long time. Going to x86 would kill Apple's portable lineup.
The fact that the new 3.06Ghz P4 consumes around 80 Watts and with HT (hyperthreading) enabled can go as high as 105 Watts should tell you that x86 is getting close to a wall. Yeah they are going to 90 nm soon but so is IBM who has been talking about 5Ghz lately. Using 90nm the 970 could probably go into a Powerbook. Intel on the other hand has started to back off speed projections including their coveted 10Ghz mark.
High end x86 processors have run more expensive than the G4 and likely the 970. How many times do people have to explain total cost of ownership? I quote one of my earlier posts
[quote] My questions is this. Would Apple really give up control of what hardware OSX is run on and throw themselves back into the "Clone Wars" again just to gain that market share or would they try to get these people to switch to Apple PPC hardware? Some would probably go to Linux rathar than buy a new PC I'm sure but as we all know running an OS on a platform where everyone and their brother builds machines with a hundreds of different possible hardware combinations (chip sets, video cards, network cards, etc) is a support nightmare. Some of the problems with PC's are certainly the fault of Windows but not all. This is where the Macintosh gets it's low total cost of ownership. <hr></blockquote>
I have seen things in PC support enviroments such as two supposedly identical machines where one refuses to use a video card that the other machine will use just fine. If you unleash OSX onto machines like this expect that and even worse to happen just like PC's with Windows on them. Again some of it is Windows but a lot of it is the nature of the platform.
if you're talking about proprietary Apple x86 hardware AND a port of OSX, the machines would likely not be much different in price at all. They still wouldn't be selling in enough volume to bring the price to the level you expect. The volume you sell does affect the price. Just ask Wal Mart.
Do you think IBM will ignore hyper-threading, or is there a possibility that it is being worked on for the Power 5, as another post suggested? (I guess if they did do hyper-threading, it would be a reason to use it only with single core chips.) Or is it going to be a contest between single core, hyper-threaded chips, and multi core PPC chips?</strong><hr></blockquote>
SMT (Symmetrical Multi-Threading) is one of the planned features of the Power5 which is expected to ship in 2004. There are many articles on the web that discuss this. Here's one:
"Another new feature in POWER5 will be simultaneous multithreading (SMT). The idea behind SMT is to share the processor hardware on a chip among multiple threads in a multiprogrammed workload. In this way, a single processor on the chip can sometimes act as two processors."
Not multicore - the G3 architecture was built for speed and simplicity, among other things by leaving out things that support multi-processing.
Similarly, the 970 is not planned to come in multicore versions - after all, it's not a Power4. The 990 family, on the other hand, is expected to include several multicore variations - but that depends entirely on what the market (read: IBM-produced unit sales) looks like by then. Three years is a long time.
Too many comments here are based on an Irish stew-perception of processor-development - "I want some of this, some of that - and then just a quick stir..."
engpjp</strong><hr></blockquote>
IBM's roadmap shows the next generation Power PC as multicore superscalar, maybe it'll be a 860 series
In addition, just because 1.8Ghz is the target speed for the first 970 chips does not mean that it won't actually debut at 2Ghz or higher once the assembly lines get going.
.</strong><hr></blockquote>
Good point, and a precidence is the Gekko, which was delivered faster than the "anounced" speed was 400 mhz, but the shipping game cubes are using a 485 mhz chip.
Umm if that is true it means my iMac 800 is much faster than a 3.06Ghz P4. I get 2703.5Megaflops while the top of the line P4 only gets 1442 Megaflops. Wow I wonder how they got the macs to score so bad in the benchmark?
Umm if that is true it means my iMac 800 is much faster than a 3.06Ghz P4. I get 2703.5Megaflops while the top of the line P4 only gets 1442 Megaflops. Wow I wonder how they got the macs to score so bad in the benchmark?</strong><hr></blockquote>
You're comparing the G4's theoretical MFlops to the P4's measured MFlops in a Linpack test, which obviously makes no use of AltiVec, and which has probably been much more thoroughly optimized for the x86 architecture than for the PPC.
The theoretical MFlops number Apple gives for the G4 in your iMac assumes AltiVec running at maximum efficiency.
Apple didn't give me that number... Altivec Fractal did. And I get 406.6 megaflobs with altivec off V.2703.6 with it on. Man altivec sure makes the G4 fast. So guess what guys the G4 without altivec is quite slow, but then again if it didn't have altivec it wouldn't be a G4. So I guess that argument is kind of defeated. Oh well its nice to know that my imac is faster than a top of the line P4. at least according to these tests.
The theoretical MFlops number Apple gives for the G4 in your iMac assumes AltiVec running at maximum efficiency.</strong><hr></blockquote>
The theoretical MFlops number Apple gives for the G4 in the iMac is also much higher than the numbers Algol stated. I don't remember exactly, but I believe it's at least double Algol's number.
The theoretical MFlops number Apple gives for the G4 in the iMac is also much higher than the numbers Algol stated. I don't remember exactly, but I believe it's at least double Algol's number.</strong><hr></blockquote>
Yeah, he clarified that it was from AltiVec Fractal. My bad.
Comments
<strong>
. . . Also, if a manufacturing flaw kills part of a hyper-threaded processor then the whole processor is dead.
Single processor failures in a dual processor chip are common enough that IBM is shipping whole lines of machines based on them! Even if it is a fairly low percentage, this practice is only going to become more common as the number of cores grows. . .
</strong><hr></blockquote>
Do you think IBM will ignore hyper-threading, or is there a possibility that it is being worked on for the Power 5, as another post suggested? (I guess if they did do hyper-threading, it would be a reason to use it only with single core chips.) Or is it going to be a contest between single core, hyper-threaded chips, and multi core PPC chips?
<strong>
The problem with a hyper-threaded core is that resources are shared so that the threads need to be in the same process, and those threads should be doing different things (i.e. integer/float, float/vector, integer/vector). Also, if a manufacturing flaw kills part of a hyper-threaded processor then the whole processor is dead.
Single processor failures in a dual processor chip are common enough that IBM is shipping whole lines of machines based on them! Even if it is a fairly low percentage, this practice is only going to become more common as the number of cores grows. If you have 16 cores and a flaw kills one (or two, or three, etc) of them but your design still allows the others to operate independently, then why not sell it as a less powerful chip? This is going to be as important in the future as selling different chips speeds is today.</strong><hr></blockquote>
Good point but I think that the reason there are so much single core POWER4 with defective 2nd cores is due to the newness of the process and processor. Once these kinks are worked out they should have a higher output of complete processors. Maybe IBM didn't want to dispose of otherwise good single core processors and used them in low end machines. When the POWER4 manufacturing process starts maturing to the point where 95% are whole dual core, it should become economically feasable to just put in the whole dual core POWER4.
<strong>Good point but I think that the reason there are so much single core POWER4 with defective 2nd cores is due to the newness of the process and processor. Once these kinks are worked out they should have a higher output of complete processors. Maybe IBM didn't want to dispose of otherwise good single core processors and used them in low end machines. When the POWER4 manufacturing process starts maturing to the point where 95% are whole dual core, it should become economically feasable to just put in the whole dual core POWER4.</strong><hr></blockquote>
I think yields in general are lower than you think -- I've heard numbers like 5-15% for the G4, and no higher than 50% even for well-tuned processors in full production. That IBM article on cellular architectures makes a big deal about being able to use chips which have one or more flaws which kill individual cores within the overall chip. That level of redundancy might be able to make yields 95% usable, albeit at a variety of core counts.
<strong>
I think yields in general are lower than you think -- I've heard numbers like 5-15% for the G4, and no higher than 50% even for well-tuned processors in full production. . .
</strong><hr></blockquote>
It has been several years since I have been connected with IC manufacturing, but that sounds about right for processes a while back. There are a lot of chips on a waffer, and they are tested and marked before packaging. The larger the chip, the greater the chance for a defect. Smaller chips do have a yield advantage. Being able to disconnect bad cores would give some of that advantage to multicore processors. However, a defect anywhere else but in a core still kills the whole chip.
<strong>
During the G4 fiasco, Apple began looking to IBM for its next generation processor. In the fall of 2000, IBM assembled its 970 development team at the request of Apple.</strong><hr></blockquote>
Forgive me, quotes such as this will explain why I don't take these rumors too seriously at all... IBM assembled it's team at the request of Apple... whatever...
Apple is in no position to be picky about speed right now. By the time the 970 is ready to go to market, Intel will be so far ahead of the game Apple won't have a chance. A better speculation would assume that Apple is trying to devise a new business design that would allow the release of OS X onto x86. Thanks to Apple's own "processor wars", consumers are wise to CPU performance. If Apple doesn't approach a standard platform to compete head-on with Microsoft it will show a blatant disregard for the needs of its customers. From a business perspective, x86 offers a lower-priced performance solution with a standard upgrade pricing ramp. Investors are happy, customers are happy--it's a win-win across the board. All that is needed at this point is a strategy to make it work--and I'm not sure Mr. Jobs is our man for the task.
:eek:
[ 12-02-2002: Message edited by: MacLuv ]</p>
Such foresight, MacLuv...
Screed
<strong>
because the upcoming G3 from IBM will have DDR-Ram support and Rapid-IO and will be multicore superscalar...? (and this all besides the fact that it'll sport a SIMD unit)
edit: i'd call this one a "G5" and the 970 "G6"
[ 11-26-2002: Message edited by: Krassy ]</strong><hr></blockquote>
Not multicore - the G3 architecture was built for speed and simplicity, among other things by leaving out things that support multi-processing.
Similarly, the 970 is not planned to come in multicore versions - after all, it's not a Power4. The 990 family, on the other hand, is expected to include several multicore variations - but that depends entirely on what the market (read: IBM-produced unit sales) looks like by then. Three years is a long time.
Too many comments here are based on an Irish stew-perception of processor-development - "I want some of this, some of that - and then just a quick stir..."
engpjp
<strong>
Not multicore - the G3 architecture was built for speed and simplicity, among other things by leaving out things that support multi-processing.
Similarly, the 970 is not planned to come in multicore versions - after all, it's not a Power4. The 990 family, on the other hand, is expected to include several multicore variations - but that depends entirely on what the market (read: IBM-produced unit sales) looks like by then. Three years is a long time.
Too many comments here are based on an Irish stew-perception of processor-development - "I want some of this, some of that - and then just a quick stir..."
engpjp</strong><hr></blockquote>
ok i will modify a bit just to clear some things up. the "upcoming G3" will be a next generation chip from IBM. so it's not a G3 anymore.
If Wintel is trying to get out of the old legacy x86 for various reasons whey would Apple try to get to x86?
Yes Intel/AMD have pushed the old cranky x86 to speeds noboyd could have dreamed about. The Pentium replacement was planned to come out 1997 and it is sligtly deolayed but accordning to Intel that is were their future CPUs will be and they ought to knew.
Win XP is made for x86. Mac OS X with the Quarts engine etc is made for CPUs with the Velocity Engine etc. My guess is that even native OS X86 applications would suffer for a long time and then we have all the emaulation code.
Moving from the struggling 68040 to the 601 was a big leap. The 601 was a powerful CPU well on par with the pentiums at similar pentiums running UNIX. In the Mac however the emulation in OS and applications killed the performance completly and the pentium totaly killed those 601 computers.
Some months ago I booted a 7200/90 into the earliest OS 7 that came with the computer. Truly terrible way slower than 8.6!
A migration to x86will set up the Mac for such trouble agian. Not a good idea. In the future when the Macintosh has leaft all the pre OS X legacy behind and the Intel is past x86 it might be a good alternative but now it is not <img src="graemlins/oyvey.gif" border="0" alt="[No]" />
Well first of all it would be nice if you based your post on real information rathar than a prediction based on previous years advances in the speed of intel processors. Intel itself has projected 3.4Ghz by around the time the 970 would be out.
In addition, just because 1.8Ghz is the target speed for the first 970 chips does not mean that it won't actually debut at 2Ghz or higher once the assembly lines get going.
Thirdly the key word for the 970 is going to be "throughput" it's not just Mhz as evidenced by the preliminary benchmark projections in comparison to the current P4.
What x86 processor do you think is going to go in an iBook a Celeron? Oh please. You know there hasn't been a x86 laptop for a long time that hasn't had a crippled or half baked excuse for a processor in it for a long time. Going to x86 would kill Apple's portable lineup.
The fact that the new 3.06Ghz P4 consumes around 80 Watts and with HT (hyperthreading) enabled can go as high as 105 Watts should tell you that x86 is getting close to a wall. Yeah they are going to 90 nm soon but so is IBM who has been talking about 5Ghz lately. Using 90nm the 970 could probably go into a Powerbook. Intel on the other hand has started to back off speed projections including their coveted 10Ghz mark.
High end x86 processors have run more expensive than the G4 and likely the 970. How many times do people have to explain total cost of ownership? I quote one of my earlier posts
[quote] My questions is this. Would Apple really give up control of what hardware OSX is run on and throw themselves back into the "Clone Wars" again just to gain that market share or would they try to get these people to switch to Apple PPC hardware? Some would probably go to Linux rathar than buy a new PC I'm sure but as we all know running an OS on a platform where everyone and their brother builds machines with a hundreds of different possible hardware combinations (chip sets, video cards, network cards, etc) is a support nightmare. Some of the problems with PC's are certainly the fault of Windows but not all. This is where the Macintosh gets it's low total cost of ownership. <hr></blockquote>
I have seen things in PC support enviroments such as two supposedly identical machines where one refuses to use a video card that the other machine will use just fine. If you unleash OSX onto machines like this expect that and even worse to happen just like PC's with Windows on them. Again some of it is Windows but a lot of it is the nature of the platform.
if you're talking about proprietary Apple x86 hardware AND a port of OSX, the machines would likely not be much different in price at all. They still wouldn't be selling in enough volume to bring the price to the level you expect. The volume you sell does affect the price. Just ask Wal Mart.
<strong>
Do you think IBM will ignore hyper-threading, or is there a possibility that it is being worked on for the Power 5, as another post suggested? (I guess if they did do hyper-threading, it would be a reason to use it only with single core chips.) Or is it going to be a contest between single core, hyper-threaded chips, and multi core PPC chips?</strong><hr></blockquote>
SMT (Symmetrical Multi-Threading) is one of the planned features of the Power5 which is expected to ship in 2004. There are many articles on the web that discuss this. Here's one:
<a href="http://www.iseriesnetwork.com/resources/artarchive/index.cfm?fuseaction=viewarticle&CO_ContentID=1523 8&channel=" target="_blank">http://www.iseriesnetwork.com/resources/artarchive/index.cfm?fuseaction=viewarticle&CO_ContentID=1523 8&channel=</a>
"Another new feature in POWER5 will be simultaneous multithreading (SMT). The idea behind SMT is to share the processor hardware on a chip among multiple threads in a multiprogrammed workload. In this way, a single processor on the chip can sometimes act as two processors."
Steve
<strong>
Not multicore - the G3 architecture was built for speed and simplicity, among other things by leaving out things that support multi-processing.
Similarly, the 970 is not planned to come in multicore versions - after all, it's not a Power4. The 990 family, on the other hand, is expected to include several multicore variations - but that depends entirely on what the market (read: IBM-produced unit sales) looks like by then. Three years is a long time.
Too many comments here are based on an Irish stew-perception of processor-development - "I want some of this, some of that - and then just a quick stir..."
engpjp</strong><hr></blockquote>
IBM's roadmap shows the next generation Power PC as multicore superscalar, maybe it'll be a 860 series
<strong>
In addition, just because 1.8Ghz is the target speed for the first 970 chips does not mean that it won't actually debut at 2Ghz or higher once the assembly lines get going.
.</strong><hr></blockquote>
Good point, and a precidence is the Gekko, which was delivered faster than the "anounced" speed was 400 mhz, but the shipping game cubes are using a 485 mhz chip.
<strong>
By the time the 970 is ready to go to market, Intel will be so far ahead of the game Apple won't have a chance.
[ 12-02-2002: Message edited by: MacLuv ]</strong><hr></blockquote>
underlining your quote ...
Source: c't 25/2002
[ 12-02-2002: Message edited by: tomk ]</p>
<strong>
underlining your quote ...
Source: c't 25/2002
[ 12-02-2002: Message edited by: tomk ]</strong><hr></blockquote>
Umm if that is true it means my iMac 800 is much faster than a 3.06Ghz P4. I get 2703.5Megaflops while the top of the line P4 only gets 1442 Megaflops. Wow I wonder how they got the macs to score so bad in the benchmark?
<strong>
Umm if that is true it means my iMac 800 is much faster than a 3.06Ghz P4. I get 2703.5Megaflops while the top of the line P4 only gets 1442 Megaflops. Wow I wonder how they got the macs to score so bad in the benchmark?</strong><hr></blockquote>
You're comparing the G4's theoretical MFlops to the P4's measured MFlops in a Linpack test, which obviously makes no use of AltiVec, and which has probably been much more thoroughly optimized for the x86 architecture than for the PPC.
The theoretical MFlops number Apple gives for the G4 in your iMac assumes AltiVec running at maximum efficiency.
<strong>
The theoretical MFlops number Apple gives for the G4 in your iMac assumes AltiVec running at maximum efficiency.</strong><hr></blockquote>
The theoretical MFlops number Apple gives for the G4 in the iMac is also much higher than the numbers Algol stated. I don't remember exactly, but I believe it's at least double Algol's number.
<strong>
The theoretical MFlops number Apple gives for the G4 in the iMac is also much higher than the numbers Algol stated. I don't remember exactly, but I believe it's at least double Algol's number.</strong><hr></blockquote>
Yeah, he clarified that it was from AltiVec Fractal. My bad.