The engineers at Intel knew there would be problems. Netburst was the bastard child of marketers who had no idea what went into chip design.
And for you to say that AMD's core design "isn't that good", you should stop acting like you know about core design, because that's entirely untrue.
This was a "content free" post. Someone with "genius" in their handle should be providing more info.
Netburst architecture wasn't about marketing. It was designed around taking a core and highly pipelining the core for high clock rates. Intel thought they'd be able to clock the core higher than their eventual ceiling. They marketed on "clockspeed" because that was the smart thing to do.
Melgross is right. Everyone is saying exactly what he is. If Barcelona doesn't clock higher than 2.3Ghz they are pretty much done. A true quad core with hypertransport isn't going to make a 2.3Ghz quad perform better than a 3.2 MCM Intel quad
They are working on it. It's one of the hardest areas in software development.
Writing multi-threaded mac apps isn't very hard. It isn't as easy, however, for Windows, but it's still my opinion that only dumb programmers have problems getting stuff multi-threaded. Yes, there are plenty of dumb programmers out there, but they're not working for the companies that make high-performance software. If there are some utility apps and the like that aren't well multi-threaded, that's not going to be much of a problem. What's more important is that the Adobe suite, Mac OS, and other major apps are nicely multi-threaded. This seems to be the case already.
Writing multi-threaded mac apps isn't very hard. It isn't as easy, however, for Windows, but it's still my opinion that only dumb programmers have problems getting stuff multi-threaded. Yes, there are plenty of dumb programmers out there, but they're not working for the companies that make high-performance software. If there are some utility apps and the like that aren't well multi-threaded, that's not going to be much of a problem. What's more important is that the Adobe suite, Mac OS, and other major apps are nicely multi-threaded. This seems to be the case already.
It may not be hard for programs that naturally lend themselves to it, such as rendering programs. But, more "normal" programs can't be multi threaded in any useful way easily. If a program is linear in nature, multi-core designs can't wring much more performance from them. Two cores is about the practical limit with many programs.
Word, and other programs will give any programmer a fit if (s)he tries to get it to efficiently use 8 cores. And in two years, at the most, we will be using 16 cores, and at most, 2 years later, 32 cores.
Multithreading programs efficiently is one of the most difficult problems out there.
Right now, PS in CS3 diesn't seem to be doing very well using those 8 cores, though it should be a natural. If you look for some of the scores that MacWorld published, for example, you will see little difference even running 4 cores.
This was a "content free" post. Someone with "genius" in their handle should be providing more info.
Netburst architecture wasn't about marketing. It was designed around taking a core and highly pipelining the core for high clock rates. Intel thought they'd be able to clock the core higher than their eventual ceiling. They marketed on "clockspeed" because that was the smart thing to do.
Melgross is right. Everyone is saying exactly what he is. If Barcelona doesn't clock higher than 2.3Ghz they are pretty much done. A true quad core with hypertransport isn't going to make a 2.3Ghz quad perform better than a 3.2 MCM Intel quad
Haven't we learned that clockspeed doesn't equal performance?
Haven't we learned that clockspeed doesn't equal performance?
No, we haven't.
The MHz myth was itself a myth.
Look, there is a great deal of misunderstanding about all of this. MHz matters. Indeed it does!
Where the concept goes wrong is when one compares different architectures.
Comparing the same architecture gives results that make sense. A chip of the same line will always perform better if it runs faster. That's pretty simple. If you compare totally different chip families, then the equation breaks down. A 2.5 GHz chip from one family may be as powerful as a 3.0 GHz chip from another family. But, if you are comparing the same families through their speed ramps, that will hold true. So, you can still make a comparison. AMD did that with the naming of their chips. They used to name them according to the INTEL speed they would perform at (at least, according to AMD).
Intel and AMD went in different directions. Intel always had the better process technology. Because of that, they were able to produce chips that AMD couldn't, e.g.. long pipelined architectures. That gave the chips great speed, and as long as the shrinking process technologies were up to it, an insurmountable lead over AMD, which over most of its history, was hapless.
AMD, on the other hand, not having as good a good process technology, needed to work around that. They found a way with their designs, which like the PPC, relied more on width than length, so to speak.
This only began to work once the 90 nm process shrink came about. Until then, smaller processes led to less current, which led to less total power, which led to less heat, which led to faster chips, which led to Intel kicking the crap out of AMD, year after year.
But, when the 90 nm shrink led to a suddenly huge increase in leakage, as well as other power related problems, Intel's relentless march to the ultimate speed came to a grinding halt.
During that same period, IBM found it had the same problem. But, IBM, like AMD, was running chips that already had a wider path, and other differences, that gave them greater processing power at a lower clock.
Because of that, when Intel took 9 months to raise the clock of their chips from 3.06 GHz to 3.31 GHz, both IBM, and AMD, were able to raise theirs by a larger percentage, because they were at a lower clock to begin with.
But, even at the peak of AMD's newfound dominance, their fastest chips never were more than about 15% better than Intel's. IBM's were about 5% less in performance than Intels, and about 20% less than AMD's.
Remember that with all of this, AMD had the on-die memory controller, and Hypertransport. Both of those gave AMD all of their performance advantages, proving that their basic chip designs in and of themselves were no better than those of Intel, possibly a bit worse, and about the same as those from IBM.
Intel's Hyperthreading had no use on most programs, and indeed, was blamed for actually slowing down its performance often, as programs weren't designed to take advantage of it.
You must have noticed that the very first generation of chips that left the Netburst line moved easily past the best that AMD could offer.
The generation past that moved far past AMD's best chips.
With all of that, the best is yet to come from Intel. In 2008 we will see on-die memory controllers, the return of Hyperthreading, now that programs will be able to use it, a much better vector processor, and even a rival to Hypertransport.
What's a chip company like AMD to do?
They can't afford to do the R&D that Intel routinely invests in. One thing they will do is to wait until Intel does the R&D, and copy it with some curves of their own, assuming that Intel doesn't patent them first.
But, they are again, a full process shrink behind. that only makes it worse.
And don't think that Intel isn't holding back!
Apple's coming out with factory authorized 3.0 GHz Cloverdales, significantly faster than the standard 2.66 GHz versions, shows that Intel's designs are reliable enough to sell to a high profile customer at faster than announced speeds.
In fact, Intel's chips overclock so well, that many in the industry think that Intel is deliberately underclocking them.
On the other hand, AMD's recent chips hardly overclock at all, showing that they have barely any margin.
It's been thought that Intel is teasing AMD with these speeds, daring them to attempt a faster chip, which they will trump as soon as it's announced.
At any rate, 2008 should be an eventful year for us in Macland.
I know enough about CPU architecture to at least compare the CPUs (thanks Anandtech)
Quite honestly if we were talking about Cell and AMD/Intel clockspeed wouldn't give us much info. However with Barcelona we know that AMD has what a 3 or 4 issue core. I believe it's 3 so the at best they are matching the Intel chips. So clockspeed makes a huge difference.
Then you movie to caching and memory bandwith which AMD has advantage with as Melgross has stated until Nehalem at least.
I see AMD keeping up but not surpassing Intel. I want AMD to stay close enough to keep Intel honest. So if I build a PC it'll be with an AMD CPU.
So what is a realistic estimate of seeing these penryns in macbook pros, late 07, or earlyish 08
Apple is certainly going through some changes. We can't even rely on "Apple Tuesdays" anymore. I have tried to run the numbers but I can't find any path that seems more likely than another path based on previous history.
The factors I use in my attempt to find an answer:
age of current system
mobo redesign difficulty
availability of new processor
remaining stock of current processor
cost difference of processor
performance difference
power usage difference
There are other factors that I can't possibly know. I've also read--on one site--that the Gilo (65nm) as the successor to Merom, but I think Apple will stick with Merom until the Penryn is out. If I had to guess, I'd say that we won't see Penryn until April 2008.
Apple is certainly going through some changes. We can't even rely on "Apple Tuesdays" anymore. I have tried to run the numbers but I can't find any path that seems more likely than another path based on previous history.
The factors I use in my attempt to find an answer:
age of current system
mobo redesign difficulty
availability of new processor
remaining stock of current processor
cost difference of processor
performance difference
power usage difference
There are other factors that I can't possibly know. I've also read--on one site--that the Gilo (65nm) as the successor to Merom, but I think Apple will stick with Merom until the Penryn is out. If I had to guess, I'd say that we won't see Penryn until April 2008.
I don't know, we might see it in January. Apple was the first to get the Yonah, they could be the first to get Penyrn as well. It's supposed to be here in October, and I'm reading here and there, that Intel could release it sooner, depending on the intro of Barcelona.
Comments
The engineers at Intel knew there would be problems. Netburst was the bastard child of marketers who had no idea what went into chip design.
And for you to say that AMD's core design "isn't that good", you should stop acting like you know about core design, because that's entirely untrue.
This was a "content free" post. Someone with "genius" in their handle should be providing more info.
Netburst architecture wasn't about marketing. It was designed around taking a core and highly pipelining the core for high clock rates. Intel thought they'd be able to clock the core higher than their eventual ceiling. They marketed on "clockspeed" because that was the smart thing to do.
Melgross is right. Everyone is saying exactly what he is. If Barcelona doesn't clock higher than 2.3Ghz they are pretty much done. A true quad core with hypertransport isn't going to make a 2.3Ghz quad perform better than a 3.2 MCM Intel quad
They are working on it. It's one of the hardest areas in software development.
Writing multi-threaded mac apps isn't very hard. It isn't as easy, however, for Windows, but it's still my opinion that only dumb programmers have problems getting stuff multi-threaded. Yes, there are plenty of dumb programmers out there, but they're not working for the companies that make high-performance software. If there are some utility apps and the like that aren't well multi-threaded, that's not going to be much of a problem. What's more important is that the Adobe suite, Mac OS, and other major apps are nicely multi-threaded. This seems to be the case already.
Writing multi-threaded mac apps isn't very hard. It isn't as easy, however, for Windows, but it's still my opinion that only dumb programmers have problems getting stuff multi-threaded. Yes, there are plenty of dumb programmers out there, but they're not working for the companies that make high-performance software. If there are some utility apps and the like that aren't well multi-threaded, that's not going to be much of a problem. What's more important is that the Adobe suite, Mac OS, and other major apps are nicely multi-threaded. This seems to be the case already.
It may not be hard for programs that naturally lend themselves to it, such as rendering programs. But, more "normal" programs can't be multi threaded in any useful way easily. If a program is linear in nature, multi-core designs can't wring much more performance from them. Two cores is about the practical limit with many programs.
Word, and other programs will give any programmer a fit if (s)he tries to get it to efficiently use 8 cores. And in two years, at the most, we will be using 16 cores, and at most, 2 years later, 32 cores.
Multithreading programs efficiently is one of the most difficult problems out there.
Right now, PS in CS3 diesn't seem to be doing very well using those 8 cores, though it should be a natural. If you look for some of the scores that MacWorld published, for example, you will see little difference even running 4 cores.
Any one have a guess when in May 2007 Santa rosa will be released?
My sources say sometime between the 1st and the 31st.
Any one have a guess when in May 2007 Santa rosa will be released?
http://computershopper.com/shoptalk/...ro_set_to_laun
The 9th, possibly.
My sources say sometime between the 1st and the 31st.
You ought to know better than to pinpoint a release date like that.
http://computershopper.com/shoptalk/...ro_set_to_laun
The 9th, possibly.
Did you to buy those huge copies of Computershopper from the newsstands like I did?
Did you to buy those huge copies of Computershopper from the newsstands like I did?
'fraid not. I merely searched for any information on a proper date, and this came up.
'fraid not. I merely searched for any information on a proper date, and this came up.
That was a big part of computer history. I still keep two copies for historical reasons.
This was a "content free" post. Someone with "genius" in their handle should be providing more info.
Netburst architecture wasn't about marketing. It was designed around taking a core and highly pipelining the core for high clock rates. Intel thought they'd be able to clock the core higher than their eventual ceiling. They marketed on "clockspeed" because that was the smart thing to do.
Melgross is right. Everyone is saying exactly what he is. If Barcelona doesn't clock higher than 2.3Ghz they are pretty much done. A true quad core with hypertransport isn't going to make a 2.3Ghz quad perform better than a 3.2 MCM Intel quad
Haven't we learned that clockspeed doesn't equal performance?
Haven't we learned that clockspeed doesn't equal performance?
No, we haven't.
The MHz myth was itself a myth.
Look, there is a great deal of misunderstanding about all of this. MHz matters. Indeed it does!
Where the concept goes wrong is when one compares different architectures.
Comparing the same architecture gives results that make sense. A chip of the same line will always perform better if it runs faster. That's pretty simple. If you compare totally different chip families, then the equation breaks down. A 2.5 GHz chip from one family may be as powerful as a 3.0 GHz chip from another family. But, if you are comparing the same families through their speed ramps, that will hold true. So, you can still make a comparison. AMD did that with the naming of their chips. They used to name them according to the INTEL speed they would perform at (at least, according to AMD).
Intel and AMD went in different directions. Intel always had the better process technology. Because of that, they were able to produce chips that AMD couldn't, e.g.. long pipelined architectures. That gave the chips great speed, and as long as the shrinking process technologies were up to it, an insurmountable lead over AMD, which over most of its history, was hapless.
AMD, on the other hand, not having as good a good process technology, needed to work around that. They found a way with their designs, which like the PPC, relied more on width than length, so to speak.
This only began to work once the 90 nm process shrink came about. Until then, smaller processes led to less current, which led to less total power, which led to less heat, which led to faster chips, which led to Intel kicking the crap out of AMD, year after year.
But, when the 90 nm shrink led to a suddenly huge increase in leakage, as well as other power related problems, Intel's relentless march to the ultimate speed came to a grinding halt.
During that same period, IBM found it had the same problem. But, IBM, like AMD, was running chips that already had a wider path, and other differences, that gave them greater processing power at a lower clock.
Because of that, when Intel took 9 months to raise the clock of their chips from 3.06 GHz to 3.31 GHz, both IBM, and AMD, were able to raise theirs by a larger percentage, because they were at a lower clock to begin with.
But, even at the peak of AMD's newfound dominance, their fastest chips never were more than about 15% better than Intel's. IBM's were about 5% less in performance than Intels, and about 20% less than AMD's.
Remember that with all of this, AMD had the on-die memory controller, and Hypertransport. Both of those gave AMD all of their performance advantages, proving that their basic chip designs in and of themselves were no better than those of Intel, possibly a bit worse, and about the same as those from IBM.
Intel's Hyperthreading had no use on most programs, and indeed, was blamed for actually slowing down its performance often, as programs weren't designed to take advantage of it.
You must have noticed that the very first generation of chips that left the Netburst line moved easily past the best that AMD could offer.
The generation past that moved far past AMD's best chips.
With all of that, the best is yet to come from Intel. In 2008 we will see on-die memory controllers, the return of Hyperthreading, now that programs will be able to use it, a much better vector processor, and even a rival to Hypertransport.
What's a chip company like AMD to do?
They can't afford to do the R&D that Intel routinely invests in. One thing they will do is to wait until Intel does the R&D, and copy it with some curves of their own, assuming that Intel doesn't patent them first.
But, they are again, a full process shrink behind. that only makes it worse.
And don't think that Intel isn't holding back!
Apple's coming out with factory authorized 3.0 GHz Cloverdales, significantly faster than the standard 2.66 GHz versions, shows that Intel's designs are reliable enough to sell to a high profile customer at faster than announced speeds.
In fact, Intel's chips overclock so well, that many in the industry think that Intel is deliberately underclocking them.
On the other hand, AMD's recent chips hardly overclock at all, showing that they have barely any margin.
It's been thought that Intel is teasing AMD with these speeds, daring them to attempt a faster chip, which they will trump as soon as it's announced.
At any rate, 2008 should be an eventful year for us in Macland.
Quite honestly if we were talking about Cell and AMD/Intel clockspeed wouldn't give us much info. However with Barcelona we know that AMD has what a 3 or 4 issue core. I believe it's 3 so the at best they are matching the Intel chips. So clockspeed makes a huge difference.
Then you movie to caching and memory bandwith which AMD has advantage with as Melgross has stated until Nehalem at least.
I see AMD keeping up but not surpassing Intel. I want AMD to stay close enough to keep Intel honest. So if I build a PC it'll be with an AMD CPU.
So what is a realistic estimate of seeing these penryns in macbook pros, late 07, or earlyish 08
Apple is certainly going through some changes. We can't even rely on "Apple Tuesdays" anymore. I have tried to run the numbers but I can't find any path that seems more likely than another path based on previous history.
The factors I use in my attempt to find an answer:
- age of current system
- mobo redesign difficulty
- availability of new processor
- remaining stock of current processor
- cost difference of processor
- performance difference
- power usage difference
There are other factors that I can't possibly know. I've also read--on one site--that the Gilo (65nm) as the successor to Merom, but I think Apple will stick with Merom until the Penryn is out. If I had to guess, I'd say that we won't see Penryn until April 2008.Apple is certainly going through some changes. We can't even rely on "Apple Tuesdays" anymore. I have tried to run the numbers but I can't find any path that seems more likely than another path based on previous history.
The factors I use in my attempt to find an answer:
- age of current system
- mobo redesign difficulty
- availability of new processor
- remaining stock of current processor
- cost difference of processor
- performance difference
- power usage difference
There are other factors that I can't possibly know. I've also read--on one site--that the Gilo (65nm) as the successor to Merom, but I think Apple will stick with Merom until the Penryn is out. If I had to guess, I'd say that we won't see Penryn until April 2008.I don't know, we might see it in January. Apple was the first to get the Yonah, they could be the first to get Penyrn as well. It's supposed to be here in October, and I'm reading here and there, that Intel could release it sooner, depending on the intro of Barcelona.
So what is a realistic estimate of seeing these penryns in macbook pros, late 07, or earlyish 08
I think Intel's stance on Penryn mobile is early 2008 so I agree with Solipsism
The original number of cores for those lines was only one reason Intel used those naming conventions, and not the most important one at all.
About the confusion: Don't worry, we are all confused, as always.
http://www.intel.com/pressroom/archi...qsub_20070328r
if we read the article carefully, it states Core 2 Quad
PENRYN - Core 2 - with newer CPU numbers
NEHALEM - Core 3
http://www.intel.com/pressroom/archi...qsub_20070328r
if we read the article carefully, it states Core 2 Quad
PENRYN - Core 2 - with newer CPU numbers
NEHALEM - Core 3
Yes, I discussed that in another post. Intel added to those families later, but the original numbers held true.