To get much better performance at that clock speed they would need to adjust the architecture of the chip: give it larger caches (which I suspect they will...
I suspect they won't. IBM wants to keep an appreciable performance gap between the Power4 and the 970. Small caches make that happen.
The Power4 also has a sick number of pins: 5500, compared to the 500ish in desktop CPUs . More L2 cache is quite likely but I wouldn't worry about the lack of L3. The G4 needed large slabs of L3 cache, especially with 2 G4s sharing one 167MHz MPX bus. Each 970 sits on its own bus running effectively at half of the CPU's core.
IIRC, the 970 is currently manufactured on a 0.13u process. How soon before it goes to 0.09u and gains a low K dielectric bonus?
Anyone want to try overclocking their PowerMac G5 when they get it?
What are the odds that IBM could make a transition of their POWER business to GPuL type chips?
mebbe not all of their business, but a large chunk of it?
If POWER chips cost lots and lots to make, mebbe they could make better margins using multiple GPuL chips in their place?
Apple seems to have a controller chip for single and duals, but I might be possible to have 4, 8, 16... GPuL's in a single box, a very expensive box, b ut no more than what a POWER based box costs you today.
That would certainly keep development of the G5/6/7... current.
What are the odds that IBM could make a transition of their POWER business to GPuL type chips?
mebbe not all of their business, but a large chunk of it?
If POWER chips cost lots and lots to make, mebbe they could make better margins using multiple GPuL chips in their place?
Apple seems to have a controller chip for single and duals, but I might be possible to have 4, 8, 16... GPuL's in a single box, a very expensive box, b ut no more than what a POWER based box costs you today.
That would certainly keep development of the G5/6/7... current.
POWER chips and the PPC 970 + successors will be aimed at different markets. The PPC 970 will really help IBM compete better with the Xeon + Opteron market while the POWER series continues to play higher up the food chain.
Don't expect either to disappear in the near future. Both have significant roadmaps ahead.
I wonder. How many bumps will Apple do? Will they bump their machines more frequently now they've got somebody who knows how to make chips?
Or will we keep the 'top goes to bottom' model every 6 months?
That's actually a good question. I suspect you'll see Apple move to a faster revision cycle with smaller bumps, more like the PC world uses, since it encourages purchases. When you know there is only going to be 2 updates a year there's far more chance you're going to wait for them.
It's not really set in stone, but that's about right.
And as far as jumping from 2 to 3 GHz (+50%) in 12 months is concerned...
Since AMD can't keep up, Intel has gotten lazy. They only plan on reaching 3.4 GHz by the end of the year...whereas they released 2.53 GHz P4s in May 2002. A 34% jump in ~18 months.
AMD has been even worse off. They 've increased clockspeeds by 32% since January 2002, going from 1.67 GHz to 2.2 GHz. They briefly had a 2.25 GHz chip but they couldn't make an appreciable number of them.
This is at best simplification and at worst plain distortion. Case in point: Pentium-M = same clock speed but faster. P4 800 FSB = same clock speed but faster. Plus, you are comparing speculation "we will do X in Y amount of time" to actual results.
What are the odds that IBM could make a transition of their POWER business to GPuL type chips?
mebbe not all of their business, but a large chunk of it?
If POWER chips cost lots and lots to make, mebbe they could make better margins using multiple GPuL chips in their place?
Apple seems to have a controller chip for single and duals, but I might be possible to have 4, 8, 16... GPuL's in a single box, a very expensive box, b ut no more than what a POWER based box costs you today.
That would certainly keep development of the G5/6/7... current.
IBM has been play'n with quads of the GUL since last July ... while Apple was play'n with dualies.
POWER chips and the PPC 970 + successors will be aimed at different markets. The PPC 970 will really help IBM compete better with the Xeon + Opteron market while the POWER series continues to play higher up the food chain.
Don't expect either to disappear in the near future. Both have significant roadmaps ahead.
Tis very true. Made even better because Dell doesn't have 64bit processing on their "workstation roadmap" till next year (at least up until April 2004). Apple has an "opportunity", I hope they capitalize on it.
"We have the new, all powerful G4. it's shipping at 550Mhz!!!"
ok, make that 500Mhz. for hte next ****ING YEAR. (or was it longer?)
in any case, the G4 initially got stuck a LONG time at its initial speeds. this would be to remove that fear with this new chip, IMO.
Spot On!, This is WHY IBM has announced the chip will scale to 3GHz within 12mths. Also, its been reported that the 970 @ .13um process will indeed scale to 2.5GHz. So within 6 months we will see this speed and then 3GHz, with the speed bump beyond 3GHz could well be the 980 chip around early 2005.
The 970 by then will be @ 0.09um and employed into the iMacs/eMacs at the very least.
The point of announcing a 3GHz in 12mths, means we (as customers) can go out and invest in Apple (buy G5's) and its OS, knowing that the hardware will only get better! Its a bit of an enterprise thing as well - im thinking xserve customers too.
That's actually a good question. I suspect you'll see Apple move to a faster revision cycle with smaller bumps, more like the PC world uses, since it encourages purchases. When you know there is only going to be 2 updates a year there's far more chance you're going to wait for them.
It has t do with a chip manufacturer, too. Who wants to use up your production capacity when the whole line can move up a step?
This is at best simplification and at worst plain distortion. Case in point: Pentium-M = same clock speed but faster. P4 800 FSB = same clock speed but faster. Plus, you are comparing speculation "we will do X in Y amount of time" to actual results.
And your point? I didn't simplify anything. I was responding to Booga's misleading post where he makes a claim about Intel and/or AMD doubling processor speed every 18 months. In reality, AMD has been stuck in the water on the PC front for nearly a year, resorting to rescaling their pseudohertz. Intel has been sitting pretty on AMD's inadequacies for about that long, delaying Prescott and being mum on Deerfield.
So what is your point? Are you betting Apple won't be at 3 GHz in about 12 months? I wouldn't be so quick to take that bet if I were you
What are the odds that IBM could make a transition of their POWER business to GPuL type chips?
mebbe not all of their business, but a large chunk of it?
If POWER chips cost lots and lots to make, mebbe they could make better margins using multiple GPuL chips in their place?
Apple seems to have a controller chip for single and duals, but I might be possible to have 4, 8, 16... GPuL's in a single box, a very expensive box, b ut no more than what a POWER based box costs you today.
That would certainly keep development of the G5/6/7... current.
As in blades?
OTOH PowerX and 9X0 processors are aimed at different markets. Big Iron on the one hand and extremely flexible and scalable server systems on the other. The line-up seems perfect to me and development isn't going to stop any time soon. It seems to me that when IBM commits it does so with no second thoughts.
And your point? I didn't simplify anything. I was responding to Booga's misleading post where he makes a claim about Intel and/or AMD doubling processor speed every 18 months. In reality, AMD has been stuck in the water on the PC front for nearly a year, resorting to rescaling their pseudohertz. Intel has been sitting pretty on AMD's inadequacies for about that long, delaying Prescott and being mum on Deerfield.
So what is your point? Are you betting Apple won't be at 3 GHz in about 12 months? I wouldn't be so quick to take that bet if I were you
Whoa, chill out there Eugene. You sound like you are about to burst a vein!
Let me break this down for you:
Booga did not make a statement about doubling processor speed. He made a statement about the rest of the industry doubling in performance every 18 months.
You misintepreted his statement regarding performance into a post about clockspeed and clockspeed alone (read your own post above). Anyone on this board should know that megahertz is not the only indicator of performance. Therefore, my follow up post noted that and showed that performance != clockspeed and cited some examples where that is the case.
While i is a fact that AMD has not been keeping with Intel of late, they have hardly been "stuck in the water" as you claimed . In fact, they went from the Thoroughbred 2200+ chip in June of last year to the Barton 3200+ in May of this year. If you think this is stuck in the water, what do you call IBM's promise of 50% increase? Also a simple rescaling?
Furthermore, you are calling their rated speed pseudohertz, Again, you are equating clock speed for performance. Numerous benchmarks on the PC side have shown that AMD comes close to (and sometimes bests) the performance of P4 while having lower clockspeed. I do not know why you of all people are stuck on this megahertz thing.
Based on that false premise, you concluded that the reason why Intel delayed Prescott and mum on Deerfield is because AMD has not been improving as fast. You have not established a causal relationship in this example either Unless you know something we don't, your guess on why Intel has delayed Prescott, is just as that, a guess that is accurate (or inaccurate) as anyone's on this board.
Lastly, where did you read that I am betting that IBM cannot reach a 50% increase in clockspeed in 12 months? Just because my correcting your simplification?
If you look at any benchmark, you'll see that the closest thing to doubling performance is clockspeed, especially vs old versions of the same chip. 800 MHz FSBs, memory bus improvements, etc. are beneficial, but relatively minor. Or do you really think the 2.2 GHz Athlon XP is twice as fast as the 1.67 GHz Athlon XP. Do you really think the 3.4 GHz P4 will be twice as fast as the 2.53 GHz P4? Pfft.
Given a 3 GHz G5 will have a 1.5 GHz FSB and have the same kind of component updates as its PC counterparts, it's all relative anyway. Your point is moot.
That's actually a good question. I suspect you'll see Apple move to a faster revision cycle with smaller bumps, more like the PC world uses, since it encourages purchases. When you know there is only going to be 2 updates a year there's far more chance you're going to wait for them.
My thinkin' also.
By the time the G5 ships...it mightn't be that long until he gets his 'Rev B'!
As for the 3 gig 970. That's at least a 6 gig G4 in overall performance by next year! That's a six gig Pentium! Will Intel be making 6 gig Prescotts next year? Dual it. You got a 12 gig G4!!! 12 gig Prescott? Any time this year, next or after?
IBM just grabbed Intel by the balls and squeezed.
Just the fact Apple/IBM announced this! Open with the road map! It's given the platform a road map. Confidence.
Is performance the same as clockspeed? Yes or no? The answer is no. Booga talked about performance. In your post, you change it to clockspeed and clockspeed alone. I merely pointed out 1) clockspeed does not equate performance and 2) you are wrong on a number of points.
First, you said AMD increased clockspeed by 32% since Jan 2002. In reality, AMD bumped its speed from 2000+ to 3200+ in that same time period - that's a 50% improvement. Intel released 2.2 ghz in Jan 2002 and now is at 3.2 ghz 18 months later. Again, a ~50% improvement as opposed to the 34% you have quoted. So, you are factually incorrect in your understanding of Intel and AMD. (For anyone who does not believe me, check http://www.sharkyextreme.com/hardware/cpu/index.php or numerous CPU review sites.)
You know, no big deal, we all make mistakes, let's move on.
But no, you do not admit that you have distorted the numbers. Instead, you accuse me of saying Apple won't reach 3 ghz in 12 months (I did not) and thinking that 3.4 GHz P4 will be twice as fast as the 2.53 GHz P4 (again, I don't know where you are getting this from).
You also flippilantly dismiss all this by simply saying my point is moot. Man, that was easy.
(Oh, and to save some of you some posting time - I am not a troll nor a Mac hater. I own a 2002 iBook and 20gb iPod. Now - that is a perfect moot point! As if whether or not I am a Mac user changes the fact that Eugene is wrong?!? Give me a break guys! )
********************
LBB: Comparing vaporware to vaporware gets no one nowhere.
Oh come on, everyone with a brain knows AMD's 3200+ rating is completely bogus. A 2.25 GHz chip minus 256k L2 cache and at 333 MHz FSB was a 2800+ not too long ago. I would venture to guess the performance increase is much much closer to 32% than 50%.
Show me another instace where the P4 made such a huge jump. You just happened to compare one of the last Willamettes with the latest Northwoods. Also, Intel *is* scaling back, if only because of AMD's inability to compete. So, Intel having introduced a 2.53 GHz Northwood in late May of last year should mean it should be well past 3.4 GHz this year. +50%/18 months is atypical, and the 970 going from 2 to 3 GHz should be much more than a 50% boost. Why is that so hard to admit?
Comments
Originally posted by JRG
To get much better performance at that clock speed they would need to adjust the architecture of the chip: give it larger caches (which I suspect they will...
I suspect they won't. IBM wants to keep an appreciable performance gap between the Power4 and the 970. Small caches make that happen.
Originally posted by synp
I suspect they won't. IBM wants to keep an appreciable performance gap between the Power4 and the 970. Small caches make that happen.
i dont think adding another meg or two fo cache to the 970 is scaring IBM...the Power4 has a sick amoutn of cache
IIRC, the 970 is currently manufactured on a 0.13u process. How soon before it goes to 0.09u and gains a low K dielectric bonus?
Anyone want to try overclocking their PowerMac G5 when they get it?
So maybe I should wait for Rev B or even C
You could do. But if you wait for the Rev B...next Spring...how long will it be before a '4 times as fast' 980 comes along to whoop ass?
I wonder. How many bumps will Apple do? Will they bump their machines more frequently now they've got somebody who knows how to make chips?
Or will we keep the 'top goes to bottom' model every 6 months?
If Apple ship G5 by August...we're looking at Rev B by Jan' 04 by this year's reckoning...
That won't be 3 gig in 12 months. It will be...? 2.5 do you think? On .13? Or .09?
Lemon Bon Bon
mebbe not all of their business, but a large chunk of it?
If POWER chips cost lots and lots to make, mebbe they could make better margins using multiple GPuL chips in their place?
Apple seems to have a controller chip for single and duals, but I might be possible to have 4, 8, 16... GPuL's in a single box, a very expensive box, b ut no more than what a POWER based box costs you today.
That would certainly keep development of the G5/6/7... current.
Originally posted by Matsu
What are the odds that IBM could make a transition of their POWER business to GPuL type chips?
mebbe not all of their business, but a large chunk of it?
If POWER chips cost lots and lots to make, mebbe they could make better margins using multiple GPuL chips in their place?
Apple seems to have a controller chip for single and duals, but I might be possible to have 4, 8, 16... GPuL's in a single box, a very expensive box, b ut no more than what a POWER based box costs you today.
That would certainly keep development of the G5/6/7... current.
POWER chips and the PPC 970 + successors will be aimed at different markets. The PPC 970 will really help IBM compete better with the Xeon + Opteron market while the POWER series continues to play higher up the food chain.
Don't expect either to disappear in the near future. Both have significant roadmaps ahead.
Originally posted by Lemon Bon Bon
I wonder. How many bumps will Apple do? Will they bump their machines more frequently now they've got somebody who knows how to make chips?
Or will we keep the 'top goes to bottom' model every 6 months?
That's actually a good question. I suspect you'll see Apple move to a faster revision cycle with smaller bumps, more like the PC world uses, since it encourages purchases. When you know there is only going to be 2 updates a year there's far more chance you're going to wait for them.
Originally posted by Eugene
It's not really set in stone, but that's about right.
And as far as jumping from 2 to 3 GHz (+50%) in 12 months is concerned...
Since AMD can't keep up, Intel has gotten lazy. They only plan on reaching 3.4 GHz by the end of the year...whereas they released 2.53 GHz P4s in May 2002. A 34% jump in ~18 months.
AMD has been even worse off. They 've increased clockspeeds by 32% since January 2002, going from 1.67 GHz to 2.2 GHz. They briefly had a 2.25 GHz chip but they couldn't make an appreciable number of them.
This is at best simplification and at worst plain distortion. Case in point: Pentium-M = same clock speed but faster. P4 800 FSB = same clock speed but faster. Plus, you are comparing speculation "we will do X in Y amount of time" to actual results.
Originally posted by Booga
Considering that the rest of the industry doubles performance every 18 months, a 50% increase in 12 months isn't exactly good news. Why the "wow"?
You're making the mistake of associating performance with MHz, that assumption may be false.
Originally posted by Matsu
What are the odds that IBM could make a transition of their POWER business to GPuL type chips?
mebbe not all of their business, but a large chunk of it?
If POWER chips cost lots and lots to make, mebbe they could make better margins using multiple GPuL chips in their place?
Apple seems to have a controller chip for single and duals, but I might be possible to have 4, 8, 16... GPuL's in a single box, a very expensive box, b ut no more than what a POWER based box costs you today.
That would certainly keep development of the G5/6/7... current.
IBM has been play'n with quads of the GUL since last July ... while Apple was play'n with dualies.
Originally posted by Telomar
POWER chips and the PPC 970 + successors will be aimed at different markets. The PPC 970 will really help IBM compete better with the Xeon + Opteron market while the POWER series continues to play higher up the food chain.
Don't expect either to disappear in the near future. Both have significant roadmaps ahead.
Tis very true. Made even better because Dell doesn't have 64bit processing on their "workstation roadmap" till next year (at least up until April 2004). Apple has an "opportunity", I hope they capitalize on it.
Originally posted by alcimedes
lol, perhaps to asuage the fears of
"We have the new, all powerful G4. it's shipping at 550Mhz!!!"
ok, make that 500Mhz. for hte next ****ING YEAR. (or was it longer?)
in any case, the G4 initially got stuck a LONG time at its initial speeds. this would be to remove that fear with this new chip, IMO.
Spot On!, This is WHY IBM has announced the chip will scale to 3GHz within 12mths. Also, its been reported that the 970 @ .13um process will indeed scale to 2.5GHz. So within 6 months we will see this speed and then 3GHz, with the speed bump beyond 3GHz could well be the 980 chip around early 2005.
The 970 by then will be @ 0.09um and employed into the iMacs/eMacs at the very least.
The point of announcing a 3GHz in 12mths, means we (as customers) can go out and invest in Apple (buy G5's) and its OS, knowing that the hardware will only get better! Its a bit of an enterprise thing as well - im thinking xserve customers too.
Originally posted by Telomar
That's actually a good question. I suspect you'll see Apple move to a faster revision cycle with smaller bumps, more like the PC world uses, since it encourages purchases. When you know there is only going to be 2 updates a year there's far more chance you're going to wait for them.
It has t do with a chip manufacturer, too. Who wants to use up your production capacity when the whole line can move up a step?
Originally posted by klinux
This is at best simplification and at worst plain distortion. Case in point: Pentium-M = same clock speed but faster. P4 800 FSB = same clock speed but faster. Plus, you are comparing speculation "we will do X in Y amount of time" to actual results.
And your point? I didn't simplify anything. I was responding to Booga's misleading post where he makes a claim about Intel and/or AMD doubling processor speed every 18 months. In reality, AMD has been stuck in the water on the PC front for nearly a year, resorting to rescaling their pseudohertz. Intel has been sitting pretty on AMD's inadequacies for about that long, delaying Prescott and being mum on Deerfield.
So what is your point? Are you betting Apple won't be at 3 GHz in about 12 months? I wouldn't be so quick to take that bet if I were you
Originally posted by Matsu
What are the odds that IBM could make a transition of their POWER business to GPuL type chips?
mebbe not all of their business, but a large chunk of it?
If POWER chips cost lots and lots to make, mebbe they could make better margins using multiple GPuL chips in their place?
Apple seems to have a controller chip for single and duals, but I might be possible to have 4, 8, 16... GPuL's in a single box, a very expensive box, b ut no more than what a POWER based box costs you today.
That would certainly keep development of the G5/6/7... current.
As in blades?
OTOH PowerX and 9X0 processors are aimed at different markets. Big Iron on the one hand and extremely flexible and scalable server systems on the other. The line-up seems perfect to me and development isn't going to stop any time soon. It seems to me that when IBM commits it does so with no second thoughts.
Originally posted by Eugene
And your point? I didn't simplify anything. I was responding to Booga's misleading post where he makes a claim about Intel and/or AMD doubling processor speed every 18 months. In reality, AMD has been stuck in the water on the PC front for nearly a year, resorting to rescaling their pseudohertz. Intel has been sitting pretty on AMD's inadequacies for about that long, delaying Prescott and being mum on Deerfield.
So what is your point? Are you betting Apple won't be at 3 GHz in about 12 months? I wouldn't be so quick to take that bet if I were you
Whoa, chill out there Eugene. You sound like you are about to burst a vein!
Let me break this down for you:
Booga did not make a statement about doubling processor speed. He made a statement about the rest of the industry doubling in performance every 18 months.
You misintepreted his statement regarding performance into a post about clockspeed and clockspeed alone (read your own post above). Anyone on this board should know that megahertz is not the only indicator of performance. Therefore, my follow up post noted that and showed that performance != clockspeed and cited some examples where that is the case.
While i is a fact that AMD has not been keeping with Intel of late, they have hardly been "stuck in the water" as you claimed . In fact, they went from the Thoroughbred 2200+ chip in June of last year to the Barton 3200+ in May of this year. If you think this is stuck in the water, what do you call IBM's promise of 50% increase? Also a simple rescaling?
Furthermore, you are calling their rated speed pseudohertz, Again, you are equating clock speed for performance. Numerous benchmarks on the PC side have shown that AMD comes close to (and sometimes bests) the performance of P4 while having lower clockspeed. I do not know why you of all people are stuck on this megahertz thing.
Based on that false premise, you concluded that the reason why Intel delayed Prescott and mum on Deerfield is because AMD has not been improving as fast. You have not established a causal relationship in this example either Unless you know something we don't, your guess on why Intel has delayed Prescott, is just as that, a guess that is accurate (or inaccurate) as anyone's on this board.
Lastly, where did you read that I am betting that IBM cannot reach a 50% increase in clockspeed in 12 months? Just because my correcting your simplification?
You gotta relax!
Given a 3 GHz G5 will have a 1.5 GHz FSB and have the same kind of component updates as its PC counterparts, it's all relative anyway. Your point is moot.
That's actually a good question. I suspect you'll see Apple move to a faster revision cycle with smaller bumps, more like the PC world uses, since it encourages purchases. When you know there is only going to be 2 updates a year there's far more chance you're going to wait for them.
My thinkin' also.
By the time the G5 ships...it mightn't be that long until he gets his 'Rev B'!
As for the 3 gig 970. That's at least a 6 gig G4 in overall performance by next year! That's a six gig Pentium! Will Intel be making 6 gig Prescotts next year? Dual it. You got a 12 gig G4!!! 12 gig Prescott? Any time this year, next or after?
IBM just grabbed Intel by the balls and squeezed.
Just the fact Apple/IBM announced this! Open with the road map! It's given the platform a road map. Confidence.
Expect Tower sales to rocket.
Mine will be in there!!! Meesa buyin'!!!
Lemon Bon Bon
Is performance the same as clockspeed? Yes or no? The answer is no. Booga talked about performance. In your post, you change it to clockspeed and clockspeed alone. I merely pointed out 1) clockspeed does not equate performance and 2) you are wrong on a number of points.
First, you said AMD increased clockspeed by 32% since Jan 2002. In reality, AMD bumped its speed from 2000+ to 3200+ in that same time period - that's a 50% improvement. Intel released 2.2 ghz in Jan 2002 and now is at 3.2 ghz 18 months later. Again, a ~50% improvement as opposed to the 34% you have quoted. So, you are factually incorrect in your understanding of Intel and AMD. (For anyone who does not believe me, check http://www.sharkyextreme.com/hardware/cpu/index.php or numerous CPU review sites.)
You know, no big deal, we all make mistakes, let's move on.
But no, you do not admit that you have distorted the numbers. Instead, you accuse me of saying Apple won't reach 3 ghz in 12 months (I did not) and thinking that 3.4 GHz P4 will be twice as fast as the 2.53 GHz P4 (again, I don't know where you are getting this from).
You also flippilantly dismiss all this by simply saying my point is moot. Man, that was easy.
(Oh, and to save some of you some posting time - I am not a troll nor a Mac hater. I own a 2002 iBook and 20gb iPod. Now - that is a perfect moot point! As if whether or not I am a Mac user changes the fact that Eugene is wrong?!? Give me a break guys! )
********************
LBB: Comparing vaporware to vaporware gets no one nowhere.
Show me another instace where the P4 made such a huge jump. You just happened to compare one of the last Willamettes with the latest Northwoods. Also, Intel *is* scaling back, if only because of AMD's inability to compete. So, Intel having introduced a 2.53 GHz Northwood in late May of last year should mean it should be well past 3.4 GHz this year. +50%/18 months is atypical, and the 970 going from 2 to 3 GHz should be much more than a 50% boost. Why is that so hard to admit?