IBMs Dave Siegel has last week at the International Supercomputer Conference confirmed that the POWER5 test systems indeed quadruples the speed of the POWER4 (when it was introduced). He named especially SPEC, which would mean 2500 SPECint and 3500 SPECfp (POWER5 is expected to run up to 2GHz)
Quote:
RBR
The 980 is the "real deal" even though the 970 is not too shabby
Seems the the 970 is shabby. And the Pentium4, Itanium2 and Opteron too.
I guess I want to contribute my little thoughts on why Apple needs to offer faster chips in the same cycle as Intel and AMD.
The chip market I think is far different than what it was when Apple was last competitive. Clearly Apple is going to have to be willing to implement better chips as fast as Intel and AMD, in that this is how the war is waged these days. You'll notice, only when AMD can't get that chip smaller/faster, do both AMD and Intel begin to play with the bus and memory. AMD because it's the only place they can milk added performance and Intel because it's better to save truly smaller or/and faster for the next true battle. If AMD is scaling clock speeds and die shrinks, then that is where they both fight the battle. As IBM/Apple enter the race with the 970, Intel find's itself flanked on all sides. I could be wrong, but I would suspect that a 970 fabbed at a smaller process than current, would be pointed by IBM directly at the Xeon space, until the advent of the Power5 which I believe they intend to run the gammit.
The other things that Apple has to consider are the needs of the 64bit workstation market. Right now, I would suppose that while Apple wants to target the Wintel market, we know that the Unix/Linux crowd, which IBM will be allegedly targeting with their Linux on 970 is probably more open to Apple, in that they have the ability to run a Unix with the productivity of MS Office and whatever else on the same machine. However, once MS offers the rumored 32-64 bit Windows for the Opteron/Athlon64 or whatever they're going to call it, you have all the lures of Apple's G5 platform sans the elegant interface that is OS X, but with the addition of the kick assed performance of the Opteron/Athlon64.
I would hazard, 970/"980" above, on par or just below AMD's Opteron/Athlon64 Apple gets the sale if productivity is a big consideration in addition to performance. Especially if Intel ends up in third. With AMD and Apple both using Hypertransport, AMD will certainly be able to derive similar gains in throughput and what not. If Apple remains in third, you're only buying an Apple workstation for OS X, because you can could get MS Office on the AMD workstation with the Intel beating performance.
Also, I think Intel has something else in their favor that they may play later. As many don't appear to know, the Pentium M, which is based on the PIII core, (far more efficient than the P4 but lost out from it's time in the sun due to internal politics (from what I've read)) is quite a little monster. At 1.6 or 1.8 GHz I believe or somewhere thereabouts it does pretty well against the P4 at 2.7 GHz or so. This is why some Intelheads are trying to get the Pentium M released for the desktop. While Intel may not be throwing their weight behind this chip publicly, there is no doubt in my mind that there is a lot to learn from it in the labs.
Now Apple might be able to beat the other platforms on price. Don't jump on me because I am just speculating on market forces. Programmer/Amorph or someone might tell me better, but does Unix/Linux on RISC tend to be a better performer barring much higher clock speed on the CISC side. Could be a stupid question. Sorry if is. I know that CISC and RISC these days are not nearly as cut and dry as before.
Whatever the case, I think Apple needs to shrink and speed as much as IBM allows them to. I'm sure this might keep IBM happier too. Especially if Apple is going to be staying with the XServe and please oh please the xU factored "XStation". Once the chips are tried and true why not. Let us not forget than Sun will be a big part of this equation. They must be working on something. While some of you may be rolling your eyes, remember how many times Apple's and AMD's demise have been forcast. IBM and AMD would do well to get SUN and SGI to consider their wares. This is not much different than everyone and their momma hawking various solutions of Intel kit. It builds brand. Linux, Solaris, OS X on the VENERABLE IBM 9XX. WHY NOT!
On the topic of AMD: Don't forget that the current AMD Athlon XP 3200+ is really just a 2200MHz chip. The 2GHz G5 however really is a 2GHz G5. The technical difference in MHz is a mere 200MHz now between AMD and IBM and may already be won for IBM.
In terms of MHz, the Athlon has scaled about 300MHz over the last 1.5 years, only the marketing numbering scheme has exploded (and become increasingly inaccurate too).
AMD is not going to die anytime soon, but their fate really depends on the success of the Opteron and on the release date and success of the Athlon64. Should these two fail, we might as well see IBM take the place of Intels largest competitor (again).
Don't forget: PC fans are touting the Opteron to be the beat-all chip already...Few of them consider that that chip is VASTLY more expensive than the Athlon, it's really the server chip, the AMD Xeon, so to say. The Athlon64 is where the desktop market really lies, and if that chip fails, AMD has a huge problem in the consumer desktop market.
Yeah, I hear ya, and I agree. I'm not saying that IBM won't go that way (dual cores), I was just bringing up a potenial issue. But you may be right, the memory bandwidth may be more of a limiting factor than the two cores sharing on bus would be.
I don't that that we will ever see a PM with more than two packaged processors (meaning a dual dualcore I could see).
Hey T.O., I definitely agree ... If Apple goes quad, then they'll most likely do it by "Dual Duals". The shared bus thing is something I didn't think of. They'd most likely have to implement IBM's CoreConnect technology within the chip, and I wonder if that would help the performance bottleneck.
I'm waiting for the 990. Now there's a machine you can apply a Guassian Blur with
Heh, I'm beginning to regret using the term "interrim". I was thinking that the 970+ didn't count as a 970 (but after more thought ... I was wrong, it is, and will make the 970 last a long time).
You'll be waiting a while for the 990 (if that's what they will call it). My suggestion is to wait, only long enough, to test drive a 970 based Mac. I think you'll succumb to the temptation (at least to lust after, maybe even buy one).
On the topic of AMD: Don't forget that the current AMD Athlon XP 3200+ is really just a 2200MHz chip. The 2GHz G5 however really is a 2GHz G5. The technical difference in MHz is a mere 200MHz now between AMD and IBM and may already be won for IBM.
In terms of MHz, the Athlon has scaled about 300MHz over the last 1.5 years, only the marketing numbering scheme has exploded (and become increasingly inaccurate too).
AMD is not going to die anytime soon, but their fate really depends on the success of the Opteron and on the release date and success of the Athlon64. Should these two fail, we might as well see IBM take the place of Intels largest competitor (again).
Don't forget: PC fans are touting the Opteron to be the beat-all chip already...Few of them consider that that chip is VASTLY more expensive than the Athlon, it's really the server chip, the AMD Xeon, so to say. The Athlon64 is where the desktop market really lies, and if that chip fails, AMD has a huge problem in the consumer desktop market.
Isn't the Athlon64 just a 64-bit enabled Athlon, with an on-die memory controller (oh, and that heatspreader )? If so, AMD REALLY needs to advance quickly with their fab tech, else the Prescott on 90nm will be a real, bad problem for AMD in the desktop space. Let's hope it's cheap at least
Going from 2 to 3 Ghz in 12 month sounds impressive - unless you consider that Moores law predicts a doubling of transistor density (translating in a doubling of chip speed) every 18 month.
IBM does not seem to be able to achieve this. With their predicted rate, they only gain 75% over the next 18 month. As far as I know, intel cpus had seen an even faster gain in clockspeed over the last years.
You idiot. If you assume a doubling every 18 months, then you're looking at an exponential distribution of the form 2^(t/18). Thus, in 12 months, the G5 should be at 3.17 GHz - pretty darn close to the quoted 3GHz figure.
This isn't a case of linear change over an 18 month period; it's an exponential distribution.
Actually, make that about one year ago (as far as Apple messing with 970 singles and duals). There were initial problems w/ Apple PI. IBM was none too happy about it either.
Jobs: Hi, Steve here. Listen, sorry about the business with 'pie, but you've got to understand that we're not used to working with CPU suppliers that don't miss deadlines and make ridiculous promises about what they can offer. When you guys go and exceed projected deadlines and specs, our engineers get confused.
IBM Dude: Just be sure you have those mobos ready to go when Neo ships, otherwise we may have to reconsider our business plan with Apple computer.
Both hang up. Jobs throws the phone across the room, picks it up, and dials human resources.
Jobs: I want a list of all the engineers working on Project Neo! I'll be meeting in the elevator with no fewer than 10 tomorrow morning!
Jobs: Hi, Steve here. Listen, sorry about the business with 'pie, but you've got to understand that we're not used to working with CPU suppliers that don't miss deadlines and make ridiculous promises about what they can offer. When you guys go and exceed projected deadlines and specs, our engineers get confused.
IBM Dude: Just be sure you have those mobos ready to go when Neo ships, otherwise we may have to reconsider our business plan with Apple computer.
Both hang up. Jobs throws the phone across the room, picks it up, and dials human resources.
Jobs: I want a list of all the engineers working on Project Neo! I'll be meeting in the elevator with no fewer than 10 tomorrow morning!
? Reduced memory latency (= on board memory controller, most likey)
They then go onto talk a lot more about multithreading on pp21-22. So it seems to me that the next major revision to the 9xx, after the bump to 90nm, will definitely be based on the Power5. We know the Power5 is coming early next year (1H) and IBM have stated that it will scale up and down their range better than the Power4 so I would not be surprised to hear at this year's MPF that they will be releasing the 980 in 2H04.
Also people, going up to 3GHz on the 970 in 12 months will be phenomenal, as some have suggested. IBM was conservative with "1.4-1.8 GHz" at release. Steve Jobs is not going to shoot himself in the foot by saying 3GHz and then only delivering 2.8GHz. He too was being conservative So I would fully expect to see a lineup that included 3.2 or 3.3 at the top end at WWDC04.
Finally, I wish they would have both singles and duals at all speed grades so as to let people have the option of a SP 2GHz or a DP 1.6 GHz if it suited their needs.
Just wait'll each core is threaded! Then you'll have two processors with two cores each, with each core pretending to be two CPUs.
There are your octos, more or less, and Apple wouldn't even have to change their current system controller!
Pshawwwww, get real, I bet most of youze guys still think Apple's gonna announce the G5 at WWDC, as if THAT'S ever going to happen, it's a developers conference F.C.S. people!
Comments
Originally posted by Transcendental Octothorpe
Indeed, but that wouldn't be a PM, now would it?
You are right, of course. That would be a POWER Mac!
RBR
The 980 is the "real deal" even though the 970 is not too shabby
Seems the the 970 is shabby. And the Pentium4, Itanium2 and Opteron too.
The chip market I think is far different than what it was when Apple was last competitive. Clearly Apple is going to have to be willing to implement better chips as fast as Intel and AMD, in that this is how the war is waged these days. You'll notice, only when AMD can't get that chip smaller/faster, do both AMD and Intel begin to play with the bus and memory. AMD because it's the only place they can milk added performance and Intel because it's better to save truly smaller or/and faster for the next true battle. If AMD is scaling clock speeds and die shrinks, then that is where they both fight the battle. As IBM/Apple enter the race with the 970, Intel find's itself flanked on all sides. I could be wrong, but I would suspect that a 970 fabbed at a smaller process than current, would be pointed by IBM directly at the Xeon space, until the advent of the Power5 which I believe they intend to run the gammit.
The other things that Apple has to consider are the needs of the 64bit workstation market. Right now, I would suppose that while Apple wants to target the Wintel market, we know that the Unix/Linux crowd, which IBM will be allegedly targeting with their Linux on 970 is probably more open to Apple, in that they have the ability to run a Unix with the productivity of MS Office and whatever else on the same machine. However, once MS offers the rumored 32-64 bit Windows for the Opteron/Athlon64 or whatever they're going to call it, you have all the lures of Apple's G5 platform sans the elegant interface that is OS X, but with the addition of the kick assed performance of the Opteron/Athlon64.
I would hazard, 970/"980" above, on par or just below AMD's Opteron/Athlon64 Apple gets the sale if productivity is a big consideration in addition to performance. Especially if Intel ends up in third. With AMD and Apple both using Hypertransport, AMD will certainly be able to derive similar gains in throughput and what not. If Apple remains in third, you're only buying an Apple workstation for OS X, because you can could get MS Office on the AMD workstation with the Intel beating performance.
Also, I think Intel has something else in their favor that they may play later. As many don't appear to know, the Pentium M, which is based on the PIII core, (far more efficient than the P4 but lost out from it's time in the sun due to internal politics (from what I've read)) is quite a little monster. At 1.6 or 1.8 GHz I believe or somewhere thereabouts it does pretty well against the P4 at 2.7 GHz or so. This is why some Intelheads are trying to get the Pentium M released for the desktop. While Intel may not be throwing their weight behind this chip publicly, there is no doubt in my mind that there is a lot to learn from it in the labs.
Now Apple might be able to beat the other platforms on price. Don't jump on me because I am just speculating on market forces. Programmer/Amorph or someone might tell me better, but does Unix/Linux on RISC tend to be a better performer barring much higher clock speed on the CISC side. Could be a stupid question. Sorry if is. I know that CISC and RISC these days are not nearly as cut and dry as before.
Whatever the case, I think Apple needs to shrink and speed as much as IBM allows them to. I'm sure this might keep IBM happier too. Especially if Apple is going to be staying with the XServe and please oh please the xU factored "XStation". Once the chips are tried and true why not. Let us not forget than Sun will be a big part of this equation. They must be working on something. While some of you may be rolling your eyes, remember how many times Apple's and AMD's demise have been forcast. IBM and AMD would do well to get SUN and SGI to consider their wares. This is not much different than everyone and their momma hawking various solutions of Intel kit. It builds brand. Linux, Solaris, OS X on the VENERABLE IBM 9XX. WHY NOT!
In terms of MHz, the Athlon has scaled about 300MHz over the last 1.5 years, only the marketing numbering scheme has exploded (and become increasingly inaccurate too).
AMD is not going to die anytime soon, but their fate really depends on the success of the Opteron and on the release date and success of the Athlon64. Should these two fail, we might as well see IBM take the place of Intels largest competitor (again).
Don't forget: PC fans are touting the Opteron to be the beat-all chip already...Few of them consider that that chip is VASTLY more expensive than the Athlon, it's really the server chip, the AMD Xeon, so to say. The Athlon64 is where the desktop market really lies, and if that chip fails, AMD has a huge problem in the consumer desktop market.
Originally posted by MacJedai
... like was said before, the 970 is an interrim chip, to really see the "magic" wait till you see the 980 in action.
Originally posted by MacJedai
... like was said before, the 970 is an interrim chip, to really see the "magic" wait till you see the 980 in action.
Originally posted by Transcendental Octothorpe
Yeah, I hear ya, and I agree. I'm not saying that IBM won't go that way (dual cores), I was just bringing up a potenial issue. But you may be right, the memory bandwidth may be more of a limiting factor than the two cores sharing on bus would be.
I don't that that we will ever see a PM with more than two packaged processors (meaning a dual dualcore I could see).
Hey T.O., I definitely agree ... If Apple goes quad, then they'll most likely do it by "Dual Duals". The shared bus thing is something I didn't think of. They'd most likely have to implement IBM's CoreConnect technology within the chip, and I wonder if that would help the performance bottleneck.
Originally posted by bluespark
I'm waiting for the 990. Now there's a machine you can apply a Guassian Blur with
Heh, I'm beginning to regret using the term "interrim". I was thinking that the 970+ didn't count as a 970 (but after more thought ... I was wrong, it is, and will make the 970 last a long time).
You'll be waiting a while for the 990 (if that's what they will call it). My suggestion is to wait, only long enough, to test drive a 970 based Mac. I think you'll succumb to the temptation (at least to lust after, maybe even buy one).
Originally posted by G-News
On the topic of AMD: Don't forget that the current AMD Athlon XP 3200+ is really just a 2200MHz chip. The 2GHz G5 however really is a 2GHz G5. The technical difference in MHz is a mere 200MHz now between AMD and IBM and may already be won for IBM.
In terms of MHz, the Athlon has scaled about 300MHz over the last 1.5 years, only the marketing numbering scheme has exploded (and become increasingly inaccurate too).
AMD is not going to die anytime soon, but their fate really depends on the success of the Opteron and on the release date and success of the Athlon64. Should these two fail, we might as well see IBM take the place of Intels largest competitor (again).
Don't forget: PC fans are touting the Opteron to be the beat-all chip already...Few of them consider that that chip is VASTLY more expensive than the Athlon, it's really the server chip, the AMD Xeon, so to say. The Athlon64 is where the desktop market really lies, and if that chip fails, AMD has a huge problem in the consumer desktop market.
Isn't the Athlon64 just a 64-bit enabled Athlon, with an on-die memory controller (oh, and that heatspreader
Originally posted by Smircle
Going from 2 to 3 Ghz in 12 month sounds impressive - unless you consider that Moores law predicts a doubling of transistor density (translating in a doubling of chip speed) every 18 month.
IBM does not seem to be able to achieve this. With their predicted rate, they only gain 75% over the next 18 month. As far as I know, intel cpus had seen an even faster gain in clockspeed over the last years.
You idiot. If you assume a doubling every 18 months, then you're looking at an exponential distribution of the form 2^(t/18). Thus, in 12 months, the G5 should be at 3.17 GHz - pretty darn close to the quoted 3GHz figure.
This isn't a case of linear change over an 18 month period; it's an exponential distribution.
Originally posted by MacJedai
Actually, make that about one year ago (as far as Apple messing with 970 singles and duals). There were initial problems w/ Apple PI. IBM was none too happy about it either.
Jobs: Hi, Steve here. Listen, sorry about the business with 'pie, but you've got to understand that we're not used to working with CPU suppliers that don't miss deadlines and make ridiculous promises about what they can offer. When you guys go and exceed projected deadlines and specs, our engineers get confused.
IBM Dude: Just be sure you have those mobos ready to go when Neo ships, otherwise we may have to reconsider our business plan with Apple computer.
Both hang up. Jobs throws the phone across the room, picks it up, and dials human resources.
Jobs: I want a list of all the engineers working on Project Neo! I'll be meeting in the elevator with no fewer than 10 tomorrow morning!
Originally posted by Junkyard Dawg
Jobs: Hi, Steve here. Listen, sorry about the business with 'pie, but you've got to understand that we're not used to working with CPU suppliers that don't miss deadlines and make ridiculous promises about what they can offer. When you guys go and exceed projected deadlines and specs, our engineers get confused.
IBM Dude: Just be sure you have those mobos ready to go when Neo ships, otherwise we may have to reconsider our business plan with Apple computer.
Both hang up. Jobs throws the phone across the room, picks it up, and dials human resources.
Jobs: I want a list of all the engineers working on Project Neo! I'll be meeting in the elevator with no fewer than 10 tomorrow morning!
LOL
https://www-914.ibm.com/events%5Cmic...s?OpenDocument
"Future Directions, 64-bit Multi-Gigahertz Embedded
PowerPC's" by Jim Rogers
In it is the following:
Looking Into the Future ? Industry Trends (p20)
Performance:
? Thread level parallelism (= Simultaneous Multithreading, p21)
? Multi-core processors
? Reduced memory latency (= on board memory controller, most likey)
They then go onto talk a lot more about multithreading on pp21-22. So it seems to me that the next major revision to the 9xx, after the bump to 90nm, will definitely be based on the Power5. We know the Power5 is coming early next year (1H) and IBM have stated that it will scale up and down their range better than the Power4 so I would not be surprised to hear at this year's MPF that they will be releasing the 980 in 2H04.
Also people, going up to 3GHz on the 970 in 12 months will be phenomenal, as some have suggested. IBM was conservative with "1.4-1.8 GHz" at release. Steve Jobs is not going to shoot himself in the foot by saying 3GHz and then only delivering 2.8GHz. He too was being conservative
Finally, I wish they would have both singles and duals at all speed grades so as to let people have the option of a SP 2GHz or a DP 1.6 GHz if it suited their needs.
Cheers
MM
Originally posted by Amorph
Just wait'll each core is threaded! Then you'll have two processors with two cores each, with each core pretending to be two CPUs.
There are your octos, more or less, and Apple wouldn't even have to change their current system controller!
Pshawwwww, get real, I bet most of youze guys still think Apple's gonna announce the G5 at WWDC, as if THAT'S ever going to happen, it's a developers conference F.C.S. people!
</fashionable skepticism flashback>