<strong>I hope they do drop the duals. It was a poor strategy from the start, and just because they've had a poor strategy in the past doesn't mean they should continue with a poor strategy.
Duals are just not efficient for mainstream computers. They're great for niche-y situations, like people who render video all day, and should be an option for those who use many dual-optimized apps. But for most of us, they're just not worthwhile. A single fast processor makes much more sense.</strong><hr></blockquote>
I have to strongly disagree with this assertion. Multi-processing is the way of the future, initially as multiple single-core chips and eventually as one or more multi-core chips. We're going to max out how fast a single chip can run, and we're going to max out how complex a single chip can be and at that point the only option will be to go parallel. This day isn't far off, and the sooner the software goes multi-threaded, the better.
If Apple's first GPUL machine is single core, single processor that won't slow down the creation of multi-threaded software -- they have already sold many multi-processor machines and the existence of the POWER4 shows the likely direction for the future. I expect that Apple will offer a mixed lineup of 1, 2, and probably more processors... they'll finally have enough bandwidth to make >2 processors worthwhile.
<strong>The first statement only applies if 1) the app is dual aware or 2) you're running two compute-intensive apps at once. Unless one of those two conditions is met, you might as well have one processor rather than two. </strong><hr></blockquote>
Not necessarily -- the OS can use a fair amount of processing power by itself (I/O, timer interrupts, Aqua, networking, various daemons, etc) and if one processor can handle all of that while the second works on the compute task uninterrupted then there is a fair amount of potential gain (fewer context switches, keeping the cache pure, no switching address spaces, etc). Going forward more apps will be multi-threaded which lets them directly leverages the multiple processors.
Well, if Apple uses a 900Mhz bus, they almost have to use multiple processors. Here's why:
The processor speed has to be a multiplier of the bus speed, generally in 0.5 increments.
a 1.8Ghz 970 would be using a bus multiplier of 2 on a 900Mhz bus.
That is the fastest the chip will be. If you use a bus multiplier of 1.5, you end up with 1350Mhz. I don't know if Apple will even dare to ship a PowerMac at that clock speed by the time this chip is available. So, what will they have to do to have 3 power mac models?
low end: 1350
mid range: 1800
high end: dual 1800
- or -
low end: single 1800
mid range: dual 1800
high end: quad 1800
It's not like the bus won't be able to feed these chips, we're talking about a bus with almost 5x as much bandwidth as the current PowerMacs and probably less latency as well.
1. Reading some other boards makes you realize that the Wintel world is awfully defensive over the 970- of course they're normally idiots anyway. But, this one makes them especially nervous because they've enjoyed a performance edge for 2+ years. Now they're not really sure of the 970's performance or its actual ship date. It's an unknown that threatens their little world.
2. The contrast between the Apple/Mot and Apple/IBM partnership is telling. Mot bedded down with Apple because they wanted to produce a high performance low cost embedded chip. On this count they didn't do to bad.
IBM has teamed with Apple, because they wanted to produce a low cost server/desktop chip. I think Mot's motivation is pretty clear. IBM's OTOH is to produce an Itanium/Hammer killer possibly to protect their server market. They accomplish this by producing low end servers (<$20k?) which I'm guessing may be a server class eating its way into their high-end server business.
IBM probably expected Itanium and Hammer-once released and/or performing satisfactorily- to accelerate the erosion. Theymay have also needed a low cost high performance server box to sell to their Linux brothers. Why sell them a PIV box when they can sell a PPC Linux Box? And Apple needs a desktop CPU with a brighter future. The low power low-cost G4 just wasn't getting it done.
3. Since, Apple needs a CPU built for the desktop why not combine IBM's goals with Apple's? IBM gets economies of scale and Apple gets power. To me the question is really who got who's way- IBM or Apple? I truly believe that Altivec inclusion was a concession to Apple in order to sell them a chip IBM wanted to build for another market niche. Were their others? And what- if anything- did Apple concede?
4. I wonder if the Mac on Intel rumors we heard was simply Apple testing OSX on an Itanium, Hammer and 970 to judge which one displayed the greatest promise. And still I wonder what "options" SJ meant would appear after the first of the year? G5 or 970?
5. Tommorrow will be interesting to say the least. But I'm betting that if SPEC numbers are released they will be so far ahead of anything on the Wintel side that it will be a few years before they could even try to catch up.
The speed advantage came from video cards,ddr RAM,and a faster bus-it hasn't existed for 2 1/2 years.The current G4s are approximately as fast as the current P4s-it depends on what you are using the chips for-the G4s excel at highly parallel processes-that is what the chip is designed for.Gigaflops are the measure of chip speed how many times do people have to be told this? What is gignificant about the 970 is that it blows away Intel's chips-not plays catchup or bests it by a little.The 1.8 gHz 970,according to benchmarks,runs at a speed roughly equivalent to what a 3.6 gHz G4,or an 8+gHz P4.Apple should not abandon multiprocessors at all,they should ramp up their production.OSX is designed at its core for multiple processors.Altivec is one of the best aspects of the processor,not simply something that was added on.
<strong>A lot of the Power 4's amazing performance comes from the fact that it has 128 MB of L3 cache.... I'm postive this chip doesn't have anywhere near that..... so I would expect performance would be no where near comparable..... any word on how much cache these things will have?</strong><hr></blockquote>
True, it'll have no where near 128 MB of cache. But this chip has a 6.4 gigabytes per second interface to memory which should make up nicely for the smaller on die cache.
I'm all 4 keeping the dual processors, or any MP configurations for that matter.
It seems this processor is possibly the best suited MP processor for desktops that anyone has ever seen. IBM seems to think it's a good idea to use it as an MP processor.
Other than that has anyone noticed we are all unregistered again?
<strong>What is gignificant about the 970 is that it blows away Intel's chips-not plays catchup or bests it by a little. The 1.8 gHz 970, according to benchmarks,runs at a speed roughly equivalent to what a 3.6 gHz G4, or an 8+gHz P4. Apple should not abandon multiprocessors at all, they should ramp up their production. OSX is designed at its core for multiple processors. Altivec is one of the best aspects of the processor,not simply something that was added on.</strong><hr></blockquote>
Do you have a source for the benchmarks... Yea I know tommorow it should all come out anyway (or most of it) but I'd love to have something fun to read before I go to bed.
"The 1.8 gHz 970, according to benchmarks,runs at a speed roughly equivalent to what a 3.6 gHz G4, or an 8+gHz P4."
Hmmm if this is even close to true then things will sure look bright come 'sometime' in 03.
<strong>Well, if Apple uses a 900Mhz bus, they almost have to use multiple processors. Here's why:
The processor speed has to be a multiplier of the bus speed, generally in 0.5 increments.
a 1.8Ghz 970 would be using a bus multiplier of 2 on a 900Mhz bus.
That is the fastest the chip will be. If you use a bus multiplier of 1.5, you end up with 1350Mhz. I don't know if Apple will even dare to ship a PowerMac at that clock speed by the time this chip is available. So, what will they have to do to have 3 power mac models?
low end: 1350
mid range: 1800
high end: dual 1800
- or -
low end: single 1800
mid range: dual 1800
high end: quad 1800
It's not like the bus won't be able to feed these chips, we're talking about a bus with almost 5x as much bandwidth as the current PowerMacs and probably less latency as well.</strong><hr></blockquote>
If the 970 starts at 1.8ghz, why would IBM make an underclocked 1.35ghz? It starts at 1.8ghz, that doesn't mean they don't have faster chips, like 1.8ghz, 2.0ghz, etc?
If Apple does go with this chip, I think it makes sense for them to drop duals as default config. Their margins go up on their basic configurations and they can get premium pricing on the dualies or quads, if they choose to offer them as well. And I do think they should continue to offer duals (at least) as built to order options.
Single proc machines will probably still toast anything we can get from Apple today, no matter what we use them for, so value would at least be maintained for regular users. (SoHo, home users etc). But for the true power users, like video, scientific etc, Apple would be able to offer, at premium pricing, truely top of the line machines. If you want these big spenders to truly view your platform as a contenter, you have to differentiate your true professional offerings from you pro-sumer offerings. Til now, the Powermacs have offered what is effectively prosumer performance. Well, it's been said before, but it all those software purchases Apple was making finally really make sense.
<strong>The speed advantage came from video cards,ddr RAM,and a faster bus-it hasn't existed for 2 1/2 years.</strong><hr></blockquote>
I agree. But, like it or not, it's the perception due to the mHz gap not all the other goodies.
[quote]Originally posted by Rick1138:
<strong>Gigaflops are the measure of chip speed how many times do people have to be told this? .</strong><hr></blockquote>
Please...
Go over to the battlefront and you'll get a nicer reception to that comment- especially since Intel publicly disavows the legitimacy of MIPS in judging processor performance- just as we do for SPEC. Wintel users repeat it ad nauseum...and they say Mac users have a heard mentality!
[quote]Originally posted by Rick1138:
<strong> What is gignificant about the 970 is that it blows away Intel's chips-not plays catchup or bests it by a little.The 1.8 gHz 970,according to benchmarks,runs at a speed roughly equivalent to what a 3.6 gHz G4,or an 8+gHz P4..</strong><hr></blockquote>
Thanks. This is exactly the performance quantification I've been wanting to read. Was "gignificant" purposeful? If not, it's still a wonderfully appropriate wat to describe the 970.
If the 970 starts at 1.8ghz, why would IBM make an underclocked 1.35ghz? It starts at 1.8ghz, that doesn't mean they don't have faster chips, like 1.8ghz, 2.0ghz, etc?</strong><hr></blockquote>
According to the IBM press release, it'll start at speeds up to 1.8GHz. 1.35GHz could be the low end in that instance.
Why does the GP-UL (oops . . . PPC970) need to have a 900Mhz bus for all the speeds of the chip? I was imagining that the bus would scale with the chip . . . ie, 900Mhz bus, 1.8GHz chip; 1GHz bus, 2GHz chip. Is there something faulty with my reasoning, or is this feasable? This would allow for several closely spec-ed models, ie 1.6, 1.8, and 2.0 (or something like it)
[quote]Do you have a source for the benchmarks?<hr></blockquote>
I can't find the articles it was in but it was from one of IBM internal testers-he stated that the one gigahertz chips that were being tested now ran at approximately twice the speed of the one gigahertz G4,and that it would be expected to scale accordingly.
If you assume that the Wintel world will be moving over to the Itanium, it might interest you to know that the current speed of the Itanium 2 -- which scores quite well on the benchmarks, btw -- is a palty 1ghz.
The people in the Wintel world who have been using ghz ratings as a measure of their penis size are going to be in for a rude awakening when the Itanium is pushed as mainstream. They very well may be forced to admit that there is something to this whole "mhz myth" after all.
One other thing to keep in mind is that the GP-UL is not just a stripped down, single-core POWER4 processor -- it isn't valid to assume that the GP-UL will simply be a scaled down POWER4 from earlier this year, with no enhancements whatsoever.
Wait until you see the benchmarks from this processor; I think you may be quite impressed.
Wait until you see the benchmarks from this processor; I think you may be quite impressed.</strong><hr></blockquote>
Do indeed listen to the man.
Intel has been working hard for years to make the Itanium very fast, but IBM has been working harder for longer, as far as I'm concerned. IBM has been making supercomputers for a long time. They know how to push around big chunks of data fast.
And regardless of whether it's single core, I'm sure it can multiprocess quite well.
<strong>The processor speed has to be a multiplier of the bus speed, generally in 0.5 increments.
a 1.8Ghz 970 would be using a bus multiplier of 2 on a 900Mhz bus.
That is the fastest the chip will be. If you use a bus multiplier of 1.5, you end up with 1350Mhz. I don't know if Apple will even dare to ship a PowerMac at that clock speed by the time this chip is available.</strong><hr></blockquote>
Drawing conclusions from sketchy clues may lead you astray.*
This is a press release. It is written by marketing, not engineers.
While it says 900MHz, could not it be DDR 450MHz or QDR 225MHz?
There is no reason that multipliers HAVE to be in increments of 0.5. It's just nice for syncing data. 0.25 would probably be possible.
This could be the beginning of a new approach.
The multiplier of EVER chip is locked at 2x, and different bus speeds are used instead. IBM did say "up to" 900MHz in the press release.
It could be that the real bus is 450MHz, with multiplier increments of 0.25x and or different busses are used.
If the speculation that the PowerPC 970 uses ApplePI is true, then you can bet your arse that the solution will be a neat one.
----------------
The bus speed is 900MHz, yet data transfer is 6.4GB/s. Sounds like 1Hz in every 9 is used to address the data, and the other 8Hz is the actual data. This is how I got my 750MHz bus, which is 1 and 1/8th of 666MHz.
----------------
BRussell, as Programmer said, you're mistaken.
----------------
Barto
*Obscure quote. I doubt anyone will know it <img src="confused.gif" border="0">
Comments
<strong>I hope they do drop the duals. It was a poor strategy from the start, and just because they've had a poor strategy in the past doesn't mean they should continue with a poor strategy.
Duals are just not efficient for mainstream computers. They're great for niche-y situations, like people who render video all day, and should be an option for those who use many dual-optimized apps. But for most of us, they're just not worthwhile. A single fast processor makes much more sense.</strong><hr></blockquote>
I have to strongly disagree with this assertion. Multi-processing is the way of the future, initially as multiple single-core chips and eventually as one or more multi-core chips. We're going to max out how fast a single chip can run, and we're going to max out how complex a single chip can be and at that point the only option will be to go parallel. This day isn't far off, and the sooner the software goes multi-threaded, the better.
If Apple's first GPUL machine is single core, single processor that won't slow down the creation of multi-threaded software -- they have already sold many multi-processor machines and the existence of the POWER4 shows the likely direction for the future. I expect that Apple will offer a mixed lineup of 1, 2, and probably more processors... they'll finally have enough bandwidth to make >2 processors worthwhile.
<strong>The first statement only applies if 1) the app is dual aware or 2) you're running two compute-intensive apps at once. Unless one of those two conditions is met, you might as well have one processor rather than two. </strong><hr></blockquote>
Not necessarily -- the OS can use a fair amount of processing power by itself (I/O, timer interrupts, Aqua, networking, various daemons, etc) and if one processor can handle all of that while the second works on the compute task uninterrupted then there is a fair amount of potential gain (fewer context switches, keeping the cache pure, no switching address spaces, etc). Going forward more apps will be multi-threaded which lets them directly leverages the multiple processors.
The processor speed has to be a multiplier of the bus speed, generally in 0.5 increments.
a 1.8Ghz 970 would be using a bus multiplier of 2 on a 900Mhz bus.
That is the fastest the chip will be. If you use a bus multiplier of 1.5, you end up with 1350Mhz. I don't know if Apple will even dare to ship a PowerMac at that clock speed by the time this chip is available. So, what will they have to do to have 3 power mac models?
low end: 1350
mid range: 1800
high end: dual 1800
- or -
low end: single 1800
mid range: dual 1800
high end: quad 1800
It's not like the bus won't be able to feed these chips, we're talking about a bus with almost 5x as much bandwidth as the current PowerMacs and probably less latency as well.
1. Reading some other boards makes you realize that the Wintel world is awfully defensive over the 970- of course they're normally idiots anyway. But, this one makes them especially nervous because they've enjoyed a performance edge for 2+ years. Now they're not really sure of the 970's performance or its actual ship date. It's an unknown that threatens their little world.
2. The contrast between the Apple/Mot and Apple/IBM partnership is telling. Mot bedded down with Apple because they wanted to produce a high performance low cost embedded chip. On this count they didn't do to bad.
IBM has teamed with Apple, because they wanted to produce a low cost server/desktop chip. I think Mot's motivation is pretty clear. IBM's OTOH is to produce an Itanium/Hammer killer possibly to protect their server market. They accomplish this by producing low end servers (<$20k?) which I'm guessing may be a server class eating its way into their high-end server business.
IBM probably expected Itanium and Hammer-once released and/or performing satisfactorily- to accelerate the erosion. Theymay have also needed a low cost high performance server box to sell to their Linux brothers. Why sell them a PIV box when they can sell a PPC Linux Box? And Apple needs a desktop CPU with a brighter future. The low power low-cost G4 just wasn't getting it done.
3. Since, Apple needs a CPU built for the desktop why not combine IBM's goals with Apple's? IBM gets economies of scale and Apple gets power. To me the question is really who got who's way- IBM or Apple? I truly believe that Altivec inclusion was a concession to Apple in order to sell them a chip IBM wanted to build for another market niche. Were their others? And what- if anything- did Apple concede?
4. I wonder if the Mac on Intel rumors we heard was simply Apple testing OSX on an Itanium, Hammer and 970 to judge which one displayed the greatest promise. And still I wonder what "options" SJ meant would appear after the first of the year? G5 or 970?
5. Tommorrow will be interesting to say the least. But I'm betting that if SPEC numbers are released they will be so far ahead of anything on the Wintel side that it will be a few years before they could even try to catch up.
<strong>A lot of the Power 4's amazing performance comes from the fact that it has 128 MB of L3 cache.... I'm postive this chip doesn't have anywhere near that..... so I would expect performance would be no where near comparable..... any word on how much cache these things will have?</strong><hr></blockquote>
True, it'll have no where near 128 MB of cache. But this chip has a 6.4 gigabytes per second interface to memory which should make up nicely for the smaller on die cache.
w00tz0r 16-CPU workstation!
<hr></blockquote>
216 gigaflops,cool.
It seems this processor is possibly the best suited MP processor for desktops that anyone has ever seen. IBM seems to think it's a good idea to use it as an MP processor.
Other than that has anyone noticed we are all unregistered again?
<strong>What is gignificant about the 970 is that it blows away Intel's chips-not plays catchup or bests it by a little. The 1.8 gHz 970, according to benchmarks,runs at a speed roughly equivalent to what a 3.6 gHz G4, or an 8+gHz P4. Apple should not abandon multiprocessors at all, they should ramp up their production. OSX is designed at its core for multiple processors. Altivec is one of the best aspects of the processor,not simply something that was added on.</strong><hr></blockquote>
Do you have a source for the benchmarks... Yea I know tommorow it should all come out anyway (or most of it) but I'd love to have something fun to read before I go to bed.
"The 1.8 gHz 970, according to benchmarks,runs at a speed roughly equivalent to what a 3.6 gHz G4, or an 8+gHz P4."
Hmmm if this is even close to true then things will sure look bright come 'sometime' in 03.
Dave
<strong>Well, if Apple uses a 900Mhz bus, they almost have to use multiple processors. Here's why:
The processor speed has to be a multiplier of the bus speed, generally in 0.5 increments.
a 1.8Ghz 970 would be using a bus multiplier of 2 on a 900Mhz bus.
That is the fastest the chip will be. If you use a bus multiplier of 1.5, you end up with 1350Mhz. I don't know if Apple will even dare to ship a PowerMac at that clock speed by the time this chip is available. So, what will they have to do to have 3 power mac models?
low end: 1350
mid range: 1800
high end: dual 1800
- or -
low end: single 1800
mid range: dual 1800
high end: quad 1800
It's not like the bus won't be able to feed these chips, we're talking about a bus with almost 5x as much bandwidth as the current PowerMacs and probably less latency as well.</strong><hr></blockquote>
If the 970 starts at 1.8ghz, why would IBM make an underclocked 1.35ghz? It starts at 1.8ghz, that doesn't mean they don't have faster chips, like 1.8ghz, 2.0ghz, etc?
Single proc machines will probably still toast anything we can get from Apple today, no matter what we use them for, so value would at least be maintained for regular users. (SoHo, home users etc). But for the true power users, like video, scientific etc, Apple would be able to offer, at premium pricing, truely top of the line machines. If you want these big spenders to truly view your platform as a contenter, you have to differentiate your true professional offerings from you pro-sumer offerings. Til now, the Powermacs have offered what is effectively prosumer performance. Well, it's been said before, but it all those software purchases Apple was making finally really make sense.
<strong>The speed advantage came from video cards,ddr RAM,and a faster bus-it hasn't existed for 2 1/2 years.</strong><hr></blockquote>
I agree. But, like it or not, it's the perception due to the mHz gap not all the other goodies.
[quote]Originally posted by Rick1138:
<strong>Gigaflops are the measure of chip speed how many times do people have to be told this? .</strong><hr></blockquote>
Please...
Go over to the battlefront and you'll get a nicer reception to that comment- especially since Intel publicly disavows the legitimacy of MIPS in judging processor performance- just as we do for SPEC. Wintel users repeat it ad nauseum...and they say Mac users have a heard mentality!
[quote]Originally posted by Rick1138:
<strong> What is gignificant about the 970 is that it blows away Intel's chips-not plays catchup or bests it by a little.The 1.8 gHz 970,according to benchmarks,runs at a speed roughly equivalent to what a 3.6 gHz G4,or an 8+gHz P4..</strong><hr></blockquote>
Thanks. This is exactly the performance quantification I've been wanting to read. Was "gignificant" purposeful? If not, it's still a wonderfully appropriate wat to describe the 970.
<strong>
If the 970 starts at 1.8ghz, why would IBM make an underclocked 1.35ghz? It starts at 1.8ghz, that doesn't mean they don't have faster chips, like 1.8ghz, 2.0ghz, etc?</strong><hr></blockquote>
According to the IBM press release, it'll start at speeds up to 1.8GHz. 1.35GHz could be the low end in that instance.
I can't find the articles it was in but it was from one of IBM internal testers-he stated that the one gigahertz chips that were being tested now ran at approximately twice the speed of the one gigahertz G4,and that it would be expected to scale accordingly.
The people in the Wintel world who have been using ghz ratings as a measure of their penis size are going to be in for a rude awakening when the Itanium is pushed as mainstream. They very well may be forced to admit that there is something to this whole "mhz myth" after all.
One other thing to keep in mind is that the GP-UL is not just a stripped down, single-core POWER4 processor -- it isn't valid to assume that the GP-UL will simply be a scaled down POWER4 from earlier this year, with no enhancements whatsoever.
Wait until you see the benchmarks from this processor; I think you may be quite impressed.
<strong>
Wait until you see the benchmarks from this processor; I think you may be quite impressed.</strong><hr></blockquote>
Do indeed listen to the man.
Intel has been working hard for years to make the Itanium very fast, but IBM has been working harder for longer, as far as I'm concerned. IBM has been making supercomputers for a long time. They know how to push around big chunks of data fast.
And regardless of whether it's single core, I'm sure it can multiprocess quite well.
<strong>The processor speed has to be a multiplier of the bus speed, generally in 0.5 increments.
a 1.8Ghz 970 would be using a bus multiplier of 2 on a 900Mhz bus.
That is the fastest the chip will be. If you use a bus multiplier of 1.5, you end up with 1350Mhz. I don't know if Apple will even dare to ship a PowerMac at that clock speed by the time this chip is available.</strong><hr></blockquote>
Drawing conclusions from sketchy clues may lead you astray.*
This is a press release. It is written by marketing, not engineers.
While it says 900MHz, could not it be DDR 450MHz or QDR 225MHz?
There is no reason that multipliers HAVE to be in increments of 0.5. It's just nice for syncing data. 0.25 would probably be possible.
This could be the beginning of a new approach.
The multiplier of EVER chip is locked at 2x, and different bus speeds are used instead. IBM did say "up to" 900MHz in the press release.
1.8GHz CPU, 900MHz bus, 6.4GB/s, dual-channel DDR 400.
1.5GHz CPU, 750MHz Bus, 5.3GB/s, dual-channel DDR 333
It could be that the real bus is 450MHz, with multiplier increments of 0.25x and or different busses are used.
If the speculation that the PowerPC 970 uses ApplePI is true, then you can bet your arse that the solution will be a neat one.
----------------
The bus speed is 900MHz, yet data transfer is 6.4GB/s. Sounds like 1Hz in every 9 is used to address the data, and the other 8Hz is the actual data. This is how I got my 750MHz bus, which is 1 and 1/8th of 666MHz.
----------------
BRussell, as Programmer said, you're mistaken.
----------------
Barto
*Obscure quote. I doubt anyone will know it <img src="confused.gif" border="0">