The delta in the number of transistors between a 32-bit 970 and a 64-bit 970 would be very small.... (maybe double to ~4 million transistors). </strong><hr></blockquote>
Thanks for the explanation.
[quote]Originally posted by moki:<strong>
Yes, it does, doesn't it. ::cryptic grin::</strong><hr></blockquote>
My fantasy - multicore 75X using Rapid I/O with Altivec, extended pipelines. The 860 finally appears. :eek:
<strong>The single best reason for Apple to stay dual is that a lot of third party software makers (and apple) went to the trouble to code their stuff for dual CPUs. Apple has a core of software that can take full advantage of multiple CPUs. Keeping that is worth making a dual 970.</strong><hr></blockquote>
Yes, but many important developers keep behind on dual processor support. We regularely get smashed on the head with benchtests where dual Mac's got trenched by a P4 2+ GHz. And mostly that's because they use apps like Photoshop or After Effects, who are not optimised for duals.
imagine i would give you the money for the (one) next PowerMac you are willing to buy (condition: i need to know the exact date of when you buy). when would you personally buy your powermac - what will be the excact date?
What exactly makes a processor designed to be used in SMP? (not questioning your knowledge, just wondering)
</strong><hr></blockquote>
Basically, it boils down to two things. An "SMP ready" processor has some way to talk to another processor, whether via a shared bus (G4) or a dedicated bus (970). It also has to have some way to keep its caches up-to-date with the caches of other processors. If you've seen a processor descibed as MEI (the G3), MESI (the 604e, the 7450) or MERSI (the 7400), this is shorthand for the set of protocols the chip supports to keep its caches coherent.
You don't have to have a shared bus and MERSI-level cache coherence on board the CPU to make an SMP machine. Be was able to ship dual G3 boards in their old BeBox. But something has to keep the CPUs talking to each other and their caches updated, so if the CPUs don't, something else on the logic board does. That means more work, more overhead and less performance, so it's not commonly done.
On the other hand, full SMP support is not something you add lightly to a CPU. It's a lot of work, and a goodly number of transistors (and in the case of the 970, costly pinouts). The 970's support for SMP is robust enough that it will be used in SMP configurations - the alternative is that IBM wasted a lot of time, money, transistors and silicon. The only questions are: by whom, and at what cost?
I don't doubt that Apple will go dual something or other. They might wait for the die shrink to .09, or for the dual-core version moki is hinting at, but they'll do it. Maybe we can hope for a dual dual-core PowerMac?
Yes, but many important developers keep behind on dual processor support. We regularely get smashed on the head with benchtests where dual Mac's got trenched by a P4 2+ GHz. And mostly that's because they use apps like Photoshop or After Effects, who are not optimised for duals.</strong><hr></blockquote>
Actually, Photoshop uses duals fairly well, even on OS 9. How well it does seems to depend on who you ask, and how they measure it. According to <a href="http://www.powerpage.org"; target="_blank">the PowerPage</a>, PC Magazine just tested the dual 1.25, and it compared well with a 3.06GHz P4. I looked for a link at <a href="http://www.pcmag.com/,"; target="_blank">http://www.pcmag.com/,</a>; but it looks like the article won't be posted until their print subscribers have had a chance to read it. So take it for what it's worth (which is not much until we have some detail on how they got that result).
After Effects appears to be scantily optimized for Apple's pro machines.
Actually, Photoshop uses duals fairly well, even on OS 9. How well it does seems to depend on who you ask, and how they measure it. According to <a href="http://www.powerpage.org"; target="_blank">the PowerPage</a>, PC Magazine just tested the dual 1.25, and it compared well with a 3.06GHz P4. I looked for a link at <a href="http://www.pcmag.com/,"; target="_blank">http://www.pcmag.com/,</a>; but it looks like the article won't be posted until their print subscribers have had a chance to read it. So take it for what it's worth (which is not much until we have some detail on how they got that result).
After Effects appears to be scantily optimized for Apple's pro machines.
<strong>Basically, it boils down to two things. An "SMP ready" processor has some way to talk to another processor, whether via a shared bus (G4) or a dedicated bus (970). It also has to have some way to keep its caches up-to-date with the caches of other processors. </strong><hr></blockquote>
There's another important piece of getting SMP going on you computer - the software. In 'thread aware' or 'SMP ready' software all the function need to be either a) 'reentrant' meaning two different CPUs are running the same function across different data just fine, or b) 'blocking'. There's some things it doesn't make any sense to allow two separate threads to manipulate simultaneously - but those need a mechanism to 'tell' the other thread "Hey, BUSY here, back off".
The nice thing about most of the software issues is that the number of processors isn't really a concern - if it was well written for 2 processors, it shouldn't matter if there are actually 32 processors. Things that block will still block (it still only makes sense to allow one function at a time to set the IP address).
That's one of the nice things about Duals across the Power line. All the performance hungry apps are either SMP-ready, or have proven through their actions that they consider MacOS a second-class citizen.
If the 970 volume production good and cost is low we will see duals across the board. But if the volume production of the 970 is lower than expected the you might see 1 or 2 high end dual and 1 or 2 single processer models.
Apple is going to do a slow release. It will release a highend model first then release the other models over a 6 - 10 week period of time. This is because they will announce the high end 2 to 4 weeks before it release, because they want to control the inventory. when you can order your model on the day they are announced they will be able to control production. Apple will be looking to sell 300,000 to 500,000 in the first 12 to 16 weeks.
By the way don't be shocked if the new powermacs are announced late February or early March with the power 970 model available beginning in april with the low end available in june. Let just say that some of apple R&D was used to assist IBM on the cost of the chip startup. Apple idea is the sooner the better. Another clue that the 970 will be here sooner than later is when you see the iMac update at the end of the month. Once the update is done everyone will be asking why buy a powermac?
well apple will take a hit on power mac sale for 30 days until the new powermacs are announced.
Actually, Photoshop uses duals fairly well, even on OS 9. How well it does seems to depend on who you ask, and how they measure it. According to <a href="http://www.powerpage.org"; target="_blank">the PowerPage</a>, PC Magazine just tested the dual 1.25, and it compared well with a 3.06GHz P4. I looked for a link at <a href="http://www.pcmag.com/,"; target="_blank">http://www.pcmag.com/,</a>; but it looks like the article won't be posted until their print subscribers have had a chance to read it. So take it for what it's worth (which is not much until we have some detail on how they got that result).
After Effects appears to be scantily optimized for Apple's pro machines.
As true as that may be, Adobe official fix or work around for the file browser causeing PS to crash is to drag the multi processor plug in out of the PS folder. So, I have no dual support here because I'm trying for stability. Myabe when Adobe releases an update with beter coding I can see how much faster PS can be given how slow it is now.
<strong>By the way don't be shocked if the new powermacs are announced late February or early March with the power 970 model available beginning in april with the low end available in june. Let just say that some of apple R&D was used to assist IBM on the cost of the chip startup. Apple idea is the sooner the better. Another clue that the 970 will be here sooner than later is when you see the iMac update at the end of the month. Once the update is done everyone will be asking why buy a powermac? </strong><hr></blockquote>
I like this statement. It leaves a nice warm afterglow - like a stiff shot of scotch. I think I am gonna print this one out - if you see it in print, it has to be true right
As true as that may be, Adobe official fix or work around for the file browser causeing PS to crash is to drag the multi processor plug in out of the PS folder.</strong><hr></blockquote>
Whoops.
The "plug in" must be an OS 9 vestige? If they dropped it (along with OS 9 support), and threaded their code they'd probably have simpler, more stable and more reliable MP support.
Not that it's necessarily easy to go back into a codebase as big and old as Photoshop's is and thread it...
imagine i would give you the money for the (one) next PowerMac you are willing to buy (condition: i need to know the exact date of when you buy). when would you personally buy your powermac - what will be the excact date? </strong><hr></blockquote>
The exact date would be the second they are available for purchase.
Honestly, I have no idea. Even engineering teams don't know exactly when products will ship. When I'm working on software, people always ask "when will it ship??" Even though I'm the lead developer, I really can't give them a good answer, other than "when it is done." It is just the nature of the beast.
I can speculate, though -- Apple is aware that if they don't release desktop systems that meet or beat comparable Windows XP systems in terms of speed by the end of the year, they are going to face further "pro" market erosion. If we don't see them in 2003, I will be very, very surprised. I'm betting on or around MacWorld/NYC, but as I said, I have no specific knowledge about release dates.
As true as that may be, Adobe official fix or work around for the file browser causeing PS to crash is to drag the multi processor plug in out of the PS folder. So, I have no dual support here because I'm trying for stability. Myabe when Adobe releases an update with beter coding I can see how much faster PS can be given how slow it is now.</strong><hr></blockquote>
Kid, see if you can find an older version of the MP plug-in. I seam to remember reading somewhere that the latest is not as fast as the last rev, it might be more stable as well, though not sure if it will work. I'm still on OS 9, Illustrator 8 and PS 6 at work...the lumbering pace of the publishing industry....
[quote]<strong>I think the first, if only line to get quad processors would be the xServe. I dont see the current design being able too keep 4 processors cool, as well as the rest of the components. Just look (err hear) how loud the current dual G4s are now. With an xServe you don't need to worry about noise, just put it in a closet.</strong><hr></blockquote>
After the initial round of 970-powered xServes (single- and double processor models), the xServe series will split up. One branch will use second-tier CPUs and retain the rack-mount server focus; the other will become the flagship and groundbreaker of Apple computer hardware design. Informally designated xRender, it will be the first to implement dual-core .09 CPUs (projected to be starting around 2.2 GHz).
Timeframe (projected): late Spring/early Summer -04 - if this holds, it will be in place to power PiXar's hitherto most ambitious venture - post-Disney.
<strong>Later also in the portable as well.</strong><hr></blockquote>
Tthe duals in laptops part is unlikely to happen any time soon, and here's why.
There is a fundamental difference between a single node machine and a multi-node machine--basically, a synchronization problem arises across the L1 and L2 caches for each processor [when a CPU needs to work on data from the same memory address bloc where another cpu has recently worked].
This is resolved via the MESI protocol--but this requires each CPU to snoop the memory bus and 'visible' caches of the other CPUs EVERY time it accesses memory.
This can be done efficiently, but ipso facto every snoop [on every cycle] the power levels on the bus must be maintained at a higher level--a fortiori, all this snooping and state-checking means a lot of transistors are cycling on each clock. Meaning even more power and even more heat.
So the board would have to be redesigned to handle higher power and heat, and the case to accommodate the additional power and heat dissipation and more components [especially when it's the bus itself which is generating a lot of heat].
I'm not saying that it cannot be done--but the cost/benefit curve is probably right under the cost/benefit curve of a vamped-up single CPU configuration.
Comments
<strong>
The delta in the number of transistors between a 32-bit 970 and a 64-bit 970 would be very small.... (maybe double to ~4 million transistors). </strong><hr></blockquote>
Thanks for the explanation.
[quote]Originally posted by moki:<strong>
Yes, it does, doesn't it. ::cryptic grin::</strong><hr></blockquote>
My fantasy - multicore 75X using Rapid I/O with Altivec, extended pipelines. The 860 finally appears. :eek:
Hope springs eternal
<strong>
this sort of response ought to be illegal- a short skirt in a bar at 2:00am is less a tease than Moki.</strong><hr></blockquote>
... "things that require a spanking Alex?"
"Bing!: 5 points"
<strong>The single best reason for Apple to stay dual is that a lot of third party software makers (and apple) went to the trouble to code their stuff for dual CPUs. Apple has a core of software that can take full advantage of multiple CPUs. Keeping that is worth making a dual 970.</strong><hr></blockquote>
Yes, but many important developers keep behind on dual processor support. We regularely get smashed on the head with benchtests where dual Mac's got trenched by a P4 2+ GHz. And mostly that's because they use apps like Photoshop or After Effects, who are not optimised for duals.
imagine i would give you the money for the (one) next PowerMac you are willing to buy (condition: i need to know the exact date of when you buy). when would you personally buy your powermac - what will be the excact date?
greets,
krassy
edit: added condition
[ 01-13-2003: Message edited by: Krassy ]</p>
<strong>
What exactly makes a processor designed to be used in SMP? (not questioning your knowledge, just wondering)
</strong><hr></blockquote>
Basically, it boils down to two things. An "SMP ready" processor has some way to talk to another processor, whether via a shared bus (G4) or a dedicated bus (970). It also has to have some way to keep its caches up-to-date with the caches of other processors. If you've seen a processor descibed as MEI (the G3), MESI (the 604e, the 7450) or MERSI (the 7400), this is shorthand for the set of protocols the chip supports to keep its caches coherent.
You don't have to have a shared bus and MERSI-level cache coherence on board the CPU to make an SMP machine. Be was able to ship dual G3 boards in their old BeBox. But something has to keep the CPUs talking to each other and their caches updated, so if the CPUs don't, something else on the logic board does. That means more work, more overhead and less performance, so it's not commonly done.
On the other hand, full SMP support is not something you add lightly to a CPU. It's a lot of work, and a goodly number of transistors (and in the case of the 970, costly pinouts). The 970's support for SMP is robust enough that it will be used in SMP configurations - the alternative is that IBM wasted a lot of time, money, transistors and silicon. The only questions are: by whom, and at what cost?
I don't doubt that Apple will go dual something or other. They might wait for the die shrink to .09, or for the dual-core version moki is hinting at, but they'll do it. Maybe we can hope for a dual dual-core PowerMac?
<strong>to moki just one little question:
imagine i would give you the money for the (one) next PowerMac you are willing to buy. when would you personally buy your powermac?
greets,
krassy
[ 01-13-2003: Message edited by: Krassy ]</strong><hr></blockquote>
He will buy it as soon the 970 powermac will be released
<strong>
Yes, but many important developers keep behind on dual processor support. We regularely get smashed on the head with benchtests where dual Mac's got trenched by a P4 2+ GHz. And mostly that's because they use apps like Photoshop or After Effects, who are not optimised for duals.</strong><hr></blockquote>
Actually, Photoshop uses duals fairly well, even on OS 9. How well it does seems to depend on who you ask, and how they measure it. According to <a href="http://www.powerpage.org"; target="_blank">the PowerPage</a>, PC Magazine just tested the dual 1.25, and it compared well with a 3.06GHz P4. I looked for a link at <a href="http://www.pcmag.com/,"; target="_blank">http://www.pcmag.com/,</a>; but it looks like the article won't be posted until their print subscribers have had a chance to read it. So take it for what it's worth (which is not much until we have some detail on how they got that result).
After Effects appears to be scantily optimized for Apple's pro machines.
[ 01-13-2003: Message edited by: Amorph ]</p>
<strong>
He will buy it as soon the 970 powermac will be released
see my edit
<strong>
Actually, Photoshop uses duals fairly well, even on OS 9. How well it does seems to depend on who you ask, and how they measure it. According to <a href="http://www.powerpage.org"; target="_blank">the PowerPage</a>, PC Magazine just tested the dual 1.25, and it compared well with a 3.06GHz P4. I looked for a link at <a href="http://www.pcmag.com/,"; target="_blank">http://www.pcmag.com/,</a>; but it looks like the article won't be posted until their print subscribers have had a chance to read it. So take it for what it's worth (which is not much until we have some detail on how they got that result).
After Effects appears to be scantily optimized for Apple's pro machines.
[ 01-13-2003: Message edited by: Amorph ]</strong><hr></blockquote>
I believe <a href="http://www.pcmag.com/article2/0,4149,810846,00.asp"; target="_blank">http://www.pcmag.com/article2/0,4149,810846,00.asp</a>; is the article they're referring to. The test is far from comprehensive (they only tested six Photoshop functions) and didn't give the actual numerical results, but it did sound good. It probably says a lot that PC Mag gave any kind of a favorable review to a Mac.
<strong>Basically, it boils down to two things. An "SMP ready" processor has some way to talk to another processor, whether via a shared bus (G4) or a dedicated bus (970). It also has to have some way to keep its caches up-to-date with the caches of other processors. </strong><hr></blockquote>
There's another important piece of getting SMP going on you computer - the software. In 'thread aware' or 'SMP ready' software all the function need to be either a) 'reentrant' meaning two different CPUs are running the same function across different data just fine, or b) 'blocking'. There's some things it doesn't make any sense to allow two separate threads to manipulate simultaneously - but those need a mechanism to 'tell' the other thread "Hey, BUSY here, back off".
The nice thing about most of the software issues is that the number of processors isn't really a concern - if it was well written for 2 processors, it shouldn't matter if there are actually 32 processors. Things that block will still block (it still only makes sense to allow one function at a time to set the IP address).
That's one of the nice things about Duals across the Power line. All the performance hungry apps are either SMP-ready, or have proven through their actions that they consider MacOS a second-class citizen.
Apple is going to do a slow release. It will release a highend model first then release the other models over a 6 - 10 week period of time. This is because they will announce the high end 2 to 4 weeks before it release, because they want to control the inventory. when you can order your model on the day they are announced they will be able to control production. Apple will be looking to sell 300,000 to 500,000 in the first 12 to 16 weeks.
By the way don't be shocked if the new powermacs are announced late February or early March with the power 970 model available beginning in april with the low end available in june. Let just say that some of apple R&D was used to assist IBM on the cost of the chip startup. Apple idea is the sooner the better. Another clue that the 970 will be here sooner than later is when you see the iMac update at the end of the month. Once the update is done everyone will be asking why buy a powermac?
well apple will take a hit on power mac sale for 30 days until the new powermacs are announced.
<strong>
Actually, Photoshop uses duals fairly well, even on OS 9. How well it does seems to depend on who you ask, and how they measure it. According to <a href="http://www.powerpage.org"; target="_blank">the PowerPage</a>, PC Magazine just tested the dual 1.25, and it compared well with a 3.06GHz P4. I looked for a link at <a href="http://www.pcmag.com/,"; target="_blank">http://www.pcmag.com/,</a>; but it looks like the article won't be posted until their print subscribers have had a chance to read it. So take it for what it's worth (which is not much until we have some detail on how they got that result).
After Effects appears to be scantily optimized for Apple's pro machines.
[ 01-13-2003: Message edited by: Amorph ]</strong><hr></blockquote>
As true as that may be, Adobe official fix or work around for the file browser causeing PS to crash is to drag the multi processor plug in out of the PS folder. So, I have no dual support here because I'm trying for stability. Myabe when Adobe releases an update with beter coding I can see how much faster PS can be given how slow it is now.
<strong>By the way don't be shocked if the new powermacs are announced late February or early March with the power 970 model available beginning in april with the low end available in june. Let just say that some of apple R&D was used to assist IBM on the cost of the chip startup. Apple idea is the sooner the better. Another clue that the 970 will be here sooner than later is when you see the iMac update at the end of the month. Once the update is done everyone will be asking why buy a powermac?
I like this statement. It leaves a nice warm afterglow - like a stiff shot of scotch. I think I am gonna print this one out - if you see it in print, it has to be true right
<strong>
As true as that may be, Adobe official fix or work around for the file browser causeing PS to crash is to drag the multi processor plug in out of the PS folder.</strong><hr></blockquote>
Whoops.
The "plug in" must be an OS 9 vestige? If they dropped it (along with OS 9 support), and threaded their code they'd probably have simpler, more stable and more reliable MP support.
Not that it's necessarily easy to go back into a codebase as big and old as Photoshop's is and thread it...
<strong>to moki just one little question:
imagine i would give you the money for the (one) next PowerMac you are willing to buy (condition: i need to know the exact date of when you buy). when would you personally buy your powermac - what will be the excact date?
The exact date would be the second they are available for purchase.
Honestly, I have no idea. Even engineering teams don't know exactly when products will ship. When I'm working on software, people always ask "when will it ship??" Even though I'm the lead developer, I really can't give them a good answer, other than "when it is done." It is just the nature of the beast.
I can speculate, though -- Apple is aware that if they don't release desktop systems that meet or beat comparable Windows XP systems in terms of speed by the end of the year, they are going to face further "pro" market erosion. If we don't see them in 2003, I will be very, very surprised. I'm betting on or around MacWorld/NYC, but as I said, I have no specific knowledge about release dates.
jalapeno: single 1.8GHz 970 (underclocked to 1.6 for earlier release)
habanero: dual 1.6GHz 970 (no faster w/out more fans)
frickin'plasma: quad 1.4-1.6GHz 970 (cooling subsystem required)
<cost? one billion dollars! muahahaha!>
by next year's .09 process, another alternative might be die-based
dual cpu (single core)
single cpu (dual core)
quad cpu (dual core)
seriously...
how many cpu's did the NeXT support (ten years ago)? 16.
how many former-NeXT folks deep in Apple?
how could they resist supercomputing support?
pooch isn't the only clustering project.
bring on the frickin'plasma
if you build it they will give you chocolate
<strong>
As true as that may be, Adobe official fix or work around for the file browser causeing PS to crash is to drag the multi processor plug in out of the PS folder. So, I have no dual support here because I'm trying for stability. Myabe when Adobe releases an update with beter coding I can see how much faster PS can be given how slow it is now.</strong><hr></blockquote>
Kid, see if you can find an older version of the MP plug-in. I seam to remember reading somewhere that the latest is not as fast as the last rev, it might be more stable as well, though not sure if it will work. I'm still on OS 9, Illustrator 8 and PS 6 at work...the lumbering pace of the publishing industry....
Zilla and fast eliptical encryption are still under development at Apple for a future release of OSX, BTW.
On demand supercomputer/render farm on your home desktop, anyone? iCray.
After the initial round of 970-powered xServes (single- and double processor models), the xServe series will split up. One branch will use second-tier CPUs and retain the rack-mount server focus; the other will become the flagship and groundbreaker of Apple computer hardware design. Informally designated xRender, it will be the first to implement dual-core .09 CPUs (projected to be starting around 2.2 GHz).
Timeframe (projected): late Spring/early Summer -04 - if this holds, it will be in place to power PiXar's hitherto most ambitious venture - post-Disney.
engpjp
<strong>Later also in the portable as well.</strong><hr></blockquote>
Tthe duals in laptops part is unlikely to happen any time soon, and here's why.
There is a fundamental difference between a single node machine and a multi-node machine--basically, a synchronization problem arises across the L1 and L2 caches for each processor [when a CPU needs to work on data from the same memory address bloc where another cpu has recently worked].
This is resolved via the MESI protocol--but this requires each CPU to snoop the memory bus and 'visible' caches of the other CPUs EVERY time it accesses memory.
This can be done efficiently, but ipso facto every snoop [on every cycle] the power levels on the bus must be maintained at a higher level--a fortiori, all this snooping and state-checking means a lot of transistors are cycling on each clock. Meaning even more power and even more heat.
So the board would have to be redesigned to handle higher power and heat, and the case to accommodate the additional power and heat dissipation and more components [especially when it's the bus itself which is generating a lot of heat].
I'm not saying that it cannot be done--but the cost/benefit curve is probably right under the cost/benefit curve of a vamped-up single CPU configuration.