What does a dual 970 configuration mean for throughput, f.e. 2*6,4? Can the rest of the hardware keep up with this massive amount?
Pim
a dual 970 config will mean that the companion chip will need to support double the bandwidth - 14,4GB/s for dual 1,8Ghz 970's. The rest of the hardware probably can't keep up, unless Apple decides to use some kind of really expensive memory. Dual channel PC3600 (450MHz/225MHz DDR) would be able to put through exactly half (3,6x2=7,2) of what the processors are capable of, or enough to feed 1 processor with data. Even though only being able to shuffle around half, more or less, of the data, Apple will still have a very, very speedy mac
Actually there was an interesting discussion on Ars about bus topologies for the 970. I had previously been assuming that a dual setup would either require a chipset with 2 ports, or 2 chipsets connected by a bridge. A third possibility was pointed out -- since the input and output are essentially on seperate buses, it would be possible for the protocol to operate in a "ring" fashion. The processors and chipset(s) would be connected in a loop and data would pass around from one device to the next (e.g. input connects upstream, output connects downstream). This has the advantage of being able to scale to multiple processors without having to re-engineer the companion chip. It does mean that bandwidth is shared and essentially 3.2 GB/sec instead of 6.4 GB/sec (@ 1.8 GHz), however.
Could it be possible that the info on the 2.5 Ghz PPC 970 was released too soon and IBM went on damage control? Apple could have 2.5 in mind for this revision but want it to be a surprise. Hell, if Jobs spills the beans at the WWDC Keynote people will begin cheering and start thinking 1.4, 1.6, and 1.8 configs. Then the roar would be even louder a few minutes later when he says it is going to top out at 2.5. I don't know, but Apple may be trying to keep our expectations down for a big surprise.
I'd expect Apple to use this to maximize performance.
Quote:
Apple should use the fastest available chip in the low end single CPU machine, and duals in the other two. PM's are still very expensive and with 970 backed performance they'll better justify their current prices but will still ALL fall in the high-end. 1500 for a machine without a display? Not good, considering that a single low speed ppc970 will not offer any kind of performance revelation over far cheaper x86 boxen,
That would just drive a bunch of users to the Entry Level Powermac. PM are expensive because no other PC offers the Design touches of a Powermac. PCs are great if you want a generic look with a swoopy bezel and a generic WinXP OS. Some people want more. OSX is Apple's Coupe De Grace here. The better it becomes the more users will flock to Apple.
Apple does have to remain competitive on price but it's not as important as some make it. Value is determined by each individual. Apple has never marketed itself as the lowest price. Consumer hear " Lowest Price Guaranteed" screams from every Retail Chain on the Planet.
LOL Matsu tell me you don't see envy in some people eyes when you whip our your Minibook? Half the people that bash Apple would LOVE to have one but just can't bring themselves to spend the additional money.
At any rate back on topic. I wonder if Apple may double up on Powermacs initially. Say something like
$1499- 1.4Ghz 970
$1899- 1.6Ghz 970
$2299- 1.8Ghz 970
$1899- 1.4Ghz 970 DP
$2299- 1.6Ghz 970 DP
$2699- 1.8Ghz 970 DP
My reasoning is why would Apple go through the expense of designing two Motherboards unless Single and DP Powermacs/Xserve would be available along the whole lineup?
The old G4 uses a shared bus (FSB), and therefore does not need different motherboard designs to support multiple processors. Bandwidth is the same, even if the need for bandwidth is double (dual processors). The PPC970 is different, and needs a different companion chip when the number of processors differ. This is to be sure that the processors are fed with data, when the number of processors double, the bandwidth they need is double, even though the 2:1 FSB feeds 6,4GB/s (that is approx. 6 times as much as the G4 bus) when the processor clocks at 1,8Ghz. The PPC970 is designed for MP, and with this design, the processors will always be fed with data, equally (if the workload if large enough of course) on 1, 2, 4 and n-way designs.
(I know this is a terrible post, but I hope it made some sense.)
Mmmm... interesting stuff - I think I just about understand it. Sounds like the 970 is going to be awesome in MP configs.
I think there might be some disinformation, or incomplete information about the mother boards. I'm not a hardware expert, however Apple's history has been pretty steadfast in using daughtercard configurations in their high end computers. There might be a technical reason that this is not possable with the 970.
I would think that they could design system using them with more contacts (pin-outs, whatever the technical term is) on the connector if needed and a companion chip on the daughter card to match the number of processors. This would be a very flexable option for Apple, allowing them to produce singles and duals on one mother board, and the second singles up to Quads (I seam to remember one of the clones using this config with a dual daughter card motherboard, the S900?). They could cover a much broader market range with this configuration, from a low end consumer single to the "Uber Mac" Quad workstation, and eventually use the same parts for Quad high end Xserves.
If they are in fact not using daughter cards, then I would expect them to move to a ZIF (or whatever the current high end "removable" connector is now) on the mother boards so that they could "scale" them to faster chips easier. The problem with this is that it opens up to easier aftermarket chip upgrades, which makes upgrading more attractive than replacing PM's, at least in the long run, which has the potential of cutting Apple's future revenue. This would also force them to spend more money to cover a broader market range for their products.
The 15" PowerBook is sticking around so that Mac OS 9 users can buy a new Mac to do whatever they need to do that OS X isn't yet capable of.
I'm going to throw my hat in the "iMac gets a 970" ring, which is a bit of a change from the position I've taken before. First of all, remember how Steve introduced the iMac? He emphasized the fact that it was not last year's technology. Well, after the pro line goes 970, the G4 will be last year's technology. If the iMac continues to more-or-less follow the PowerBook, then I expect it to sport a 970 late this fall, or perhaps at MWSF. Both can use the lower-voltage (and thus, lower wattage and slower clocked) 970s, leaving the PowerMacs to keep the highest MHz rating (in addition to whatever other enhancements Apple decides to throw into their workstation line).
I don't think either of the boards that MacWhispers is talking about is an iMac board, or a PowerBook board. Not yet. I don't think it's a PowerMac board either.
There seems to be some controversy over whether Panther is sufficient to reschedule WWDC, or if the 970 must be involved as well. I think the first is closer, but both alternatives miss the point. First: As Programmer and Kickaha and others have pointed out, very little will need to be done in order to move from the G4 to the 970, all else being equal. Most programs won't even need a recompile, because the 970 already supports PPC and VMX, and most programs can live happily within 2GB of virtual memory. Faster performance will come automatically and transparently. So the 970, in and of itself, is no reason to postpone a developer conference. I think Apple is right: Panther is the big news. But why?
Well, think about it: An operating system's job is to exploit the hardware; more precisely, to package the capabilities of the hardware into a developer-and user-friendly form. Apple has gotten really good at making sure that OS X uses the available hardware efficiently (QE being one example). So, if all the hardware is going to be is 970-based PowerMacs that are as close to the older variety as is practical, and which are intended to be used in much the same way, then there's really nothing new or radical for the OS to do. Panther would be more evolutionary than revolutionary. But if the hardware were capable of behaving in altogether new (to PCs) ways, and Panther exposed, say, pervasive Rendezvous enabled distributed computing, assisted by Cocoa's existing ability to disguise a remote process as just another task running next door, and if it was implemented as core functionality, then Apple would be introducing a new (again, to PCs) paradigm, and they'd be deeply interested in getting as many of their developers on board as possible.
Actually there was an interesting discussion on Ars about bus topologies for the 970. I had previously been assuming that a dual setup would either require a chipset with 2 ports, or 2 chipsets connected by a bridge. A third possibility was pointed out -- since the input and output are essentially on seperate buses, it would be possible for the protocol to operate in a "ring" fashion. The processors and chipset(s) would be connected in a loop and data would pass around from one device to the next (e.g. input connects upstream, output connects downstream). This has the advantage of being able to scale to multiple processors without having to re-engineer the companion chip. It does mean that bandwidth is shared and essentially 3.2 GB/sec instead of 6.4 GB/sec (@ 1.8 GHz), however.
Uhm, I were unaware of this. Maybe that's how they plan to have their first incarnation of the dual PPC970 designed, and later upgrade to a more complex, expensive companion chip (with support for the bandwidth the dual processors supports), when they finish it, or when it's feasible to produce, if it is not currently or in the near future.
I think there might be some disinformation, or incomplete information about the mother boards. I'm not a hardware expert, however Apple's history has been pretty steadfast in using daughtercard configurations in their high end computers. There might be a technical reason that this is not possable with the 970...
The information about motherboards isn't what I would call reliable, but the argument that Apple has used daughtercards in the past/current machines and will continue to on the new ones doesn't hold water. The new 970 bus design runs at 900 MHz (and more in the future) and thus has very tight timing & connection restrictions. It may simply not be possible to use a board-to-board connection between the processor(s) and the companion chip(s). The companion chip probably wants to be on the same board as its memory. At that point there is so much on the one board that they might as well just put everything on one board (or else all the I/O on a daughtercard if the companion chip isn't one big fully integrated monster). Alternatively the motherboard information might just be inaccurate, or might omit the fact that it comes in two pieces. The rumours are a bit thin, and the facts are even thinner.
Uhm, I were unaware of this. Maybe that's how they plan to have their first incarnation of the dual PPC970 designed, and later upgrade to a more complex, expensive companion chip (with support for the bandwidth the dual processors supports), when they finish it, or when it's feasible to produce, if it is not currently or in the near future.
If this is the case we may never see a non-ring implementation. Instead they'd probably just increase the clock rate and/or move to QDR. This isn't necessarily a bad thing, however, as its possible that the ring topology could actually deliver its promised bandwidth, compared to the x86 buses which only hit their claimed numbers is very specialized situations (if at all). I do expect the 970 to be very efficient at using bandwidth, and if Apple's current MPX memory controller is any indication the companion chip will be very efficient as well.
Apple should use the fastest available chip in the low end single CPU machine, and duals in the other two. PM's are still very expensive and with 970 backed performance they'll better justify their current prices but will still ALL fall in the high-end. 1500 for a machine without a display? Not good, considering that a single low speed ppc970 will not offer any kind of performance revelation over far cheaper x86 boxen, the game is either to use the fastest 970's available or duals.
1.8 single, dual 1.4, and dual 1.8. Same configs as currently used. Or they could possibly use something just a mite slower in the low end and include a superdrive, which at the price ought to be included in all PM's, or they could, gasp, further lower the price of the entry level tower.
Didn't Apple once have a lineup with a single CPU model that had a higher clock speed than a dual model, and people were getting confused?
I don't think either of the boards that MacWhispers is talking about is an iMac board, or a PowerBook board. Not yet. I don't think it's a PowerMac board either.
There seems to be some controversy over whether Panther is sufficient to reschedule WWDC, or if the 970 must be involved as well. ... So the 970, in and of itself, is no reason to postpone a developer conference. I think Apple is right: Panther is the big news. But why?
Well, think about it: An operating system's job is to exploit the hardware; more precisely, to package the capabilities of the hardware into a developer-and user-friendly form. ... Apple would be introducing a new (again, to PCs) paradigm, and they'd be deeply interested in getting as many of their developers on board as possible.
Food for thought. [/B]
Feeding my thoughts is that the most likely Mac to get the 970 first is the Xserve. Panther would be sufficient to reschedule WWDC if it let the "Cat" out of the bag about the 970. If it has obvious 64-bitness in it's construction then the 970 is a GO!
The information about motherboards isn't what I would call reliable, but the argument that Apple has used daughtercards in the past/current machines and will continue to on the new ones doesn't hold water. The new 970 bus design runs at 900 MHz (and more in the future) and thus has very tight timing & connection restrictions. It may simply not be possible to use a board-to-board connection between the processor(s) and the companion chip(s). The companion chip probably wants to be on the same board as its memory. At that point there is so much on the one board that they might as well just put everything on one board (or else all the I/O on a daughtercard if the companion chip isn't one big fully integrated monster). Alternatively the motherboard information might just be inaccurate, or might omit the fact that it comes in two pieces. The rumours are a bit thin, and the facts are even thinner.
Programmer, I did not mean to imply that since Apple has in the past used daughter cards they will in the future. But, rather that they have used them to great effect in the past, and that if possable I think that they will use them in the future. This gives them much more flexability in manufacturing and marketing computers, allowing them to cover a 1 to 4 processor range with 2 mother boards in stead of 4. I do see Apple moving beyond the dual configuration, they need it to address the video market and make a more powerfull pressence in the server market.
Some questions that need to be resolved are, can they put the companion chip on the daughter card, or can they design a companion chip that would be on the mother board which could cover both a single and dual configuration? For a Quad set-up would they need 2 companion chips, 4, or would they need yet another chip designed specifically for Quads? If a unique chip needs to be designed for each configuration then that creates some major problems in implementing MP systems. How does IBM handle this with the Power 4?
..I highly doubt that we will see the 970 in an Xserve first. Apple sells Xserves in quantities totalling less than 5,000/qtr, IIRC. So, I don't think Apple is gonna pop a 970 into a machine that sells in such low quantities and completely destroy any PowerMac sales until the introduction of the 970 based PowerMac.
Nope, when the 970 comes, it will either arrive as a 1.2GHz 970 in a 15" PowerBook, or as the expected 1.4, 1.6, 1.8 configs bandied about in the PowerMacs, or both. Apple can't afford to strangle its sales by doing a slow roll out of the 970. They know that if they do sales will stagnate while consumers wait for the other shoe to drop.
If it is introduced in an out of the ordinary fashion, Apple might introduce it in workstation class machines to milk the bleeding edgers first, but I don't think even that would pay off as regular consumers would just bide their time waiting for it to trickle down.
...Not good, considering that a single low speed ppc970 will not offer any kind of performance revelation over far cheaper x86 boxen...
We won't know actual performance for awhile yet, of course, but the ppc 970 @ 1.4GHz should be pretty respectable going by the (non-AV, non-SMP) estimated spec benchmarks. To the point that, yes, there will be cheaper x86 that beat it in test1, test2 etc., but... the way the features are aligned I do _not_ think that that it will be so completely outclassed at _all_ benchmarks.
That is:
_IF_ you have a program that really does benefit from 64-bits (Oracle)
_OR_ you have a program that needs lots of double precision FPU (Lightwave)
_OR_ you have Altivec programs that are crippled by the FSB (Anything Altivecable)
_THEN_ there's going to be three areas of drastic improvement.
There's also the recurring rumors that something _else_ is in the pipeline. GPU-as-CPU's-assistant, or multi-GPUs... If true (or something else interesting is going on) that's another enticing piece of the puzzle.
I don't _ever_ expect the PowerMac's (or whatever the 'pro' line is called) to beat x86 boxes on price.
The areas it's supposed to be competitive in are the (price/performance) and pure performance areas. The 970, even the lower clocked ones, should be quite good _at_specific_jobs_.
The Make-Matsu-Happy machine (a one-slot desktop with max CPUs & a $1000 price tag) isn't in the cards. The closest that will happen is a the iMacs dropping in price & improving in power as the 970 gets adopted. But Matsu, it _won't_ make you happy.
Then they _won't_ sell me a desktop. Not that steve cares, but the way the numbers play out they won't sell a whole lot of other people desktops either. And as long as I remain in a dual platform environment at home, dropping the mac is as easy as a quick trip over to eBay. I like my PB (at the price I paid for it) but that really ought to be the regular price to begin with. Still this is all besides the point. I buy to suit me, the miniPB (discounted) suits me and my budget, when Apple provides more machines that do that, I will own more Apples, if they don't, I will own more windows, and if Apple drags its feet too much, I will own only windows. What Steve needs to understand is that macs are nice, but they simply aren't that nice.
Speaking of cube form factors, the shuttle miniXPC form factor is steadily growing in PC circles -- 1 PCI, 1 AGP (often in addition to on-board graphics and sound) one CPU SOCKET, one optical bay and one/two HDD bays -- hmmm... why does that sound so familiar? Yeah, I must be making untowards demands on Apple again. Earth to Apple, 1000USD no longer describes a low-end or even a lower mid-range machine. A 1000 headless (monitorless) USD box is now firmly in the upper mid-range of desktop machinery. Most people aren't interested in a barebones tower at a workstation price, nor are they interested in an planned obsolescence AIO at a high-end tower price.
Go ahead, get smug about it, I'm not the one losing market-share on a quaterly basis. Has Apple ever strung together more than 2 quaters of growth (as a percentage of marketshare) in the last 5 years? I don't care about excuses, if you can't build a price and feature competitive box then maybe you don't deserve to be in business.
Then they _won't_ sell me a desktop. Not that steve cares, but the way the numbers play out they won't sell a whole lot of other people desktops either
The only numbers that matter are Margin and Revenue. Steve doesn't care. He never has. He's always had above average tastes. Because Joe Bloe Walmart shopper doesn't share those those tastes is no concern of his.
Quote:
Go ahead, get smug about it, I'm not the one losing market-share on a quaterly basis. Has Apple ever strung together more than 2 quaters of growth (as a percentage of marketshare) in the last 5 years? I don't care about excuses, if you can't build a price and feature competitive box then maybe you don't deserve to be in business.
Nevyn is not being smug. Whether Matsu buys a Mac or a PC has no bearing on what the rest of the 6 billion people on this earth do. You are entitled to govern your life as you see fit. If Apple is not in your plans then that's the way it is. The fact is Apple IS in business, they've acquired companies and are planned for future growth.
The thing is Matsu you are just a consumer. You think and react like one. You feel like Price is the biggest motivator and to Millions it is. But there also exist Millions who feel that a value propostion must meet more than just price.
despite your constant complaints about price you will not control the fate of Apple.
Actually there was an interesting discussion on Ars about bus topologies for the 970. I had previously been assuming that a dual setup would either require a chipset with 2 ports, or 2 chipsets connected by a bridge. A third possibility was pointed out -- since the input and output are essentially on seperate buses, it would be possible for the protocol to operate in a "ring" fashion. The processors and chipset(s) would be connected in a loop and data would pass around from one device to the next (e.g. input connects upstream, output connects downstream). This has the advantage of being able to scale to multiple processors without having to re-engineer the companion chip. It does mean that bandwidth is shared and essentially 3.2 GB/sec instead of 6.4 GB/sec (@ 1.8 GHz), however.
That's 3.2GB/sec per processor, for a total bandwidth to the memory of 6.4GB/sec. If the bus was architected not as a ring but with two independent connections to the memory controller, then the total bandwidth would be 12.8GB/sec, which is more than even dual-channel DDR400 could provide. In other words - it doesn't really matter, because no memory can feed that right now. If future generations of the 970 retain the 4:1 DDR FSB, then we'll always have more bus bandwidth than memory bandwidth either way.
On a side note, it's a welcome change to have a faster bus than memory instead of the other way around.
Feeding my thoughts is that the most likely Mac to get the 970 first is the Xserve. Panther would be sufficient to reschedule WWDC if it let the "Cat" out of the bag about the 970. If it has obvious 64-bitness in it's construction then the 970 is a GO!
...
I can't agree I'm afraid...the point of the xserve is reliability not out and out speed.
As the situation stands the xserve is slightly slower than the top PM's.
There's no need to put a hot expensive chip in a server box that needs to be stable and problem free.
Comments
Originally posted by pim_fortuyn
What does a dual 970 configuration mean for throughput, f.e. 2*6,4? Can the rest of the hardware keep up with this massive amount?
Pim
a dual 970 config will mean that the companion chip will need to support double the bandwidth - 14,4GB/s for dual 1,8Ghz 970's. The rest of the hardware probably can't keep up, unless Apple decides to use some kind of really expensive memory. Dual channel PC3600 (450MHz/225MHz DDR) would be able to put through exactly half (3,6x2=7,2) of what the processors are capable of, or enough to feed 1 processor with data. Even though only being able to shuffle around half, more or less, of the data, Apple will still have a very, very speedy mac
The Strange Case of Jack Campbell
I'd expect Apple to use this to maximize performance.
Apple should use the fastest available chip in the low end single CPU machine, and duals in the other two. PM's are still very expensive and with 970 backed performance they'll better justify their current prices but will still ALL fall in the high-end. 1500 for a machine without a display? Not good, considering that a single low speed ppc970 will not offer any kind of performance revelation over far cheaper x86 boxen,
That would just drive a bunch of users to the Entry Level Powermac. PM are expensive because no other PC offers the Design touches of a Powermac. PCs are great if you want a generic look with a swoopy bezel and a generic WinXP OS. Some people want more. OSX is Apple's Coupe De Grace here. The better it becomes the more users will flock to Apple.
Apple does have to remain competitive on price but it's not as important as some make it. Value is determined by each individual. Apple has never marketed itself as the lowest price. Consumer hear " Lowest Price Guaranteed" screams from every Retail Chain on the Planet.
LOL Matsu tell me you don't see envy in some people eyes when you whip our your Minibook? Half the people that bash Apple would LOVE to have one but just can't bring themselves to spend the additional money.
At any rate back on topic. I wonder if Apple may double up on Powermacs initially. Say something like
$1499- 1.4Ghz 970
$1899- 1.6Ghz 970
$2299- 1.8Ghz 970
$1899- 1.4Ghz 970 DP
$2299- 1.6Ghz 970 DP
$2699- 1.8Ghz 970 DP
My reasoning is why would Apple go through the expense of designing two Motherboards unless Single and DP Powermacs/Xserve would be available along the whole lineup?
Originally posted by r-0X#Zapchud
The old G4 uses a shared bus (FSB), and therefore does not need different motherboard designs to support multiple processors. Bandwidth is the same, even if the need for bandwidth is double (dual processors). The PPC970 is different, and needs a different companion chip when the number of processors differ. This is to be sure that the processors are fed with data, when the number of processors double, the bandwidth they need is double, even though the 2:1 FSB feeds 6,4GB/s (that is approx. 6 times as much as the G4 bus) when the processor clocks at 1,8Ghz. The PPC970 is designed for MP, and with this design, the processors will always be fed with data, equally (if the workload if large enough of course) on 1, 2, 4 and n-way designs.
(I know this is a terrible post, but I hope it made some sense.)
Mmmm... interesting stuff - I think I just about understand it. Sounds like the 970 is going to be awesome in MP configs.
I would think that they could design system using them with more contacts (pin-outs, whatever the technical term is) on the connector if needed and a companion chip on the daughter card to match the number of processors. This would be a very flexable option for Apple, allowing them to produce singles and duals on one mother board, and the second singles up to Quads (I seam to remember one of the clones using this config with a dual daughter card motherboard, the S900?). They could cover a much broader market range with this configuration, from a low end consumer single to the "Uber Mac" Quad workstation, and eventually use the same parts for Quad high end Xserves.
If they are in fact not using daughter cards, then I would expect them to move to a ZIF (or whatever the current high end "removable" connector is now) on the mother boards so that they could "scale" them to faster chips easier. The problem with this is that it opens up to easier aftermarket chip upgrades, which makes upgrading more attractive than replacing PM's, at least in the long run, which has the potential of cutting Apple's future revenue. This would also force them to spend more money to cover a broader market range for their products.
I'm going to throw my hat in the "iMac gets a 970" ring, which is a bit of a change from the position I've taken before. First of all, remember how Steve introduced the iMac? He emphasized the fact that it was not last year's technology. Well, after the pro line goes 970, the G4 will be last year's technology. If the iMac continues to more-or-less follow the PowerBook, then I expect it to sport a 970 late this fall, or perhaps at MWSF. Both can use the lower-voltage (and thus, lower wattage and slower clocked) 970s, leaving the PowerMacs to keep the highest MHz rating (in addition to whatever other enhancements Apple decides to throw into their workstation line).
I don't think either of the boards that MacWhispers is talking about is an iMac board, or a PowerBook board. Not yet. I don't think it's a PowerMac board either.
There seems to be some controversy over whether Panther is sufficient to reschedule WWDC, or if the 970 must be involved as well. I think the first is closer, but both alternatives miss the point. First: As Programmer and Kickaha and others have pointed out, very little will need to be done in order to move from the G4 to the 970, all else being equal. Most programs won't even need a recompile, because the 970 already supports PPC and VMX, and most programs can live happily within 2GB of virtual memory. Faster performance will come automatically and transparently. So the 970, in and of itself, is no reason to postpone a developer conference. I think Apple is right: Panther is the big news. But why?
Well, think about it: An operating system's job is to exploit the hardware; more precisely, to package the capabilities of the hardware into a developer-and user-friendly form. Apple has gotten really good at making sure that OS X uses the available hardware efficiently (QE being one example). So, if all the hardware is going to be is 970-based PowerMacs that are as close to the older variety as is practical, and which are intended to be used in much the same way, then there's really nothing new or radical for the OS to do. Panther would be more evolutionary than revolutionary. But if the hardware were capable of behaving in altogether new (to PCs) ways, and Panther exposed, say, pervasive Rendezvous enabled distributed computing, assisted by Cocoa's existing ability to disguise a remote process as just another task running next door, and if it was implemented as core functionality, then Apple would be introducing a new (again, to PCs) paradigm, and they'd be deeply interested in getting as many of their developers on board as possible.
Food for thought.
Originally posted by Programmer
Actually there was an interesting discussion on Ars about bus topologies for the 970. I had previously been assuming that a dual setup would either require a chipset with 2 ports, or 2 chipsets connected by a bridge. A third possibility was pointed out -- since the input and output are essentially on seperate buses, it would be possible for the protocol to operate in a "ring" fashion. The processors and chipset(s) would be connected in a loop and data would pass around from one device to the next (e.g. input connects upstream, output connects downstream). This has the advantage of being able to scale to multiple processors without having to re-engineer the companion chip. It does mean that bandwidth is shared and essentially 3.2 GB/sec instead of 6.4 GB/sec (@ 1.8 GHz), however.
Uhm, I were unaware of this. Maybe that's how they plan to have their first incarnation of the dual PPC970 designed, and later upgrade to a more complex, expensive companion chip (with support for the bandwidth the dual processors supports), when they finish it, or when it's feasible to produce, if it is not currently or in the near future.
Originally posted by @homenow
I think there might be some disinformation, or incomplete information about the mother boards. I'm not a hardware expert, however Apple's history has been pretty steadfast in using daughtercard configurations in their high end computers. There might be a technical reason that this is not possable with the 970...
The information about motherboards isn't what I would call reliable, but the argument that Apple has used daughtercards in the past/current machines and will continue to on the new ones doesn't hold water. The new 970 bus design runs at 900 MHz (and more in the future) and thus has very tight timing & connection restrictions. It may simply not be possible to use a board-to-board connection between the processor(s) and the companion chip(s). The companion chip probably wants to be on the same board as its memory. At that point there is so much on the one board that they might as well just put everything on one board (or else all the I/O on a daughtercard if the companion chip isn't one big fully integrated monster). Alternatively the motherboard information might just be inaccurate, or might omit the fact that it comes in two pieces. The rumours are a bit thin, and the facts are even thinner.
Originally posted by r-0X#Zapchud
Uhm, I were unaware of this. Maybe that's how they plan to have their first incarnation of the dual PPC970 designed, and later upgrade to a more complex, expensive companion chip (with support for the bandwidth the dual processors supports), when they finish it, or when it's feasible to produce, if it is not currently or in the near future.
If this is the case we may never see a non-ring implementation. Instead they'd probably just increase the clock rate and/or move to QDR. This isn't necessarily a bad thing, however, as its possible that the ring topology could actually deliver its promised bandwidth, compared to the x86 buses which only hit their claimed numbers is very specialized situations (if at all). I do expect the 970 to be very efficient at using bandwidth, and if Apple's current MPX memory controller is any indication the companion chip will be very efficient as well.
Originally posted by Matsu
Apple should use the fastest available chip in the low end single CPU machine, and duals in the other two. PM's are still very expensive and with 970 backed performance they'll better justify their current prices but will still ALL fall in the high-end. 1500 for a machine without a display? Not good, considering that a single low speed ppc970 will not offer any kind of performance revelation over far cheaper x86 boxen, the game is either to use the fastest 970's available or duals.
1.8 single, dual 1.4, and dual 1.8. Same configs as currently used. Or they could possibly use something just a mite slower in the low end and include a superdrive, which at the price ought to be included in all PM's, or they could, gasp, further lower the price of the entry level tower.
Didn't Apple once have a lineup with a single CPU model that had a higher clock speed than a dual model, and people were getting confused?
I think that we'll see:
-Single 1.4
-Single 1.6
-Dual 1.7
-Dual 1.9
Originally posted by Amorph
I don't think either of the boards that MacWhispers is talking about is an iMac board, or a PowerBook board. Not yet. I don't think it's a PowerMac board either.
There seems to be some controversy over whether Panther is sufficient to reschedule WWDC, or if the 970 must be involved as well. ... So the 970, in and of itself, is no reason to postpone a developer conference. I think Apple is right: Panther is the big news. But why?
Well, think about it: An operating system's job is to exploit the hardware; more precisely, to package the capabilities of the hardware into a developer-and user-friendly form. ... Apple would be introducing a new (again, to PCs) paradigm, and they'd be deeply interested in getting as many of their developers on board as possible.
Food for thought.
Feeding my thoughts is that the most likely Mac to get the 970 first is the Xserve. Panther would be sufficient to reschedule WWDC if it let the "Cat" out of the bag about the 970. If it has obvious 64-bitness in it's construction then the 970 is a GO!
...
Originally posted by Programmer
The information about motherboards isn't what I would call reliable, but the argument that Apple has used daughtercards in the past/current machines and will continue to on the new ones doesn't hold water. The new 970 bus design runs at 900 MHz (and more in the future) and thus has very tight timing & connection restrictions. It may simply not be possible to use a board-to-board connection between the processor(s) and the companion chip(s). The companion chip probably wants to be on the same board as its memory. At that point there is so much on the one board that they might as well just put everything on one board (or else all the I/O on a daughtercard if the companion chip isn't one big fully integrated monster). Alternatively the motherboard information might just be inaccurate, or might omit the fact that it comes in two pieces. The rumours are a bit thin, and the facts are even thinner.
Programmer, I did not mean to imply that since Apple has in the past used daughter cards they will in the future. But, rather that they have used them to great effect in the past, and that if possable I think that they will use them in the future. This gives them much more flexability in manufacturing and marketing computers, allowing them to cover a 1 to 4 processor range with 2 mother boards in stead of 4. I do see Apple moving beyond the dual configuration, they need it to address the video market and make a more powerfull pressence in the server market.
Some questions that need to be resolved are, can they put the companion chip on the daughter card, or can they design a companion chip that would be on the mother board which could cover both a single and dual configuration? For a Quad set-up would they need 2 companion chips, 4, or would they need yet another chip designed specifically for Quads? If a unique chip needs to be designed for each configuration then that creates some major problems in implementing MP systems. How does IBM handle this with the Power 4?
Nope, when the 970 comes, it will either arrive as a 1.2GHz 970 in a 15" PowerBook, or as the expected 1.4, 1.6, 1.8 configs bandied about in the PowerMacs, or both. Apple can't afford to strangle its sales by doing a slow roll out of the 970. They know that if they do sales will stagnate while consumers wait for the other shoe to drop.
If it is introduced in an out of the ordinary fashion, Apple might introduce it in workstation class machines to milk the bleeding edgers first, but I don't think even that would pay off as regular consumers would just bide their time waiting for it to trickle down.
Originally posted by Matsu
...Not good, considering that a single low speed ppc970 will not offer any kind of performance revelation over far cheaper x86 boxen...
We won't know actual performance for awhile yet, of course, but the ppc 970 @ 1.4GHz should be pretty respectable going by the (non-AV, non-SMP) estimated spec benchmarks. To the point that, yes, there will be cheaper x86 that beat it in test1, test2 etc., but... the way the features are aligned I do _not_ think that that it will be so completely outclassed at _all_ benchmarks.
That is:
_IF_ you have a program that really does benefit from 64-bits (Oracle)
_OR_ you have a program that needs lots of double precision FPU (Lightwave)
_OR_ you have Altivec programs that are crippled by the FSB (Anything Altivecable)
_THEN_ there's going to be three areas of drastic improvement.
There's also the recurring rumors that something _else_ is in the pipeline. GPU-as-CPU's-assistant, or multi-GPUs... If true (or something else interesting is going on) that's another enticing piece of the puzzle.
I don't _ever_ expect the PowerMac's (or whatever the 'pro' line is called) to beat x86 boxes on price.
The areas it's supposed to be competitive in are the (price/performance) and pure performance areas. The 970, even the lower clocked ones, should be quite good _at_specific_jobs_.
The Make-Matsu-Happy machine (a one-slot desktop with max CPUs & a $1000 price tag) isn't in the cards. The closest that will happen is a the iMacs dropping in price & improving in power as the 970 gets adopted. But Matsu, it _won't_ make you happy.
Speaking of cube form factors, the shuttle miniXPC form factor is steadily growing in PC circles -- 1 PCI, 1 AGP (often in addition to on-board graphics and sound) one CPU SOCKET, one optical bay and one/two HDD bays -- hmmm... why does that sound so familiar? Yeah, I must be making untowards demands on Apple again. Earth to Apple, 1000USD no longer describes a low-end or even a lower mid-range machine. A 1000 headless (monitorless) USD box is now firmly in the upper mid-range of desktop machinery. Most people aren't interested in a barebones tower at a workstation price, nor are they interested in an planned obsolescence AIO at a high-end tower price.
PRICE, OPTIONS, DISPLAY FLEXIBILITY/CHOICE, EXPANSION AND UPGRADABILITY! POWER, PRICE, PRICE, PRICE, PRICE!!!
Go ahead, get smug about it, I'm not the one losing market-share on a quaterly basis. Has Apple ever strung together more than 2 quaters of growth (as a percentage of marketshare) in the last 5 years? I don't care about excuses, if you can't build a price and feature competitive box then maybe you don't deserve to be in business.
Then they _won't_ sell me a desktop. Not that steve cares, but the way the numbers play out they won't sell a whole lot of other people desktops either
The only numbers that matter are Margin and Revenue. Steve doesn't care. He never has. He's always had above average tastes. Because Joe Bloe Walmart shopper doesn't share those those tastes is no concern of his.
Go ahead, get smug about it, I'm not the one losing market-share on a quaterly basis. Has Apple ever strung together more than 2 quaters of growth (as a percentage of marketshare) in the last 5 years? I don't care about excuses, if you can't build a price and feature competitive box then maybe you don't deserve to be in business.
Nevyn is not being smug. Whether Matsu buys a Mac or a PC has no bearing on what the rest of the 6 billion people on this earth do. You are entitled to govern your life as you see fit. If Apple is not in your plans then that's the way it is. The fact is Apple IS in business, they've acquired companies and are planned for future growth.
The thing is Matsu you are just a consumer. You think and react like one. You feel like Price is the biggest motivator and to Millions it is. But there also exist Millions who feel that a value propostion must meet more than just price.
despite your constant complaints about price you will not control the fate of Apple.
Originally posted by Programmer
Actually there was an interesting discussion on Ars about bus topologies for the 970. I had previously been assuming that a dual setup would either require a chipset with 2 ports, or 2 chipsets connected by a bridge. A third possibility was pointed out -- since the input and output are essentially on seperate buses, it would be possible for the protocol to operate in a "ring" fashion. The processors and chipset(s) would be connected in a loop and data would pass around from one device to the next (e.g. input connects upstream, output connects downstream). This has the advantage of being able to scale to multiple processors without having to re-engineer the companion chip. It does mean that bandwidth is shared and essentially 3.2 GB/sec instead of 6.4 GB/sec (@ 1.8 GHz), however.
That's 3.2GB/sec per processor, for a total bandwidth to the memory of 6.4GB/sec. If the bus was architected not as a ring but with two independent connections to the memory controller, then the total bandwidth would be 12.8GB/sec, which is more than even dual-channel DDR400 could provide. In other words - it doesn't really matter, because no memory can feed that right now. If future generations of the 970 retain the 4:1 DDR FSB, then we'll always have more bus bandwidth than memory bandwidth either way.
On a side note, it's a welcome change to have a faster bus than memory instead of the other way around.
Originally posted by Aphelion
Feeding my thoughts is that the most likely Mac to get the 970 first is the Xserve. Panther would be sufficient to reschedule WWDC if it let the "Cat" out of the bag about the 970. If it has obvious 64-bitness in it's construction then the 970 is a GO!
...
I can't agree I'm afraid...the point of the xserve is reliability not out and out speed.
As the situation stands the xserve is slightly slower than the top PM's.
There's no need to put a hot expensive chip in a server box that needs to be stable and problem free.