<strong>IBM are under a legal obligation, as a result of an anti-trust settlement, to announce (reasonably) accurately their forthcoming plans. If they say sampling in Q2 and production in Q3, a mere six months before the first of those dates, then they cannot wildly diverge without serious questions being asked.</strong><hr></blockquote>
In reality, this supports the argument that we will see volume production before the announced date. IBM has to be conservative (read allow for slippage) in its announcements in order to comply with the consent decree. But nothing says they can't ship product before the announced date. The consent decree merely prevents them from disseminating FUD -- announcing products that won't be ready for some time just to prevent sales going to the competition.
[quote]<strong>However, were they to make a variation on the 970 for a specific customer as a custom chip, they don't have to pre-announce it at all, and may be contractually required not to.
< Wild speculation mode> If Apple came along some years ago and spoke to IBM, saying "How about taking your plans for a POWER4 lite and making us a special version, dual core, shared L2, on-board memory controller, and our nice ApplePI interface?". IBM would probably have said "Show us the money and we'll do it." I like to think of it as the PPC 977, could appear any time, and, being a custom design, knowledge would not even be widespread within IBM. </Wild speculation mode></strong><hr></blockquote>
Yes but this would be extremely difficult to keep under wraps. You'd think we would of heard something about this by this late date.
Where I'd expect the PPC 970 to be a very nice chip is in 2 way, 4 way or 8 way solutions against the likes of Xeons or Opterons. I rather expect IBM is planning it that way too it is just a question of cost now. Personally I think that would be the best argument for why Apple should finally release some much higher end (and higher margin) workstations or servers using 4 or even 8 processors.<hr></blockquote>
As my signature has said since 1999, Apple has always had the power to win the MHz Wars by going multiple CPU's.
The by now obvious drawback to this approach has been the limitations of the bus architecture to support the processors. I was cheered to see the Powermac go to all duals, even with the bus limits imposed by the current motherboard. The main advantage right now of the duals is that it has encouraged the developers (like Adobe) to write multi-threaded applications.
The 970 will erase those limitations, and I expect Apple to jump on this, to finally, and for all time, erase the performance lead by Intel and AMD.
About cost: Many people assume that the cost of the IBM 970 will preclude it from being offered in dual, much less quad or octo configurations. I think that it may well be cheaper than the current Moto offerings because it is designed to be produced in a modern, high efficiency fab. Just going to the 300mm wafer insures significant cost savings, and a rapid move to a 90nm process will further increase per wafer yields and decrease cost.
The purchase of Nothing Real, and other high end applications insures that Apple will have a workstation class offering with at least 4 CPU's in my opinion. I don't expect that one to be cheap, but I do expect the dual Powermacs to be at the same, or lower pricing level.
Back on Topic: Although I share everyone's hope against hope that the 970 will be here soonest, Steve's proclaiming 2003 as "the year of the laptop" somewhat dashed those hopes for me and now my best and most optimistic prediction would be finally some mention, or better yet a Xsrve with the 970 announced at MWNY in July, with the Powermac 970 in September so the iMacs can go to the highest available (7457) speed for Xmas sales.
An IBM 970 dual Powermac by MWSF 2004 being the safer, albeit pessimistic, bet.
<strong>About cost: Many people assume that the cost of the IBM 970 will preclude it from being offered in dual, much less quad or octo configurations. I think that it may well be cheaper than the current Moto offerings because it is designed to be produced in a modern, high efficiency fab. Just going to the 300mm wafer insures significant cost savings, and a rapid move to a 90nm process will further increase per wafer yields and decrease cost.
</strong><hr></blockquote>
The 0.13 micron 970 shouldn't be any more expensive, roughly speaking, than the 7455 currently used in Apple's machines (I doubt it'll be much cheaper though). The 7457 will be cheaper & faster than the 7455 because of its migration to a 0.13 process, but it will go into lower end and portable machines.
SMP machines have their downsides too -- they don't benchmark that well since most benchmarks are single processor (at least so far), they generate twice the heat, they require a more complex motherboard or they have to share memory bandwidth. More than 2 G4's just doesn't make sense, and won't until they go with an RIO-based NUMA architecture. Certainly the 970 looks like it is ideally suited for large SMP machines, but those aren't going to be cheap machines.
Multi-core is going to start showing up on the desktop in the next couple of years, and IBM is definitely leading that charge. This has the advantage of keeping system complexity down while giving the advantages of SMP.
Back on Topic: Although I share everyone's hope against hope that the 970 will be here soonest, Steve's proclaiming 2003 as "the year of the laptop" somewhat dashed those hopes for me and now my best and most optimistic prediction would be finally some mention, or better yet a Xsrve with the 970 announced at MWNY in July, with the Powermac 970 in September so the iMacs can go to the highest available (7457) speed for Xmas sales.
SMP machines have their downsides too -- More than 2 G4's just doesn't make sense, and won't until they go with an RIO-based NUMA architecture.
Multi-core is going to start showing up on the desktop in the next couple of years, and IBM is definitely leading that charge. This has the advantage of keeping system complexity down while giving the advantages of SMP.<hr></blockquote>
I defer to your greater understanding of such things, programmer, and in 1999 when I first wrote my sig I did not have a full understanding of the technical difficulties in implementing quads and octos, but at that time the 7410 was the best offering and it is MERCI compliant (correct me if I'm wrong, or have the acronym scrambled).
But that was then and this is now... I have one question though, would two dual core CPU's be considered a "quad", and by extension four dual CPU's a "octo"?
By the way I had high hopes that Apple would buy SGI when it went to $.50 a share, and actually bought some SGI in hopes of that acquisition. It never happened, but I did make a small killing on SGI when I sold @ $4.33.
We've gotta be looking at a typical Apple scenario of July announcment with September shipping. I thnk multi-processor high end machines are very likely...Apple have been dancing with the high-end 3D and film editing companies for two years now and they HAVE to be telling those guys something seriously juicy. My personal wish...a 2/3U rackmount half the depth of the Xserve with quad 970's.
<strong>I defer to your greater understanding of such things, programmer, and in 1999 when I first wrote my sig I did not have a full understanding of the technical difficulties in implementing quads and octos, but at that time the 7410 was the best offering and it is MERCI compliant (correct me if I'm wrong, or have the acronym scrambled).
</strong><hr></blockquote>
Back in the days of the 604 the situation wasn't so bad -- processors couldn't gobble data as fast as a 1+ GHz G4. Since then, however, computing power has grown very quickly and the PowerPC memory interface hasn't really kept pace.
MERSI, by the way, describes the possible states of each chunk of cache (generally 32-byte blocks called "cache lines"). Each cache line is a copy of what is supposed to be in memory at some location. The memory copy isn't kept completely up to date at all times, however, because that would slow things down too much. Instead the cache keeps a few state bits, describing what is going on with the cache line (note that this is from memory so there might be inaccuracies):
M - modified, which means that this cache line contains the latest data and it has been modified from the value that was in RAM. Nobody else has a copy.
E - exclusive, which means that the cache line is unmodified but the only copy of this data is this one.
S - shared, which means that the cache line and at least one other processor's cache have unmodified copies of this data.
I - invalid, which means that this cache line currently isn't used.
R - reserved (only used by the 7400/7410, I believe), which means that this cache line has a modified copy of the data and it can provide it directly to a different processor's cache without updating memory in the process. This was a significant speed optimization since processor could trade data back and forth without having to wait for a RAM read-modify-write cycle. I'm not sure why it was dropped from later processors.
These MERSI bits are used to mark each cache line, and a processor on the bus watches all of the bus traffic from all processors (this is called bus snooping and is why it needs to be shared bus). When it sees a memory transaction involving data that it has in one of its cache lines, it can either intervene or just update its own MERSI bits to track what is going on.
[quote]<strong>
But that was then and this is now... I have one question though, would two dual core CPU's be considered a "quad", and by extension four dual CPU's a "octo"?
</strong><hr></blockquote>
I don't know that there is "official" terminology yet. My guess is that we'll start talking about the number of cores in a system. Its not going to be very long (<5 years) before we start seeing large numbers of cores on a single die. On the desktop this may replace having to have seperate processor chips. IBM is talking about having 16+ processors in a "cell" configuration on one die.
[quote]<strong>
By the way I had high hopes that Apple would buy SGI when it went to $.50 a share, and actually bought some SGI in hopes of that acquisition. It never happened, but I did make a small killing on SGI when I sold @ $4.33. </strong><hr></blockquote>
I've given up trying to have reasons while investing -- they are usually completely wrong, and whether I make money or not is completely independent of whether my reasons were right or wrong.
<strong>I've given up trying to have reasons while investing -- they are usually completely wrong, and whether I make money or not is completely independent of whether my reasons were right or wrong.</strong><hr></blockquote>
Yes, SGI went up more dramatically than I had even hoped for with an Apple buy-out (and attendant stock swap). Too bad my continued AAPL accumulation has not paid off yet. The 970 event (RSN) is what I'm hoping for to have a similar effect.
Yes but this would be extremely difficult to keep under wraps. You'd think we would of heard something about this by this late date.</strong><hr></blockquote>
Nobody got any sort of hint of the new Powerbooks 'til the night before though...
[quote]Nobody got any sort of hint of the new Powerbooks 'til the night before though...<hr></blockquote>The new powerbooks didn't need to have software developers qualify a "new" SIMD implementation or have hardware developers test for a whole new mobo architecture. Such large-scale changes should be much harder to keep under wraps.
802.11g was a bigger surprise as that is not even a ratified standard--yet.
Ok. This is a bad thread...let's face it. IBM and Apple surely aren't giving any clues as to when it will be introduced in a Powermac, assuming Apple will still call the workstations that , but we can see the only timelines put forth are IBM's 3Q estimate - July - Sept, so figure Apple, the biggest buyer of said chip, is going to get them a helluva lot sooner...so say Feb intro and March release. :eek:
<strong>Apple have been dancing with the high-end 3D and film editing companies for two years now and they HAVE to be telling those guys something seriously juicy.</strong><hr></blockquote>
As well as Apple gets along with the film and press industries I rather expect some of the greatest demand for them lies in the scientific communities, which traditionally use a unix workstation and a PC. OS X has really wrapped both into one and gives the opportunity for significant cost savings especially for the educational markets.
<strong>SMP machines have their downsides too ...benchmarks...heat...complex...but those aren't going to be cheap machines.</strong><hr></blockquote>
The point I see in there is that (assuming a SMP-ready sort of FSB) the 970 is very well situated for SMP.
All the issues you state do, of course, apply. But the competition is MP Xeons -> it is an area where Apple should be able to (handily!) compete in price based soley on relative CPU price.
<strong>Actually I'd rather see Apple use Power5 chips as high-high-end... Xserve for example!
It wouldnt require that much work to get the Power5 working with OSX.. Except of course the 64 bits, but 970 has already made that possible.. </strong><hr></blockquote>
I'm sure Apple will consider the possibility as soon as the POWER5 is a reality. I also suspect it won't be quite as "high-high-end" as you seem to think.
<strong>As for duals and quads not benching as well due to apps being written for single proc:
Virtual Single Processor is all i have to say.
VSD</strong><hr></blockquote>
I read the description of the VSD technology and it does not mean that unthreaded apps will suddenly use multiple processors. You still need to thread your app, but it makes the distribution of threads between processors transparent even across non-shared memory systems. If I'm wrong about this please point me at the link where they say different.
How much of a performance jump would one see using VSD? If I understand this right it is a way to run the same process through both CPUs making two CPUs a lot more like one really fast one. If VSD is being used on the 970 I would expect to see more than one dual powermac. I imagine it will be like today. Bottom line single, middle and top dual.
Question: Is there a way to run a dual system through two separate busses. Where each CPU has it's own bus and they are connect at the system controller? If apple ever mad quads for there Xserves would this not help performance? Just wondering.
Also can anyone come up with a good reason as to why the 7455 is running at 1.42GHz but we had rumors of Moto having trouble getting the 7457 past 1.42Ghz? Kind of weird in my opinion. I was wondering if apple perhaps turned down the 7457 in favor of the 970. Moto thus deciding not to bother with it.
Comments
<strong>
I think the marketing strategy has devolved more to "don't get slaughtered like arthritic, myopic old mutton."</strong><hr></blockquote>
Ow ow ow ow. <img src="graemlins/lol.gif" border="0" alt="[Laughing]" />
<strong>IBM are under a legal obligation, as a result of an anti-trust settlement, to announce (reasonably) accurately their forthcoming plans. If they say sampling in Q2 and production in Q3, a mere six months before the first of those dates, then they cannot wildly diverge without serious questions being asked.</strong><hr></blockquote>
In reality, this supports the argument that we will see volume production before the announced date. IBM has to be conservative (read allow for slippage) in its announcements in order to comply with the consent decree. But nothing says they can't ship product before the announced date. The consent decree merely prevents them from disseminating FUD -- announcing products that won't be ready for some time just to prevent sales going to the competition.
[quote]<strong>However, were they to make a variation on the 970 for a specific customer as a custom chip, they don't have to pre-announce it at all, and may be contractually required not to.
< Wild speculation mode> If Apple came along some years ago and spoke to IBM, saying "How about taking your plans for a POWER4 lite and making us a special version, dual core, shared L2, on-board memory controller, and our nice ApplePI interface?". IBM would probably have said "Show us the money and we'll do it." I like to think of it as the PPC 977, could appear any time, and, being a custom design, knowledge would not even be widespread within IBM. </Wild speculation mode></strong><hr></blockquote>
Yes but this would be extremely difficult to keep under wraps. You'd think we would of heard something about this by this late date.
Where I'd expect the PPC 970 to be a very nice chip is in 2 way, 4 way or 8 way solutions against the likes of Xeons or Opterons. I rather expect IBM is planning it that way too it is just a question of cost now. Personally I think that would be the best argument for why Apple should finally release some much higher end (and higher margin) workstations or servers using 4 or even 8 processors.<hr></blockquote>
As my signature has said since 1999, Apple has always had the power to win the MHz Wars by going multiple CPU's.
The by now obvious drawback to this approach has been the limitations of the bus architecture to support the processors. I was cheered to see the Powermac go to all duals, even with the bus limits imposed by the current motherboard. The main advantage right now of the duals is that it has encouraged the developers (like Adobe) to write multi-threaded applications.
The 970 will erase those limitations, and I expect Apple to jump on this, to finally, and for all time, erase the performance lead by Intel and AMD.
About cost: Many people assume that the cost of the IBM 970 will preclude it from being offered in dual, much less quad or octo configurations. I think that it may well be cheaper than the current Moto offerings because it is designed to be produced in a modern, high efficiency fab. Just going to the 300mm wafer insures significant cost savings, and a rapid move to a 90nm process will further increase per wafer yields and decrease cost.
The purchase of Nothing Real, and other high end applications insures that Apple will have a workstation class offering with at least 4 CPU's in my opinion. I don't expect that one to be cheap, but I do expect the dual Powermacs to be at the same, or lower pricing level.
Back on Topic: Although I share everyone's hope against hope that the 970 will be here soonest, Steve's proclaiming 2003 as "the year of the laptop" somewhat dashed those hopes for me and now my best and most optimistic prediction would be finally some mention, or better yet a Xsrve with the 970 announced at MWNY in July, with the Powermac 970 in September so the iMacs can go to the highest available (7457) speed for Xmas sales.
An IBM 970 dual Powermac by MWSF 2004 being the safer, albeit pessimistic, bet.
...
[ 01-26-2003: Message edited by: Aphelion ]</p>
<strong>About cost: Many people assume that the cost of the IBM 970 will preclude it from being offered in dual, much less quad or octo configurations. I think that it may well be cheaper than the current Moto offerings because it is designed to be produced in a modern, high efficiency fab. Just going to the 300mm wafer insures significant cost savings, and a rapid move to a 90nm process will further increase per wafer yields and decrease cost.
</strong><hr></blockquote>
The 0.13 micron 970 shouldn't be any more expensive, roughly speaking, than the 7455 currently used in Apple's machines (I doubt it'll be much cheaper though). The 7457 will be cheaper & faster than the 7455 because of its migration to a 0.13 process, but it will go into lower end and portable machines.
SMP machines have their downsides too -- they don't benchmark that well since most benchmarks are single processor (at least so far), they generate twice the heat, they require a more complex motherboard or they have to share memory bandwidth. More than 2 G4's just doesn't make sense, and won't until they go with an RIO-based NUMA architecture. Certainly the 970 looks like it is ideally suited for large SMP machines, but those aren't going to be cheap machines.
Multi-core is going to start showing up on the desktop in the next couple of years, and IBM is definitely leading that charge. This has the advantage of keeping system complexity down while giving the advantages of SMP.
<strong>
Back on Topic: Although I share everyone's hope against hope that the 970 will be here soonest, Steve's proclaiming 2003 as "the year of the laptop" somewhat dashed those hopes for me and now my best and most optimistic prediction would be finally some mention, or better yet a Xsrve with the 970 announced at MWNY in July, with the Powermac 970 in September so the iMacs can go to the highest available (7457) speed for Xmas sales.
[ 01-26-2003: Message edited by: Aphelion ]</strong><hr></blockquote>
2002 was the year of the TFT's, death of the CRT's and the eMac came
just a thought
SMP machines have their downsides too -- More than 2 G4's just doesn't make sense, and won't until they go with an RIO-based NUMA architecture.
Multi-core is going to start showing up on the desktop in the next couple of years, and IBM is definitely leading that charge. This has the advantage of keeping system complexity down while giving the advantages of SMP.<hr></blockquote>
I defer to your greater understanding of such things, programmer, and in 1999 when I first wrote my sig I did not have a full understanding of the technical difficulties in implementing quads and octos, but at that time the 7410 was the best offering and it is MERCI compliant (correct me if I'm wrong, or have the acronym scrambled).
But that was then and this is now... I have one question though, would two dual core CPU's be considered a "quad", and by extension four dual CPU's a "octo"?
By the way I had high hopes that Apple would buy SGI when it went to $.50 a share, and actually bought some SGI in hopes of that acquisition. It never happened, but I did make a small killing on SGI when I sold @ $4.33.
...
My personal wish...a 2/3U rackmount half the depth of the Xserve with quad 970's.<hr></blockquote>
Yes! and then put it in a vertical case and sell it as the PowerStation.
...
<strong>I defer to your greater understanding of such things, programmer, and in 1999 when I first wrote my sig I did not have a full understanding of the technical difficulties in implementing quads and octos, but at that time the 7410 was the best offering and it is MERCI compliant (correct me if I'm wrong, or have the acronym scrambled).
</strong><hr></blockquote>
Back in the days of the 604 the situation wasn't so bad -- processors couldn't gobble data as fast as a 1+ GHz G4. Since then, however, computing power has grown very quickly and the PowerPC memory interface hasn't really kept pace.
MERSI, by the way, describes the possible states of each chunk of cache (generally 32-byte blocks called "cache lines"). Each cache line is a copy of what is supposed to be in memory at some location. The memory copy isn't kept completely up to date at all times, however, because that would slow things down too much. Instead the cache keeps a few state bits, describing what is going on with the cache line (note that this is from memory so there might be inaccuracies):
M - modified, which means that this cache line contains the latest data and it has been modified from the value that was in RAM. Nobody else has a copy.
E - exclusive, which means that the cache line is unmodified but the only copy of this data is this one.
S - shared, which means that the cache line and at least one other processor's cache have unmodified copies of this data.
I - invalid, which means that this cache line currently isn't used.
R - reserved (only used by the 7400/7410, I believe), which means that this cache line has a modified copy of the data and it can provide it directly to a different processor's cache without updating memory in the process. This was a significant speed optimization since processor could trade data back and forth without having to wait for a RAM read-modify-write cycle. I'm not sure why it was dropped from later processors.
These MERSI bits are used to mark each cache line, and a processor on the bus watches all of the bus traffic from all processors (this is called bus snooping and is why it needs to be shared bus). When it sees a memory transaction involving data that it has in one of its cache lines, it can either intervene or just update its own MERSI bits to track what is going on.
[quote]<strong>
But that was then and this is now... I have one question though, would two dual core CPU's be considered a "quad", and by extension four dual CPU's a "octo"?
</strong><hr></blockquote>
I don't know that there is "official" terminology yet. My guess is that we'll start talking about the number of cores in a system. Its not going to be very long (<5 years) before we start seeing large numbers of cores on a single die. On the desktop this may replace having to have seperate processor chips. IBM is talking about having 16+ processors in a "cell" configuration on one die.
[quote]<strong>
By the way I had high hopes that Apple would buy SGI when it went to $.50 a share, and actually bought some SGI in hopes of that acquisition. It never happened, but I did make a small killing on SGI when I sold @ $4.33. </strong><hr></blockquote>
I've given up trying to have reasons while investing -- they are usually completely wrong, and whether I make money or not is completely independent of whether my reasons were right or wrong.
<strong>I've given up trying to have reasons while investing -- they are usually completely wrong, and whether I make money or not is completely independent of whether my reasons were right or wrong.</strong><hr></blockquote>
Yes, SGI went up more dramatically than I had even hoped for with an Apple buy-out (and attendant stock swap). Too bad my continued AAPL accumulation has not paid off yet. The 970 event (RSN) is what I'm hoping for to have a similar effect.
...
<strong>
Yes but this would be extremely difficult to keep under wraps. You'd think we would of heard something about this by this late date.</strong><hr></blockquote>
Nobody got any sort of hint of the new Powerbooks 'til the night before though...
802.11g was a bigger surprise as that is not even a ratified standard--yet.
[ 01-26-2003: Message edited by: cowerd ]</p>
<strong>Apple have been dancing with the high-end 3D and film editing companies for two years now and they HAVE to be telling those guys something seriously juicy.</strong><hr></blockquote>
As well as Apple gets along with the film and press industries I rather expect some of the greatest demand for them lies in the scientific communities, which traditionally use a unix workstation and a PC. OS X has really wrapped both into one and gives the opportunity for significant cost savings especially for the educational markets.
<strong>SMP machines have their downsides too ...benchmarks...heat...complex...but those aren't going to be cheap machines.</strong><hr></blockquote>
The point I see in there is that (assuming a SMP-ready sort of FSB) the 970 is very well situated for SMP.
All the issues you state do, of course, apply. But the competition is MP Xeons -> it is an area where Apple should be able to (handily!) compete in price based soley on relative CPU price.
It wouldnt require that much work to get the Power5 working with OSX.. Except of course the 64 bits, but 970 has already made that possible..
<strong>Actually I'd rather see Apple use Power5 chips as high-high-end... Xserve for example!
It wouldnt require that much work to get the Power5 working with OSX.. Except of course the 64 bits, but 970 has already made that possible..
I'm sure Apple will consider the possibility as soon as the POWER5 is a reality. I also suspect it won't be quite as "high-high-end" as you seem to think.
Virtual Single Processor is all i have to say.
VSD
<strong>As for duals and quads not benching as well due to apps being written for single proc:
Virtual Single Processor is all i have to say.
VSD</strong><hr></blockquote>
I read the description of the VSD technology and it does not mean that unthreaded apps will suddenly use multiple processors. You still need to thread your app, but it makes the distribution of threads between processors transparent even across non-shared memory systems. If I'm wrong about this please point me at the link where they say different.
Question: Is there a way to run a dual system through two separate busses. Where each CPU has it's own bus and they are connect at the system controller? If apple ever mad quads for there Xserves would this not help performance? Just wondering.
Also can anyone come up with a good reason as to why the 7455 is running at 1.42GHz but we had rumors of Moto having trouble getting the 7457 past 1.42Ghz? Kind of weird in my opinion. I was wondering if apple perhaps turned down the 7457 in favor of the 970. Moto thus deciding not to bother with it.