Diggin' the "trojan horse" strategy of licensing OS X server to IBM for their Power 4/5 servers. In through the back door, and bitch-slap all the IT windows dorks.
Holy Schnikies!!! The PPC 9x0 is poised to scale like a motherfscker! Never would have happend without Steve Jobs, that man is a genius. Seems very likely that it was a tough sell to IBM that Apple had the future potential to make designing a CPU for them worthwhile, but since Jobs has reworked Apple back into "his" image, w/ OS X, software, and a real direction...damn. The more I learn, the more this new Apple/IBM alliance keeps getting better and better!
Now I've got my own upgrade path: G4 400 MHz --> 1.33 GHz Gigadesigns upgrade card -->PPC 9x0 4 GHz.
Just waiting for the 970s to go on sale so the bottom drops out of the G4 upgrade card prices.
Pentium4 did it -- started at 1.4 GHz and is now at 3 GHz, heading up. IBM started considerably later, but as a result gains the advantage of having the more advanced processes more quickly. The pipelines on the 970 are roughly as long as those on the Pentium4, and we don't yet know enough about other internal details to have any idea how it will scale compared to the Intel design. The 4.5 GHz number is for the 9x0 chips, after the POWER5 arrives which is probably more than a year from now. The fact that the POWER4 has gained 70% in 2.5 years shows that IBM is capable of scaling quickly, even on a design intended more for reliability and robustness.
I totally agree with you programmer. Just wanted to bring myself down to earth for a while there. If IBM is able to make the 9x0 hit 4.5ghz in about 1.5 years, that would be impressive to say the least. Just imagine what performance these comps would have. I guess IBM already is working hard on it So if the 980 is able to reach 4-4.5 ghz by the beginning of 2005, that would give Apple impressive half-year upgrades if they follow their usaul upgrade cycle.
Second half 2003 - 970 introduced at 1.8
MWSF 2004 - 970 2.3/2.5 max
Second half 2004 - 970+ 3+
MWSF 2005 - 980? 4.0-4.5
With this rate of improvements I would think that a lot of people would be tempted to replace their pm's more frequently than has been the case for the last few years. With these improvements in speed, there would be a real "hands on" performance improvement with every update. I think a lot of people would be tempted to update they set-up a little ofter than has been the case for the last couple 'o years.
And we must not forget that they may already be getting a nice yield of ~2 ghz 970's. The 1.8 ghz introduction speed has been the quoted max speed for over a year now, and much can happen in this time, so it's not impossible that we're in for a little surprise. But then again, if the yields of these processors is not that great, Apple would be better off waiting for the yields to improve, or wait until IBM has fine-tuned their prod. process to get greater yeilds.
And with the possibility of the 970 reaching the vicinity of 2.5 ghz by christmas, they would have a nice amount of "lower-end" processors in the 1.4 - 1.8 range that would make a great fit for both powerbooks and iMacs.
Wasn't the Power5 (and I assume the 980) suppose to have some form of hyperthreading? Can't seem to locate any info and don't think that was the term they used.
If you mean multithreading there's some info here but not much:
Programmer I think is on the right path here. It's not OS X or the technical details of 970 that leads me to think that Apple will soon be gunning for SGI and HP
Talk about kicking a company when they are down... gunning for SGI would simply be heartless.
4) Apple just dropped the price points significantly. For those of you expecting a $5000 970 - you're wrong. Apple pretty much keeps price points steady and simply improves the package for the price - that's what all of those nutty deals are for RAM, Office, etc. even for brand-new systems. Apple isn't blowing out excess hardware - they're sweetening the pot so they don't have to move price points. Don't forget what happens when Apple raises prices - even a little. The large drop in price points reflects Apple ramping up to compete more broadly, and the shift of profits from the core desktop/laptop line to software and fringe products (iPod, music, etc.) January/February product revs and price drops was a warning shot for everybody that big things were coming in summer:
"Now Apple makes almost as much operating profit on each iPod it sells as it does on each iMac, even though the iPod costs a fraction as much to manufacture." Fortune article
The more Apple isn't dependent on the desktop, laptop, and server lines for profits, the more licensing makes sense. More and more Apple can compete with Dell by operating like Microsoft. So, once Apple cuts it's profit margins on conventional hardware to the 10% range, licensing can turn on with minimal risk. Even if IBM does sell a consumer box, Apple will still rake it in with iPods, .mac, iLife, and everything else.
OS X 10.3 + X11 + 970 = something IBM can finally use. I think the license was part of the deal for getting the 970 in our hands. At least I dearly hope so...
You know, Johnsonwax? That IS the best post I've seen on the 'insider boards to date...and I've see a few good 'uns.
Finally, someone who finally 'gets' the 'new' Apple.
If you read Simon Jary's insightful-as-ever' editorials for Macworld UK you can see that Apple is moving 'away' from its traditional philosophies to incorporate a broader 'non-Mac' userbase as part of its strategy for profit and growth.
For die-hard Mac-heads on this forum, it may be hard to swallow or take.
But, iPods for Windows, X-Serves for Enterprise, a Music store for Mac AND Windows. Internet Explorer is about to be booted off 'X' coinciding with Panther? An alliance with the old enemy IBM with 'X' licensing? That is one powerful alliance and it helps IBM's coffers but really helps Apple break into new markets. The 'lucrative' 'HP' 'boost-your-yearly-bottom-line' market.
What next, 'iOffice' to take on M$? A low-end headless Mac? Mac for Windows?
For that puny 2% to turn into 10%...something drastic is going to have to happen. Apple needs to map out new markets and a 'newly confident' Apple can dominate these markets. Apple have found you can't make enough money from an ever-loyal but shrinking userbase. Everybody now has a computer that is fast enough. You can't make a profit from a market unless you're actually in it. iPod rules! M$ nil. Apple will maintain and redefine traditional markets. The process has already begun. The 'last' sacred cow is on its way to the abatoir (slaughter house...) I can hardly wait.
It does both single- and double-precision, depending. A lot could be made of the single-precision Altivec support considering that a lot of FEA code can be efficiently parallellized, though I can't guess as to how much time it spends working on single-precision (yet another point for double-precision Altivec)
We would not want double-precision within AltiVec. There are more effective ways to utilize the silicon. We've been through this.. Well, unless AltiVec was widened to 256bit (Programmer?). Search a few of my other posts. It's been discussed here to death ;-)
I think one thing you have to keep in mind is that CAD, unlike photorealistic 3D rendering/modeling is not an area where fast double-precision calcs are necessary, but even if they were, I'm sure that the 970 will have some impressive double-precision scores anyway.
On the idea that Apple might license MacOS X to IBM...
Programmer gets the idea.
McCrab gets the idea.
Johnsonwax, you get it...
Johnsonwax writes:
Quote:
Lets face it, though. We've had 64 bit workstations with how many unix OSes and variants now? And how many are still out there? How many has IBM supported? OS X is something different and special here and they might see that.
Yeah, it's not Winders... And I think IBM is still pissed that Micro$oft screwed them over OS/2. It might be a little about "get-back"
Yep, that's what I thought I had seen, or try this one.
I hope this makes it into the 980 along with a dual-core.
Boy Mac people are sure quick to jump into the megahertz myth. How time (and bandwagons) change in the tech world.
Multithreading sounds really good on top of the higher CPU speed. What are they going to use for a FSB multiplier with a MP 2.5 GHZ machine let alone a 4.5 GHZ?
Boy Mac people are sure quick to jump into the megahertz myth. How time (and bandwagons) change in the tech world.
We're not comparing PowerPC's against x86's here, we're comparing PowerPC's with PowerPC's, and in that case, there's much more to MHz than mere myth ...
We're not comparing PowerPC's against x86's here, we're comparing PowerPC's with PowerPC's, and in that case, there's much more to MHz than mere myth ...
We would not want double-precision within AltiVec. There are more effective ways to utilize the silicon. We've been through this.. Well, unless AltiVec was widened to 256bit (Programmer?). Search a few of my other posts. It's been discussed here to death ;-)
That's all I was getting at. You're right, let's not open that can up again.
Quote:
On the idea that Apple might license MacOS X to IBM...
Yeah, it's not Winders... And I think IBM is still pissed that Micro$oft screwed them over OS/2. It might be a little about "get-back"
Nah, companies like IBM don't hold grudges. They might, however, currently be a little pissed about MS licensing and interaction. If IBM can sell baby-POWER workstations and reduce the MS tax, and keep a little extra profit because they made the CPU, well, that's a whole other thing.
There is one other thing that I didn't mention in my previous post, is that the same argument for licensing the OS to a company like IBM also extends to licensing the OS to a company like IBM for running on non-PowerPC hardware - provided that the success of the licensing plan isn't dependent on consumer software. It would only work initially for X11, Cocoa, and non-GUI apps, so IBM might again be the right place to look.
Multithreading sounds really good on top of the higher CPU speed. What are they going to use for a FSB multiplier with a MP 2.5 GHZ machine let alone a 4.5 GHZ?
Well, assuming they keep the same system for determining, a 2.5GHz 970 would have a 1.25GHz FSB (effective, 625.5 MHz real). A 4.5GHz 970+/980 would have a 2.25GHz bus (effective 1.125 GHz real).
Well, assuming they keep the same system for determining, a 2.5GHz 970 would have a 1.25GHz FSB (effective, 625.5 MHz real). A 4.5GHz 970+/980 would have a 2.25GHz bus (effective 1.125 GHz real).
My point was: What are they going to use for a memory system.
Looks like Hannibal has finally finished the second of his two 970 articles. 11,000 VERY interesting words to throw on the performance debate fire. Looks like integer and altivec performance is not as good as expected but the FPU and bus show great promise
Thanks Programmer, Amorph, et alii. Your comments have helped me better understand the discussion in this thread.
I don't however agree that IBM would license OSX from Apple. IBM went on record with their intent to put the 970 in blades for their linux customers. Aside from serious linux geeks who might use linux for their primary OS on the desktop, I could be wrong, but I would hazard a guess that most that would run linux would do so for headless servers and clusters. Such installations are neither the abode of Apple nor MS. With the spate of articles recently regarding IBM withholding the rather impressive performance of some their top of the line Intel gear, it is the general concensus that IBM only offers Intel cpus to the customer that absolutely positively has to have it, to insure that either way IBM makes the sale. However, who'd doubt that Big Blue would brook such insurgence for long as clearly at IBM, IBM iron will be supreme and that is just the start. I wouldn't recommend that Apple even attempt to jump into that battle if they can help it.
I don't think Apple and IBM have the same customers typically. It would make sense perhaps to buy a truck load of IBM blades and then maybe sell a 970 Mac as the sysops desktop machine, from which they'd administer the drones. IBM and Apple should have a mutual sales agreement where they are each other's authorized sellers, giving them the ability to offer each others wares. From what I've read, IBM's 970s blades won't have VMX/Altivec onboard. Thus if a Blue customer needs it, sale Apple's kit. Conversely, were one to approach Apple, but they don't need VMX, sale IBM's kit. I more than anyone would love to see Apple get every sale it possibly can, but I think right now, Apple needs a great relationship with their supplier. That supplier just happens to be Big Blue and with Apple needing a foothold into IT departments, IBM is the kind of credential they need. Then who knows, IBM gets the sale either way, maybe they don't care. >>my head hurts!!<<
Most imperative, IMHO, is that Apple and IBM begin a PPC 970/980/990 "Intel Inside" type branding campaign. Hell with the Power 4/5 heritage, why not build on big brothers success. Almost as with the Cray when the G4 first premiered. I read over at Macosrumors that Apple may drop the "G5" altogether, which I encourage. The curtain is coming down around the MHz myth as I've read somewhere or other that there are some trying to get Intel to sell Centrino (Pentium M(based on PIII core more efficient design than PIV)) for the desktop in that even though it is clocked lower it actually performs better.
Whatever the case, Intel is vulnerable right now and it looks like the world is ready for a little diversification. Anyone that steps right can make quite a bit of head way. Between AMD with X86-64 and IBM's like strategy, the market will be getting the same message from both of Intel's biggest competitors. This could be as big as Windows was to the IBM landscape. INDEED.
I don't however agree that IBM would license OSX from Apple. IBM went on record with their intent to put the 970 in blades for their linux customers. Aside from serious linux geeks who might use linux for their primary OS on the desktop, I could be wrong, but I would hazard a guess that most that would run linux would do so for headless servers and clusters. Such installations are neither the abode of Apple nor MS. With the spate of articles recently regarding IBM withholding the rather impressive performance of some their top of the line Intel gear, it is the general concensus that IBM only offers Intel cpus to the customer that absolutely positively has to have it, to insure that either way IBM makes the sale. However, who'd doubt that Big Blue would brook such insurgence for long as clearly at IBM, IBM iron will be supreme and that is just the start. I wouldn't recommend that Apple even attempt to jump into that battle if they can help it.
Actually, the whole reason for Apple to license OS X Server to IBM is to avoid jumping into that battle themselves. Apple has access to robust and battle-tested clustering technology, courtesy of NeXT. The NeXT kernel (Mach) could scale up to 24 processors. The potential is there. The main concern would be Apple not treading directly on IBM's turf (by releasing big hardware) and IBM not treading directly on Apple's. Mac OS X Server also brings something to the table that IBM currently can't offer, which is a drop-in, secure, plug-and-play UNIX. If Apple makes inroads into professional markets like video, those markets probably need servers too, and IBM could offer seamlessly OS X compatible servers. IBM doesn't care, as long as you buy from them. For Apple, it's another line of revenue they didn't have before, and a way to get past one of their most annoying obstacles: The sheer number of Windows servers out there that treat Macs as second-class citizens.
Quote:
I don't think Apple and IBM have the same customers typically. It would make sense perhaps to buy a truck load of IBM blades and then maybe sell a 970 Mac as the sysops desktop machine, from which they'd administer the drones. IBM and Apple should have a mutual sales agreement where they are each other's authorized sellers, giving them the ability to offer each others wares. From what I've read, IBM's 970s blades won't have VMX/Altivec onboard. Thus if a Blue customer needs it, sale Apple's kit.
This is exactly it: The mutual interest lies in the fact that Apple's and IBM's kit is different, but mutually compatible. Servers are getting more and more common, now that network bandwidth no longer sucks. It's quite common for people to open and save documents directly to network drives and/or databases (e.g. Cumulus), and forms applications are increasingly web-driven.
The 970 is the 970. I don't think IBM's engineers "get" SIMD yet, so the VMX unit on board might spend a lot of time twiddling its thumbs in IBM machines. But the functionality should be there. There are currently no indications whatever that a SIMD-less 970 exists.
Even in IBM applications, the VMX (AltiVec) unit might get used by customers, or third-party apps, or even by Red Hat Linux compiled with auto-vectorization (which might not help much at all, since vectorization is still best done by hand, but it might eke out a bit more performance). The pure-RISC mindset will die hard within IBM.
Quote:
Most imperative, IMHO, is that Apple and IBM begin a PPC 970/980/990 "Intel Inside" type branding campaign.
They actually did that way back when, with the PowerPC logo (and, of course, the name "Power Macintosh"). It would be interesting to see them bring that back, and it might happen. I believe "PowerPC" is IBM's trademark.
The 970 is the 970. I don't think IBM's engineers "get" SIMD yet, so the VMX unit on board might spend a lot of time twiddling its thumbs in IBM machines. But the functionality should be there. There are currently no indications whatever that a SIMD-less 970 exists.
Even in IBM applications, the VMX (AltiVec) unit might get used by customers, or third-party apps, or even by Red Hat Linux compiled with auto-vectorization (which might not help much at all, since vectorization is still best done by hand, but it might eke out a bit more performance). The pure-RISC mindset will die hard within IBM.
I think eventually, IBM will get a clue when they see what Apple is doing with SIMD, clustering and the scientific community - especially in BLAST software, etc - in the meantime, if some at IBM want to pretend it's still 1998 - that's just more scientists in the Apple camp.
Anybody 'round here ever even tried setting up a Beowulf cluster?
Anybody 'round here ever even tried setting up a Beowulf cluster?
Exactly.
Even if OS X can't muster the raw performance of Linux, and even if the hardware isn't quite as cheap as the sort of stuff usually used to build Beowulf clusters, do not underestimate the appeal of an Apple-slick, Rendezvous-enabled clustering solution. People have spent money for convenience since the beginning of time, and I'm sure there are plenty of scientists who'd rather spend their time doing science (or, say, 3D artists who'd rather spend their time doing 3D art) and have the machines just work. The huge appeal of Macs in places like NASA and Lawrence Livermore is a testimony to that fact.
Which is actually quite strange. POWER and PowerPC aren't "pure-RISC" in the original sense of the term RISC. Their instruction set has hundreds of instructions, some of which are quite powerful. The POWER and PowerPC instruction sets are really just designed to be more orthogonal, based around a fixed instruction word size, and using a load/store architecture. The VMX design uses exactly the same principles, just extending it to the problem of dealing with streaming SIMD data.
I think the hold-outs at IBM are just the ones whose problems don't lend themselves to vector implementation. Or at least they don't think they do. IBM has clearly changed its perspective, however, since a major feature of POWER5 is this mysterious "FastPath" that they keep talking about.
Comments
Holy Schnikies!!! The PPC 9x0 is poised to scale like a motherfscker! Never would have happend without Steve Jobs, that man is a genius. Seems very likely that it was a tough sell to IBM that Apple had the future potential to make designing a CPU for them worthwhile, but since Jobs has reworked Apple back into "his" image, w/ OS X, software, and a real direction...damn. The more I learn, the more this new Apple/IBM alliance keeps getting better and better!
Now I've got my own upgrade path: G4 400 MHz --> 1.33 GHz Gigadesigns upgrade card -->PPC 9x0 4 GHz.
Just waiting for the 970s to go on sale so the bottom drops out of the G4 upgrade card prices.
Originally posted by Programmer
Pentium4 did it -- started at 1.4 GHz and is now at 3 GHz, heading up. IBM started considerably later, but as a result gains the advantage of having the more advanced processes more quickly. The pipelines on the 970 are roughly as long as those on the Pentium4, and we don't yet know enough about other internal details to have any idea how it will scale compared to the Intel design. The 4.5 GHz number is for the 9x0 chips, after the POWER5 arrives which is probably more than a year from now. The fact that the POWER4 has gained 70% in 2.5 years shows that IBM is capable of scaling quickly, even on a design intended more for reliability and robustness.
I totally agree with you programmer. Just wanted to bring myself down to earth for a while there. If IBM is able to make the 9x0 hit 4.5ghz in about 1.5 years, that would be impressive to say the least. Just imagine what performance these comps would have. I guess IBM already is working hard on it
Second half 2003 - 970 introduced at 1.8
MWSF 2004 - 970 2.3/2.5 max
Second half 2004 - 970+ 3+
MWSF 2005 - 980? 4.0-4.5
With this rate of improvements I would think that a lot of people would be tempted to replace their pm's more frequently than has been the case for the last few years. With these improvements in speed, there would be a real "hands on" performance improvement with every update. I think a lot of people would be tempted to update they set-up a little ofter than has been the case for the last couple 'o years.
And we must not forget that they may already be getting a nice yield of ~2 ghz 970's. The 1.8 ghz introduction speed has been the quoted max speed for over a year now, and much can happen in this time, so it's not impossible that we're in for a little surprise. But then again, if the yields of these processors is not that great, Apple would be better off waiting for the yields to improve, or wait until IBM has fine-tuned their prod. process to get greater yeilds.
And with the possibility of the 970 reaching the vicinity of 2.5 ghz by christmas, they would have a nice amount of "lower-end" processors in the 1.4 - 1.8 range that would make a great fit for both powerbooks and iMacs.
And man, that Dual 4.5 980 is gonna rock!!!
Originally posted by Bigc
Wasn't the Power5 (and I assume the 980) suppose to have some form of hyperthreading? Can't seem to locate any info and don't think that was the term they used.
If you mean multithreading there's some info here but not much:
IBM Power5 will be multicore and multithread - the inquirer
Some News about the IBM Power5 processor - hoise
Originally posted by johnsonwax
Programmer I think is on the right path here. It's not OS X or the technical details of 970 that leads me to think that Apple will soon be gunning for SGI and HP
Talk about kicking a company when they are down... gunning for SGI would simply be heartless.
Talk about kicking a company when they are down... gunning for SGI would simply be heartless.
Pound them to crap and buy Maya from them very cheaply?
Lemon Bon Bon 8)
4) Apple just dropped the price points significantly. For those of you expecting a $5000 970 - you're wrong. Apple pretty much keeps price points steady and simply improves the package for the price - that's what all of those nutty deals are for RAM, Office, etc. even for brand-new systems. Apple isn't blowing out excess hardware - they're sweetening the pot so they don't have to move price points. Don't forget what happens when Apple raises prices - even a little. The large drop in price points reflects Apple ramping up to compete more broadly, and the shift of profits from the core desktop/laptop line to software and fringe products (iPod, music, etc.) January/February product revs and price drops was a warning shot for everybody that big things were coming in summer:
"Now Apple makes almost as much operating profit on each iPod it sells as it does on each iMac, even though the iPod costs a fraction as much to manufacture." Fortune article
The more Apple isn't dependent on the desktop, laptop, and server lines for profits, the more licensing makes sense. More and more Apple can compete with Dell by operating like Microsoft. So, once Apple cuts it's profit margins on conventional hardware to the 10% range, licensing can turn on with minimal risk. Even if IBM does sell a consumer box, Apple will still rake it in with iPods, .mac, iLife, and everything else.
OS X 10.3 + X11 + 970 = something IBM can finally use. I think the license was part of the deal for getting the 970 in our hands. At least I dearly hope so...
You know, Johnsonwax? That IS the best post I've seen on the 'insider boards to date...and I've see a few good 'uns.
Finally, someone who finally 'gets' the 'new' Apple.
If you read Simon Jary's insightful-as-ever' editorials for Macworld UK you can see that Apple is moving 'away' from its traditional philosophies to incorporate a broader 'non-Mac' userbase as part of its strategy for profit and growth.
For die-hard Mac-heads on this forum, it may be hard to swallow or take.
But, iPods for Windows, X-Serves for Enterprise, a Music store for Mac AND Windows. Internet Explorer is about to be booted off 'X' coinciding with Panther? An alliance with the old enemy IBM with 'X' licensing? That is one powerful alliance and it helps IBM's coffers but really helps Apple break into new markets. The 'lucrative' 'HP' 'boost-your-yearly-bottom-line' market.
What next, 'iOffice' to take on M$? A low-end headless Mac? Mac for Windows?
For that puny 2% to turn into 10%...something drastic is going to have to happen. Apple needs to map out new markets and a 'newly confident' Apple can dominate these markets. Apple have found you can't make enough money from an ever-loyal but shrinking userbase. Everybody now has a computer that is fast enough. You can't make a profit from a market unless you're actually in it. iPod rules! M$ nil. Apple will maintain and redefine traditional markets. The process has already begun. The 'last' sacred cow is on its way to the abatoir (slaughter house...) I can hardly wait.
'Here's daddy!'
Lemon Bon Bon
It does both single- and double-precision, depending. A lot could be made of the single-precision Altivec support considering that a lot of FEA code can be efficiently parallellized, though I can't guess as to how much time it spends working on single-precision (yet another point for double-precision Altivec)
We would not want double-precision within AltiVec. There are more effective ways to utilize the silicon. We've been through this.. Well, unless AltiVec was widened to 256bit (Programmer?). Search a few of my other posts. It's been discussed here to death ;-)
I think one thing you have to keep in mind is that CAD, unlike photorealistic 3D rendering/modeling is not an area where fast double-precision calcs are necessary, but even if they were, I'm sure that the 970 will have some impressive double-precision scores anyway.
On the idea that Apple might license MacOS X to IBM...
Programmer gets the idea.
McCrab gets the idea.
Johnsonwax, you get it...
Johnsonwax writes:
Lets face it, though. We've had 64 bit workstations with how many unix OSes and variants now? And how many are still out there? How many has IBM supported? OS X is something different and special here and they might see that.
Yeah, it's not Winders... And I think IBM is still pissed that Micro$oft screwed them over OS/2. It might be a little about "get-back"
\t
--
Ed
Originally posted by NETROMac
If you mean multithreading there's some info here but not much:
IBM Power5 will be multicore and multithread - the inquirer
Some News about the IBM Power5 processor - hoise
Yep, that's what I thought I had seen, or try this one.
I hope this makes it into the 980 along with a dual-core.
Boy Mac people are sure quick to jump into the megahertz myth.
Multithreading sounds really good on top of the higher CPU speed. What are they going to use for a FSB multiplier with a MP 2.5 GHZ machine let alone a 4.5 GHZ?
Originally posted by Bigc
Boy Mac people are sure quick to jump into the megahertz myth.
We're not comparing PowerPC's against x86's here, we're comparing PowerPC's with PowerPC's, and in that case, there's much more to MHz than mere myth ...
Originally posted by OverToasty
We're not comparing PowerPC's against x86's here, we're comparing PowerPC's with PowerPC's, and in that case, there's much more to MHz than mere myth ...
Oh, right , I get it
Originally posted by Lemon Bon Bon
For die-hard Mac-heads on this forum, it may be hard to swallow or take.
...but not for shareholders or others wanting to see Apple succeed in the long-run.
Originally posted by Ed M.
We would not want double-precision within AltiVec. There are more effective ways to utilize the silicon. We've been through this.. Well, unless AltiVec was widened to 256bit (Programmer?). Search a few of my other posts. It's been discussed here to death ;-)
That's all I was getting at. You're right, let's not open that can up again.
On the idea that Apple might license MacOS X to IBM...
Yeah, it's not Winders... And I think IBM is still pissed that Micro$oft screwed them over OS/2. It might be a little about "get-back"
Nah, companies like IBM don't hold grudges. They might, however, currently be a little pissed about MS licensing and interaction. If IBM can sell baby-POWER workstations and reduce the MS tax, and keep a little extra profit because they made the CPU, well, that's a whole other thing.
There is one other thing that I didn't mention in my previous post, is that the same argument for licensing the OS to a company like IBM also extends to licensing the OS to a company like IBM for running on non-PowerPC hardware - provided that the success of the licensing plan isn't dependent on consumer software. It would only work initially for X11, Cocoa, and non-GUI apps, so IBM might again be the right place to look.
Originally posted by Bigc
Multithreading sounds really good on top of the higher CPU speed. What are they going to use for a FSB multiplier with a MP 2.5 GHZ machine let alone a 4.5 GHZ?
Well, assuming they keep the same system for determining, a 2.5GHz 970 would have a 1.25GHz FSB (effective, 625.5 MHz real). A 4.5GHz 970+/980 would have a 2.25GHz bus (effective 1.125 GHz real).
Originally posted by kupan787
Well, assuming they keep the same system for determining, a 2.5GHz 970 would have a 1.25GHz FSB (effective, 625.5 MHz real). A 4.5GHz 970+/980 would have a 2.25GHz bus (effective 1.125 GHz real).
My point was: What are they going to use for a memory system.
Originally posted by jamm
Looks like Hannibal has finally finished the second of his two 970 articles. 11,000 VERY interesting words to throw on the performance debate fire. Looks like integer and altivec performance is not as good as expected but the FPU and bus show great promise
Second Ars PPC 970 Article
Unfotunately, the basis for the article is speculation (though to his credit, Hannibal does acknowledge this).
I don't however agree that IBM would license OSX from Apple. IBM went on record with their intent to put the 970 in blades for their linux customers. Aside from serious linux geeks who might use linux for their primary OS on the desktop, I could be wrong, but I would hazard a guess that most that would run linux would do so for headless servers and clusters. Such installations are neither the abode of Apple nor MS. With the spate of articles recently regarding IBM withholding the rather impressive performance of some their top of the line Intel gear, it is the general concensus that IBM only offers Intel cpus to the customer that absolutely positively has to have it, to insure that either way IBM makes the sale. However, who'd doubt that Big Blue would brook such insurgence for long as clearly at IBM, IBM iron will be supreme and that is just the start. I wouldn't recommend that Apple even attempt to jump into that battle if they can help it.
I don't think Apple and IBM have the same customers typically. It would make sense perhaps to buy a truck load of IBM blades and then maybe sell a 970 Mac as the sysops desktop machine, from which they'd administer the drones. IBM and Apple should have a mutual sales agreement where they are each other's authorized sellers, giving them the ability to offer each others wares. From what I've read, IBM's 970s blades won't have VMX/Altivec onboard. Thus if a Blue customer needs it, sale Apple's kit. Conversely, were one to approach Apple, but they don't need VMX, sale IBM's kit. I more than anyone would love to see Apple get every sale it possibly can, but I think right now, Apple needs a great relationship with their supplier. That supplier just happens to be Big Blue and with Apple needing a foothold into IT departments, IBM is the kind of credential they need. Then who knows, IBM gets the sale either way, maybe they don't care.
Most imperative, IMHO, is that Apple and IBM begin a PPC 970/980/990 "Intel Inside" type branding campaign. Hell with the Power 4/5 heritage, why not build on big brothers success. Almost as with the Cray when the G4 first premiered. I read over at Macosrumors that Apple may drop the "G5" altogether, which I encourage. The curtain is coming down around the MHz myth as I've read somewhere or other that there are some trying to get Intel to sell Centrino (Pentium M(based on PIII core more efficient design than PIV)) for the desktop in that even though it is clocked lower it actually performs better.
Whatever the case, Intel is vulnerable right now and it looks like the world is ready for a little diversification. Anyone that steps right can make quite a bit of head way. Between AMD with X86-64 and IBM's like strategy, the market will be getting the same message from both of Intel's biggest competitors. This could be as big as Windows was to the IBM landscape. INDEED.
Originally posted by ArkAngel
I don't however agree that IBM would license OSX from Apple. IBM went on record with their intent to put the 970 in blades for their linux customers. Aside from serious linux geeks who might use linux for their primary OS on the desktop, I could be wrong, but I would hazard a guess that most that would run linux would do so for headless servers and clusters. Such installations are neither the abode of Apple nor MS. With the spate of articles recently regarding IBM withholding the rather impressive performance of some their top of the line Intel gear, it is the general concensus that IBM only offers Intel cpus to the customer that absolutely positively has to have it, to insure that either way IBM makes the sale. However, who'd doubt that Big Blue would brook such insurgence for long as clearly at IBM, IBM iron will be supreme and that is just the start. I wouldn't recommend that Apple even attempt to jump into that battle if they can help it.
Actually, the whole reason for Apple to license OS X Server to IBM is to avoid jumping into that battle themselves. Apple has access to robust and battle-tested clustering technology, courtesy of NeXT. The NeXT kernel (Mach) could scale up to 24 processors. The potential is there. The main concern would be Apple not treading directly on IBM's turf (by releasing big hardware) and IBM not treading directly on Apple's. Mac OS X Server also brings something to the table that IBM currently can't offer, which is a drop-in, secure, plug-and-play UNIX. If Apple makes inroads into professional markets like video, those markets probably need servers too, and IBM could offer seamlessly OS X compatible servers. IBM doesn't care, as long as you buy from them.
I don't think Apple and IBM have the same customers typically. It would make sense perhaps to buy a truck load of IBM blades and then maybe sell a 970 Mac as the sysops desktop machine, from which they'd administer the drones. IBM and Apple should have a mutual sales agreement where they are each other's authorized sellers, giving them the ability to offer each others wares. From what I've read, IBM's 970s blades won't have VMX/Altivec onboard. Thus if a Blue customer needs it, sale Apple's kit.
This is exactly it: The mutual interest lies in the fact that Apple's and IBM's kit is different, but mutually compatible. Servers are getting more and more common, now that network bandwidth no longer sucks. It's quite common for people to open and save documents directly to network drives and/or databases (e.g. Cumulus), and forms applications are increasingly web-driven.
The 970 is the 970. I don't think IBM's engineers "get" SIMD yet, so the VMX unit on board might spend a lot of time twiddling its thumbs in IBM machines. But the functionality should be there. There are currently no indications whatever that a SIMD-less 970 exists.
Even in IBM applications, the VMX (AltiVec) unit might get used by customers, or third-party apps, or even by Red Hat Linux compiled with auto-vectorization (which might not help much at all, since vectorization is still best done by hand, but it might eke out a bit more performance). The pure-RISC mindset will die hard within IBM.
Most imperative, IMHO, is that Apple and IBM begin a PPC 970/980/990 "Intel Inside" type branding campaign.
They actually did that way back when, with the PowerPC logo (and, of course, the name "Power Macintosh"). It would be interesting to see them bring that back, and it might happen. I believe "PowerPC" is IBM's trademark.
Originally posted by Amorph
The 970 is the 970. I don't think IBM's engineers "get" SIMD yet, so the VMX unit on board might spend a lot of time twiddling its thumbs in IBM machines. But the functionality should be there. There are currently no indications whatever that a SIMD-less 970 exists.
Even in IBM applications, the VMX (AltiVec) unit might get used by customers, or third-party apps, or even by Red Hat Linux compiled with auto-vectorization (which might not help much at all, since vectorization is still best done by hand, but it might eke out a bit more performance). The pure-RISC mindset will die hard within IBM.
I think eventually, IBM will get a clue when they see what Apple is doing with SIMD, clustering and the scientific community - especially in BLAST software, etc - in the meantime, if some at IBM want to pretend it's still 1998 - that's just more scientists in the Apple camp.
Anybody 'round here ever even tried setting up a Beowulf cluster?
eeeach! ...
Originally posted by OverToasty
Anybody 'round here ever even tried setting up a Beowulf cluster?
Exactly.
Even if OS X can't muster the raw performance of Linux, and even if the hardware isn't quite as cheap as the sort of stuff usually used to build Beowulf clusters, do not underestimate the appeal of an Apple-slick, Rendezvous-enabled clustering solution. People have spent money for convenience since the beginning of time, and I'm sure there are plenty of scientists who'd rather spend their time doing science (or, say, 3D artists who'd rather spend their time doing 3D art) and have the machines just work. The huge appeal of Macs in places like NASA and Lawrence Livermore is a testimony to that fact.
Originally posted by Amorph
The pure-RISC mindset will die hard within IBM.
Which is actually quite strange. POWER and PowerPC aren't "pure-RISC" in the original sense of the term RISC. Their instruction set has hundreds of instructions, some of which are quite powerful. The POWER and PowerPC instruction sets are really just designed to be more orthogonal, based around a fixed instruction word size, and using a load/store architecture. The VMX design uses exactly the same principles, just extending it to the problem of dealing with streaming SIMD data.
I think the hold-outs at IBM are just the ones whose problems don't lend themselves to vector implementation. Or at least they don't think they do. IBM has clearly changed its perspective, however, since a major feature of POWER5 is this mysterious "FastPath" that they keep talking about.