It's faster, the SSOI process improves transistor performance by 35% on the same process according to IBM. IBM is using faster transistors in their CU-08 process that are about 30% faster anyway. The combination with SSOI should give the transistors near a 50% performance boost. All in all the 970FX is about 20-30% faster than the 970 at the same clock speed. IBM's product brochure bears this out. However SpecInt and SpecFP scores were calculated with the GCC 3.3 compiler and not IBM's high-performance iLC compiler. So there is a lot of cloak and dagger secrecy about the 970FX's real performance, probably due to the Apple relationship. Steve Job's wants to announce this himself at a grand media event to get the most media hype.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
IBM Announces New Chip Making Method
- Thread starter MacRumors
- Start date
- Sort by reaction score
It's faster, the SSOI process improves transistor performance by 35% on the same process according to IBM. IBM is using faster transistors in their CU-08 process that are about 30% faster anyway. The combination with SSOI should give the transistors near a 50% performance boost. All in all the 970FX is about 20-30% faster than the 970 at the same clock speed. IBM's product brochure bears this out. However SpecInt and SpecFP scores were calculated with the GCC 3.3 compiler and not IBM's high-performance iLC compiler. So there is a lot of cloak and dagger secrecy about the 970FX's real performance, probably due to the Apple relationship. Steve Job's wants to announce this himself at a grand media event to get the most media hype.
That's a pretty battle front comment as I have heard, and naive. The 970 is a 'superset' of the Power4 core. It contains additional instructions, it contains a VMX unit that the Power4 does not. It uses faster components and contains better pipelining to take it to a much faster frequency while allowing it to operate at a lower voltage. The new 970FX contains PowerTune extensions that do not exist on the Power4 but are promised for the Power5 later this year.
The difference between the two is the dual core and the Power4's 4 way dual core multy module that allow's for 8 core processors in one module. However one dual 970 G5 is more powerful than a single dual core Power4. The difference is that you will never see a 64-Way 970, but the 970 is able to go to 8-way due to its three coherent interconnects (2^3 = 8).
I'll say it--QUAD!
When I see these things, I get a little too excited, I guess!
If the system bus improvements are such that now the CPUs/RAM can't push utilize all of the bus bandwidth, why not add two more CPUs? For non-multiple-CPU applications it would be a total waste, but for those programs which can take advantage of it--yikes. Could be a crusher.
No, I am not naive enough to think that this wouldn't take some tremendous rearchitecture to accomplish--mobo, system controller, etc.--but if the heat per CPU can be handled by the enclosure and enough big-name programs could be rewritten to take advantage of it, why not? I'm not talking about thousands of titles, either. Maybe the top twenty video/animation/graphic design/science/math apps currently running on OS X. The marquee players.
More expensive? Of course. But there are plenty of idiots out there who would buy a "render-farm-in-a-box" like this for $4-6k for the base or $10k maxed out w/RAM without blinking an eye. High-end Mac buyers have deep pockets and love to pay the premium to be on the bleeding edge.
Go ahead. I am bracing myself for the inevitable bubble-bursting.
When I see these things, I get a little too excited, I guess!
If the system bus improvements are such that now the CPUs/RAM can't push utilize all of the bus bandwidth, why not add two more CPUs? For non-multiple-CPU applications it would be a total waste, but for those programs which can take advantage of it--yikes. Could be a crusher.
No, I am not naive enough to think that this wouldn't take some tremendous rearchitecture to accomplish--mobo, system controller, etc.--but if the heat per CPU can be handled by the enclosure and enough big-name programs could be rewritten to take advantage of it, why not? I'm not talking about thousands of titles, either. Maybe the top twenty video/animation/graphic design/science/math apps currently running on OS X. The marquee players.
More expensive? Of course. But there are plenty of idiots out there who would buy a "render-farm-in-a-box" like this for $4-6k for the base or $10k maxed out w/RAM without blinking an eye. High-end Mac buyers have deep pockets and love to pay the premium to be on the bleeding edge.
Go ahead. I am bracing myself for the inevitable bubble-bursting.
Stolid
macrumors regular
I've not looked too much into the details of the difference; I'm mainly going on what Varadarajan said at NASA (which I quoted almost word for word, he made the comment that it was ironic that a chip designed for supercomputer use was modified for a desktop then that desktop chip was put in a supercomputer, as well as a comment the 970 was 'essentially a subset of the Power4'). The additional parts will, of course, be important to any factor. I didn't say "everything will be immediatly usable" but the vast majority of improvements should still be. The VMX unit was admitably ignored by Varadarajan and Tech (most of the things the cluster is to be used for is double-precision, the VMX is single); and he said they wanted to use it if they could but hadn't tried anything yet.
Lower voltage would lower heat (all other things equal); and except in the case of a Power4 optimization on voltage or pipelining (which would either already be in the G5 or have semi decent odds of remaining transferable) I don't see the difference. The fab process is, TTBOMK, the same (or incredibly similar). If the differences in G5 and Power4 are in the Power5 (in favor of G5 similarity) then just replace 'watch the P4' with 'watch the P5'
(back to top of 2nd quote)
Faster components will generate more heat, generally speaking; but as I stated earlier; better cooling and use techniques should still be transferable. I'm talking about better use of things like SoI and whatnot; as IBM is more likely, if you ask me, to put those in their super-computer chips first as those can normally take the 'development cost' hit better.
Certain developments on the Power4/Power5 line would be silly or strange to transfer; such as clock rate increases, and 'multicore' technology. But developments in cooling and power consumption seem - for the most part - to be the same quinessential chip.
Most importantly; I can't see the vector unit inside the G5 making more (and I doubt as much) heat as a 'core' in the Power4s (in terms of watts over area and overall size of chip) -- a Power line of the same generation as the G5 would put out more heat and more/equal heat per square-cm.
I'm not saying the chips are equal; but I'm saying that from a fabrication type (SoI and other IBM techs) and cooling approach (CPU die cooling) the chips are fundamentally very similar. What I'm saying here is admitably somewhat guess-work but I'm going off a combination of the knowledge I have from building my own watercooling systems and the research I did for that and 2 years of Computer Architecture and Design classes. I'm not an expert by any means; but I hope I can at least hazzard a good guess after all that. 😛
~Stolid~
Far more than I imagined
I must say I'm impressed, IBM is delivering far more technology in the 90nm chip than I expected. This is obviously a GOOD THING!!!!!
I do wonder what the executives rational was for declaring that the chip could drive a Laptop. Unless those power usage figures, floating around, are maxiums I still see this chip as being far to hot for a laptop. Maybe it is me but I'd hate to see 30 or 40 watts go out the door to run the north bridge and the CPU at full speed.
Maybe Apple has new battery technology up its sleeve. If not I think some people will be disappointed with any 970 base laptop.
Thanks
Dave
I must say I'm impressed, IBM is delivering far more technology in the 90nm chip than I expected. This is obviously a GOOD THING!!!!!
I do wonder what the executives rational was for declaring that the chip could drive a Laptop. Unless those power usage figures, floating around, are maxiums I still see this chip as being far to hot for a laptop. Maybe it is me but I'd hate to see 30 or 40 watts go out the door to run the north bridge and the CPU at full speed.
Maybe Apple has new battery technology up its sleeve. If not I think some people will be disappointed with any 970 base laptop.
Thanks
Dave
Computer / car analogy is closer then anyone thinks - just as a computer needs software to be useful, as does a car need parts. Walk into any auto parts store and ask for an alternator for a Beamer or an air intake manifold for any straight 8 Benz. Chances are they will have it for a GM or Ford product. Software is not the "gas" in the anaolgy.
I have to stronly disagree here. One of the things that has kept me off the MAC platform for the last couple of years has been performance. If Apple can deliver high perfromance hardware at a reasonable price I'm inclined to start looking again.
I'm a Linux user, I can tell you that there are a huge number of dissatisfied Windows customers out there. Most of these people would not be happy with Linux due to its nature at the moment. Thus Apple has potential customers out there. The trick is to draw them in, good performance for the buck is part of that equation.
The thing to consider is just where this will allow Apple to go performance wise that AMD and Intel will have a hard time following them. You can bet on dual core chips and other high integration technologies working there way into the product matrix. Sure
AMD will eventually have a dual core processor but it will be very easy now for Apple to lead the way on this and other future technologies.
Dave
I'm a Linux user, I can tell you that there are a huge number of dissatisfied Windows customers out there. Most of these people would not be happy with Linux due to its nature at the moment. Thus Apple has potential customers out there. The trick is to draw them in, good performance for the buck is part of that equation.
The thing to consider is just where this will allow Apple to go performance wise that AMD and Intel will have a hard time following them. You can bet on dual core chips and other high integration technologies working there way into the product matrix. Sure
AMD will eventually have a dual core processor but it will be very easy now for Apple to lead the way on this and other future technologies.
Dave
Even though the HT has the interconnects, the 970 does not use the HT interconnect in a fashion that allows it to scale well beyond two processors for most apps. The G5s to date have all been vanilla SMP architectures. HT is capable of NUMA (the Opterons are actually NUMA systems masquerading as SMP and scale almost perfectly out to about 7-8 processors using HT), but the PPC970 systems are not designed in this fashion. Apple would have to re-architect their G5 systems before they would have a quad+ system that was worth the money spent.
As currently engineered, you will get diminishing returns beyond two processors for the PPC970. After that, you'll have serious memory contention issues to deal with. As it is, the PPC970 only gets about 60% of the memory throughput that the Opteron gets in real world tests. (Ironically, the PPC970 gives the best memory performance when using the GCC 2.9x compilers. The IBM XLC compiler falls in the middle and GCC 3.3 gives the worst performance.) This is the standard trade-off between simple SMP and ccNUMA architectures. For dual processors, SMP is cheaper and often faster all other things being equal (which they rarely are), but SMP doesn't scale beyond two processors while NUMA scales very nicely to several.
It does make me wonder what IBMs PPC970 servers will actually look like. I am curious as to whether or not they'll be ccNUMA systems (like Opterons) or SMP (like G5s).
If Apple was really with it, they would develop and package some killer apps designed specifically for teaching/learning that work hand in hand with pedagogy and classroom practices. And bundle these with educational purposes -- It leverages the idea that people buy macs because of the iLife apps. If they could do this for kids and teachers, then they are adding a value to the education market that cannot be purchased at any price on the PC. In anycase -- they must revive HyperCard or something close to it (and bundle it with education purchases). the value of hypercard in education 10-15 years ago could not be understated and it is a shame that it has atrophied.
Just curious Dave, what do you need the performance for? Video editing? Number crunching? Software development?
I'd stay with x86 and Linux if the software suits your needs. Apple isn't really going to give you more unless you value pretty hardware and Apple's software.
Re: I'll say it--QUAD!
The G5 is architected as an SMP system, which means that it won't scale well beyond two processors. Apple would have to do some heavy engineering to make the PPC970 work well as a quad system, which may make it more expensive than it is worth just to say they can. The HyperTransport will scale nicely to quite a few processors if it is used in ccNUMA mode (which is how the Opteron uses it), but 1) the PPC970 currently doesn't know how to do ccNUMA with HyperTransport, and 2) It's memory latency is already so poor in SMP mode that it is arguably not worth the latency hit you would get by converting the architecture to NUMA.
In short, they COULD produce a quad processor PPC970 pretty easily, but it would only give very marginal increases in performance because of the way the system is currently put together and few people would want to spend the money for such meager real-world speed increases. Apple may do NUMA based systems in the future, since the HT works very well for that kind of thing (HT was originally designed for Cray NUMA systems -- tech trivia), but it will probably require a major processor generation upgrade before it would be supported. In other words, you might see it in the PPC980 systems if they re-engineer them to be NUMA friendly, but it will probably never show up in a 970 based chip.
The G5 is architected as an SMP system, which means that it won't scale well beyond two processors. Apple would have to do some heavy engineering to make the PPC970 work well as a quad system, which may make it more expensive than it is worth just to say they can. The HyperTransport will scale nicely to quite a few processors if it is used in ccNUMA mode (which is how the Opteron uses it), but 1) the PPC970 currently doesn't know how to do ccNUMA with HyperTransport, and 2) It's memory latency is already so poor in SMP mode that it is arguably not worth the latency hit you would get by converting the architecture to NUMA.
In short, they COULD produce a quad processor PPC970 pretty easily, but it would only give very marginal increases in performance because of the way the system is currently put together and few people would want to spend the money for such meager real-world speed increases. Apple may do NUMA based systems in the future, since the HT works very well for that kind of thing (HT was originally designed for Cray NUMA systems -- tech trivia), but it will probably require a major processor generation upgrade before it would be supported. In other words, you might see it in the PPC980 systems if they re-engineer them to be NUMA friendly, but it will probably never show up in a 970 based chip.
mpopkin
macrumors 6502
I believe people need to understand something about processors and this applies to PC's as well. Power does not matter. While the 970fx is an amazing chip in all rumors, i want everyone to find a pic of the Xserve G5, you will notice NO size difference between it and the Powermac G5, I own a Dual 2ghz machine and yes the processors are Huge. In addition to finding a picture of the Xserve G5, take a look at a PowerPC G4 or G3 Chip, back when they were made, what seperated Apple from the Wintel folks so much was, was the very fact that Apple processors were 1/3 to 1/4 the size of their Intel/Amd counterparts, the G4/G3 are tiny, they are smaller in dimension almost than memory(thicker though), you can take apart an imac if you want(g3) or look at a powermac g4. When the G5 was created, it was not rushed per say, but it was made using a larger process because they had SO many more transistors and parts that went into producing it and it was powerful. Even at 90nm(current G5 are 120nm) there are still a pretty big processor. It is clear that the 970FX is not a Mobile processor nor intended to be used in a powerbook( maybe a desktop replace pbook, but that would be going downhill and copying the wintel clunkers). One of the reasons the powerbook is so amazing is its size, power and portability. The only way to incorporate the 970fx is to sacrifice all the portability and qualities that we have come to expect and desire. There is no doubt that there will be a powerbook g5, but i doubt that it will be out soon or maybe even based on the powerpc 970(any version) it is possible that Apple is waiting for the PowerPC 980 which is based on IBM Power5 Processor and is much much more powerful than the g5 currently is and is made on a new die process below 65nm and both using/producing less power/heat than the current g5 or PowerPc 970fx could possibly achieve. It is logical to assume that Apple will release a New Powerbook, IMAC(current generation sales have tanked) and Emac(time is ripe for a new one). But i am simply stating: IT WILL NOT be this month or next month or the month after, rather look for it to be actually shipped around August-September, that is not saying that they will not announce it in May-June. They may very well do so.
Stolid
macrumors regular
Nothing stops Apple from making a 'thin' powerbook with all the "classic" features and also a desktop replacement.
A heavy/thick laptop doesn't bug me -- at the moment I'm using an Inspiron 8200 that I love to death and its over 10 pounds. Thin and light might be "sexy" and good advertising but (to metaphore) the army uses tanks as well as James Bond. They both have a place there and they both have a place in a full lineup.
Apple might need to hire a few extra designers (or 'overtime' currents) and the new fabrication of cases and whatnot; but they had to do that for 17" and 12" PBs too -- not a big deal for a large company.
A heavy/thick laptop doesn't bug me -- at the moment I'm using an Inspiron 8200 that I love to death and its over 10 pounds. Thin and light might be "sexy" and good advertising but (to metaphore) the army uses tanks as well as James Bond. They both have a place there and they both have a place in a full lineup.
Apple might need to hire a few extra designers (or 'overtime' currents) and the new fabrication of cases and whatnot; but they had to do that for 17" and 12" PBs too -- not a big deal for a large company.
Remember the IBM PPC970 blades?
Does anybody remember which version of the PPC970 is used in the IBM blades (1U servers) that IBM presented two or three months ago? I only remember that they were clocked up to 1.8Ghz.
EDIT:
Just checked myself, they are dual 1.6Ghz with only two harddrives. I suppose they use the 130nm version, since they were presented already November or Dezember, guess it's two harddrives only because these processors produce more heat.
EDIT EDIT:
The two harddrives are 40GB each, which sounds really small, a Powerbook can have a single 80GB drive. The new XServes look much more attractive compared to them. I wonder what kind of system controller they use, presumably not Apple's.
Does anybody remember which version of the PPC970 is used in the IBM blades (1U servers) that IBM presented two or three months ago? I only remember that they were clocked up to 1.8Ghz.
EDIT:
Just checked myself, they are dual 1.6Ghz with only two harddrives. I suppose they use the 130nm version, since they were presented already November or Dezember, guess it's two harddrives only because these processors produce more heat.
EDIT EDIT:
The two harddrives are 40GB each, which sounds really small, a Powerbook can have a single 80GB drive. The new XServes look much more attractive compared to them. I wonder what kind of system controller they use, presumably not Apple's.
Re: Re: IBM Announces New Chip Making Method
Yes! "Hello Moto" is becoming "Bye Moto". And not many in the Mac community will miss them. Well at least not as long as IBM delivers 🙂
Yes! "Hello Moto" is becoming "Bye Moto". And not many in the Mac community will miss them. Well at least not as long as IBM delivers 🙂
mpopkin
macrumors 6502
What do u know
i have an Dell Inspiron 8200 and compared to my other compute( Powermac G5 D 2ghz, it is a piece of crap, low performance lug of a machine, I would not have it accept it was a graduation gift from my parents, Wintel is not Apple. I switched platforms because i am tired of those cumbersome machines which crash all the time. Do not assume that a lug of a machine like the Inspiron 8200 is any bit the equal of a powerbook, my brother has a powerbook g4 1.25 ghz and it creams the Dell 8200 in performance. Smaller does not mean less powerful, but with Apple you get power and size instead of power and Huge.
i have an Dell Inspiron 8200 and compared to my other compute( Powermac G5 D 2ghz, it is a piece of crap, low performance lug of a machine, I would not have it accept it was a graduation gift from my parents, Wintel is not Apple. I switched platforms because i am tired of those cumbersome machines which crash all the time. Do not assume that a lug of a machine like the Inspiron 8200 is any bit the equal of a powerbook, my brother has a powerbook g4 1.25 ghz and it creams the Dell 8200 in performance. Smaller does not mean less powerful, but with Apple you get power and size instead of power and Huge.
Stolid
macrumors regular
Amazing... You're comparing a DESKTOP to a LAPTOP.
Wintel is not apple - Amazing revelation there too. I thought they were the same company.
Crash all the time? I do not REMEMBER the last crash my 8200 had. I think it had a series of them; but I was doing some weird stuff on it with kernel level stuff (making fake SCSI devices; not fun stuff)
A G4 1.25 creams the 8200? Okay. By what metric?
By simple 'user feel'? If that's the case then I've got to say that I've got to use a G5 system because I've used a top of the line G4 at a CompUSA and felt a graphical lag just doing something as simple as moving windows around in Photoshop on it, I've never seen that on my 8200. Now I've turned off 'show window while moving' so that's probably why; but it goes to show why 'user feel' isn't a good metric. So if you've got another metric please tell me what it is.
Because this is what I keep seeing:
- Rob Galbraith
Smaller does not mean less powerful? Then why does a PowerMac have faster procs (and 2 of them no less) than a PowerBook?
I am not advocating completely ignoring size; but I don't see the harm in a desktop replacement being added to the line. Nothing MAKES you buy that system if Apple makes it; just as nothing MAKES me buy what they've got now. Which I might add is the reason I haven't yet; I'm not at all happy with the state of the mobiles and I can't convince myself I'd use a desktop often enough to justify it, I'm just too mobile.
So by what I've seen of your logic; that means I should stop looking at the Apple platform; its far more important they maintain being "sexy" than have customers. 🙄
Sorry if this sounds flamy; but I see claims that a G4 PowerBook is faster than an 8200 and have to shake my head. It's just not justified (unless of course you get a bottom of the line 8200, but then you'd have to compare it to equivilently lower end "MacTops").
And let's try to steer this more on topic in our replies. 😛