2.5ghz is the top end for the .9nm version of the 970. The version or run we get first will only go to 1.8ghz like mentioned when the 970 was first announced to the public.
[quote]2.5ghz is the top end for the .9nm version of the 970. The version or run we get first will only go to 1.8ghz like mentioned when the 970 was first announced to the public.<hr></blockquote>
Actually, it was confirmed today that the 2.3 GHz and 2.5 Ghz are .13 not .009, but that they won't be ready to go initially. The 1.8 GHz 970 is going to be the 'prime time' top of the line chip.
<strong>Well Linux and AIX don't support Altivec, so I think that's what they mean. They don't want their customers to think it's going to make their software faster just because it has a vector unit. I think that there is support in gcc for Altivec though, so it's possible that Linux on the 970 might have some support for it. But he didn't know anything about that.
</strong><hr></blockquote>
As someone who ports Linux to various PowerPC processors for a living, I can assure you that Linux supports Altivec (at least, for the Motorola 74xx series, if not the IBM). However, for the kernel (which Linux is), this merely means storing the altivec related registers on context switches if necessary.
The GNU toolsuite (gcc, glibc, etc) also support Altivec. What the whole GNU/Linux package doesn't have is specific optimization of the GUI and/or applications for this (since it is expected to run on a variety of processors, this makes sense).
But, altivec enhanced applications could certainly be written and properly handled by the OS (say, an mp3 encoder or graphical tool).
Think about it though. What does a blade server really need vector processing for? I think it's more for the graphically challenged desktop/workstation user.</strong><hr></blockquote>
Can't altivec can be used to speed up encryption and decryption?
Actually, it was confirmed today that the 2.3 GHz and 2.5 Ghz are .13 not .009, but that they won't be ready to go initially. The 1.8 GHz 970 is going to be the 'prime time' top of the line chip.</strong><hr></blockquote>
Really? Damn, must've have missed that. So the move to .009 would get the 970 up there then. Nice, holds well for Apple's future.
Actually, it was confirmed today that the 2.3 GHz and 2.5 Ghz are .13 not .009, but that they won't be ready to go initially. The 1.8 GHz 970 is going to be the 'prime time' top of the line chip.</strong><hr></blockquote>
I did not see this. Do you happen to have a link?
[Edit]: Ach! Nevermind. I see the info at MacRumors.
<strong>Well Linux and AIX don't support Altivec, so I think that's what they mean. They don't want their customers to think it's going to make their software faster just because it has a vector unit. I think that there is support in gcc for Altivec though, so it's possible that Linux on the 970 might have some support for it. But he didn't know anything about that.
As for Apple and Sony, he did NOT suggest that Apple and Sony were in any way connected. I'm not saying it isn't true, but there was no suggestion of Sony and Apple working together on the PS3, 970 or anything else with IBM.
John</strong><hr></blockquote>
Precisely! Altivec is so much wasted silicon for customers whose software makes no use of it. Unless or until the customers have use of it there is no purpose being served by adding expense to the production process and extra spaces to reduce the potential yield.
To clarify, here is what the IBM rep said about Altivec.
1) It is for Apple. He said they needed it because their systems were "graphics intensive." His words, not mine. He told me that they wouldn't be using it and that they make a point of informing their customers of that. He also suggested that other people have caught on that it could be for Apple when they see the vector unit on the slide.
2) During the presentation he circled the little box that said vector unit and specifically mentioned that they weren't utilizing it. That could mean anything, but I took it to be a marketing message to their customers not to expect anything from it. Who knows what reality is. I know that IBM certainly mentions it in their press releases, so it doesn't all jive. But that's what the guy said both to the group and to me personally.
I think we're splitting hairs though. The key is that IBM without any doubt confirmed to me that Apple is going to use the 970. None of us doubted this, but it sure is nice to know for sure. This guy is almost certainly NOT a Macintosh user, so the fact that he was so obviously aware of Apple and the 970 really hit home to me.
He seemed impressed that I knew about the Apple connection and the chip details in general. There was no NDA, btw. I didn't get the impression that they were discussing anything that wasn't available on their website. The only thing that might have been somewhat secretive was the PS3 talk. But much like the 970 chip he discussed that with me privately and also to the whole group. So it couldn't have been that big a deal. There were several other reps their too. So either they're clueless or the info isn't that secret.
<strong>1) It is for Apple. He said they needed it because their systems were "graphics intensive." His words, not mine. He told me that they wouldn't be using it and that they make a point of informing their customers of that. He also suggested that other people have caught on that it could be for Apple when they see the vector unit on the slide.
2) During the presentation he circled the little box that said vector unit and specifically mentioned that they weren't utilizing it. That could mean anything, but I took it to be a marketing message to their customers not to expect anything from it. </strong><hr></blockquote>
I can say with absolute certainty if he said that he doesn't really know (or he has been misinterpreted). IBM definitely isn't planning to withhold the Altivec unit from customers.
[quote]Originally posted by johnpg:
<strong>So either they're clueless or the info isn't that secret.</strong><hr></blockquote>
Everything there is pretty much freely available or was due to be available at the end of the CeBit conference except the stuff about Apple.
Most likely IBM just won't be supporting the altivec through their marketing plans. That doesn't mean it won't be usable, just that it won't be supported by IBM.
It will be there for Apple and anybody else who can figure out how to benefit from it. No doubt, there are a lot of folks who will figure out how to use it, and in the end that will likely benefit development for Apple as well.
As to the 2.5ghz chip, my understanding from the beginning was that they are getting a surprisingly larger number of 2.5 ghz yields then expected out of the .13 process, but probably not enough for initial introduction and release. Therefore they will be introduced later, possibly as an incremental increase prior to the eventual 980 on the .09 process.
What I am expecting is that sometime not too long after MacWorld (not at Macworld) the 970 Apple boxes will be introduced with a single 1.4, dual 1.4 and dual 1.8 at very reasonable pricepoints. The 1.4 970's will be noticiably faster then the 1.42 motorola G4 with true high speed bus and a lot more. Even the low end single 1.4 will be quite nice, the rough equivelent of a 2 ghz G4.
The 1.2 ghz will be reseved for portables, with its low heat and miserly power use, but will outperform the current 1 gig Powerbooks by a comfortable margin.
Then again, that is purely my speculation, and I have been wrong before.
The new limiting factor will be the speed of "affordable" memory.
Aw, man, this is no fun. We finally know about the next generation processor for sure, and it's not due out for half a year. BAH! Remember all those good times the G5 (the MOT 8500) gave us? There were at least three excitedly over-hyped expos before the idea started dying. This is all *too* certain.
Ah, I miss the good old days. <img src="graemlins/lol.gif" border="0" alt="[Laughing]" />
A full PowerPC Blade-rack armed with 970 with Altivec, run on Linux with applications like BLAST most certainly will make use of Altivec even if IBM them selves doesn't.
Sun are releasing special blade-modules just for SSL-encryption som I guess that IBM could make something similar and take advantage of Altivec's strength when it comes to encryption.
The 1.2 ghz will be reseved for portables, with its low heat and miserly power use, but will outperform the current 1 gig Powerbooks by a comfortable margin.
Then again, that is purely my speculation, and I have been wrong before.
The new limiting factor will be the speed of "affordable" memory.
</strong><hr></blockquote>
It's even possible we see another G4 speed bump in the powerbooks to a 1.25Ghz G4 before we get a 1.2 Ghz 970 in them. that's still a much larger step than the transition from G3/400Mhz to a G4/400Mhz
i also think that the G4 7457 still has a future in apple's. albeit i would love to see ppc 970's across the line. i think that will take another year/year and a half.
besides that i think the limiting factor won't be the speed of affordable memory, but will be apples roadmap. what do they plan to release after the release of the powermac with 970.
The problem with the G3 G4 transition was several fold, with the main one that the G4 did not scale higher then the G3 for a very long time and that the OS did not support SMP.
Who would buy a G4/500 tower if a iMac G3/450 was sold for a fraction of the price? --Very few!
Now having dual in towers and servers and single in ther other line is a viable option.
Also, replacing a 1GHz G3 with a 1GHz G4 does little to the performance outside strict Altivec performance. Replacing a 1GHz G4/G3 with a 1GHz 970 will at least double the performance.
A iMac is better of with 1 GHz 970 than 1.6 GHz G4 if they can be inplemented at similar total costs. Not only will the computer be faster but Apple would by using 970 invest in a CPU that has a future as a desktop CPU. Buying G4 is investing in a dead end desktop CPU, the fact that is very good in embedded applications is a good thing for CISCO but not for Apple.
I can not imagine that Apple could replace all G4 and G3 computers overnight with 970 but I do not see any problem with doing it if it would have been possible, if you understand what I mean.
ASAP:First the towers, shortly there after the servers. The other ones can wait for their natrual replacement cycles to incorporate the 970.
I would guess that the G4 hold out the longest in the iBook. With 970 in both PB and iB the two lines will merge totaly or partly.
So no more G4 speed bumps in the towers ever and no speed bumps next year at all to the G4. So: 2003 in with the 970 in towers and servers and plain vanilla speed bumps for the rest, 2004 out with the G4. <img src="graemlins/lol.gif" border="0" alt="[Laughing]" />
Comments
<strong>That's pretty interesting.
I've heard that 2.5Ghz was the top, but 1.7 seems reasonable for actual retail production. 2.5 is the number for future implementation I'm asserting.
[ 03-12-2003: Message edited by: willywalloo ]</strong><hr></blockquote>
2.5ghz is the top end for the .9nm version of the 970. The version or run we get first will only go to 1.8ghz like mentioned when the 970 was first announced to the public.
<strong>
This came out about a <a href="http://maccentral.macworld.com/news/0203/11.redhat.php" target="_blank">year ago</a>. Note that Red Hat has <a href="http://sources.redhat.com/binutils/docs-2.12/as.info/PowerPC-Opts.html" target="_blank">supported Altivec</a> for quite some time.</strong><hr></blockquote>
Thanks Rhumgod. I'm not too informed about Linux . . . Guess I will be more informed soon though.
Thanks Again
Actually, it was confirmed today that the 2.3 GHz and 2.5 Ghz are .13 not .009, but that they won't be ready to go initially. The 1.8 GHz 970 is going to be the 'prime time' top of the line chip.
<strong>Well Linux and AIX don't support Altivec, so I think that's what they mean. They don't want their customers to think it's going to make their software faster just because it has a vector unit. I think that there is support in gcc for Altivec though, so it's possible that Linux on the 970 might have some support for it. But he didn't know anything about that.
</strong><hr></blockquote>
As someone who ports Linux to various PowerPC processors for a living, I can assure you that Linux supports Altivec (at least, for the Motorola 74xx series, if not the IBM). However, for the kernel (which Linux is), this merely means storing the altivec related registers on context switches if necessary.
The GNU toolsuite (gcc, glibc, etc) also support Altivec. What the whole GNU/Linux package doesn't have is specific optimization of the GUI and/or applications for this (since it is expected to run on a variety of processors, this makes sense).
But, altivec enhanced applications could certainly be written and properly handled by the OS (say, an mp3 encoder or graphical tool).
John
<strong>
Think about it though. What does a blade server really need vector processing for? I think it's more for the graphically challenged desktop/workstation user.</strong><hr></blockquote>
Can't altivec can be used to speed up encryption and decryption?
<strong>Can't altivec can be used to speed up encryption and decryption?</strong><hr></blockquote>
I am not sure if there is sufficient data intensity in encryption/decryption that would benefit from alitvec processing.
<strong>
Actually, it was confirmed today that the 2.3 GHz and 2.5 Ghz are .13 not .009, but that they won't be ready to go initially. The 1.8 GHz 970 is going to be the 'prime time' top of the line chip.</strong><hr></blockquote>
Really? Damn, must've have missed that. So the move to .009 would get the 970 up there then. Nice, holds well for Apple's future.
<strong>
Actually, it was confirmed today that the 2.3 GHz and 2.5 Ghz are .13 not .009, but that they won't be ready to go initially. The 1.8 GHz 970 is going to be the 'prime time' top of the line chip.</strong><hr></blockquote>
I did not see this. Do you happen to have a link?
[Edit]: Ach! Nevermind. I see the info at MacRumors.
[ 03-12-2003: Message edited by: LudwigVan ]</p>
There'll be 2.5 GHz 970s running around one way or another :cool:
<strong>I don't think they will need to.</strong><hr></blockquote>
They will.....2 is 2, 1.8 is 1.8.....people want NUMBER!
Just snap a 20 lb copper heatsink on top of the chip it should be fine, I think
[ 03-13-2003: Message edited by: Leonis ]</p>
<strong>Well Linux and AIX don't support Altivec, so I think that's what they mean. They don't want their customers to think it's going to make their software faster just because it has a vector unit. I think that there is support in gcc for Altivec though, so it's possible that Linux on the 970 might have some support for it. But he didn't know anything about that.
As for Apple and Sony, he did NOT suggest that Apple and Sony were in any way connected. I'm not saying it isn't true, but there was no suggestion of Sony and Apple working together on the PS3, 970 or anything else with IBM.
John</strong><hr></blockquote>
Precisely! Altivec is so much wasted silicon for customers whose software makes no use of it. Unless or until the customers have use of it there is no purpose being served by adding expense to the production process and extra spaces to reduce the potential yield.
1) It is for Apple. He said they needed it because their systems were "graphics intensive." His words, not mine. He told me that they wouldn't be using it and that they make a point of informing their customers of that. He also suggested that other people have caught on that it could be for Apple when they see the vector unit on the slide.
2) During the presentation he circled the little box that said vector unit and specifically mentioned that they weren't utilizing it. That could mean anything, but I took it to be a marketing message to their customers not to expect anything from it. Who knows what reality is. I know that IBM certainly mentions it in their press releases, so it doesn't all jive. But that's what the guy said both to the group and to me personally.
I think we're splitting hairs though. The key is that IBM without any doubt confirmed to me that Apple is going to use the 970. None of us doubted this, but it sure is nice to know for sure. This guy is almost certainly NOT a Macintosh user, so the fact that he was so obviously aware of Apple and the 970 really hit home to me.
He seemed impressed that I knew about the Apple connection and the chip details in general. There was no NDA, btw. I didn't get the impression that they were discussing anything that wasn't available on their website. The only thing that might have been somewhat secretive was the PS3 talk. But much like the 970 chip he discussed that with me privately and also to the whole group. So it couldn't have been that big a deal. There were several other reps their too. So either they're clueless or the info isn't that secret.
Cheers,
John
<strong>1) It is for Apple. He said they needed it because their systems were "graphics intensive." His words, not mine. He told me that they wouldn't be using it and that they make a point of informing their customers of that. He also suggested that other people have caught on that it could be for Apple when they see the vector unit on the slide.
2) During the presentation he circled the little box that said vector unit and specifically mentioned that they weren't utilizing it. That could mean anything, but I took it to be a marketing message to their customers not to expect anything from it. </strong><hr></blockquote>
I can say with absolute certainty if he said that he doesn't really know (or he has been misinterpreted). IBM definitely isn't planning to withhold the Altivec unit from customers.
[quote]Originally posted by johnpg:
<strong>So either they're clueless or the info isn't that secret.</strong><hr></blockquote>
Everything there is pretty much freely available or was due to be available at the end of the CeBit conference except the stuff about Apple.
It will be there for Apple and anybody else who can figure out how to benefit from it. No doubt, there are a lot of folks who will figure out how to use it, and in the end that will likely benefit development for Apple as well.
As to the 2.5ghz chip, my understanding from the beginning was that they are getting a surprisingly larger number of 2.5 ghz yields then expected out of the .13 process, but probably not enough for initial introduction and release. Therefore they will be introduced later, possibly as an incremental increase prior to the eventual 980 on the .09 process.
What I am expecting is that sometime not too long after MacWorld (not at Macworld) the 970 Apple boxes will be introduced with a single 1.4, dual 1.4 and dual 1.8 at very reasonable pricepoints. The 1.4 970's will be noticiably faster then the 1.42 motorola G4 with true high speed bus and a lot more. Even the low end single 1.4 will be quite nice, the rough equivelent of a 2 ghz G4.
The 1.2 ghz will be reseved for portables, with its low heat and miserly power use, but will outperform the current 1 gig Powerbooks by a comfortable margin.
Then again, that is purely my speculation, and I have been wrong before.
The new limiting factor will be the speed of "affordable" memory.
[ 03-13-2003: Message edited by: Shaktai ]</p>
Ah, I miss the good old days.
Sun are releasing special blade-modules just for SSL-encryption som I guess that IBM could make something similar and take advantage of Altivec's strength when it comes to encryption.
<strong>...
The 1.2 ghz will be reseved for portables, with its low heat and miserly power use, but will outperform the current 1 gig Powerbooks by a comfortable margin.
Then again, that is purely my speculation, and I have been wrong before.
The new limiting factor will be the speed of "affordable" memory.
</strong><hr></blockquote>
It's even possible we see another G4 speed bump in the powerbooks to a 1.25Ghz G4 before we get a 1.2 Ghz 970 in them. that's still a much larger step than the transition from G3/400Mhz to a G4/400Mhz
i also think that the G4 7457 still has a future in apple's. albeit i would love to see ppc 970's across the line. i think that will take another year/year and a half.
besides that i think the limiting factor won't be the speed of affordable memory, but will be apples roadmap. what do they plan to release after the release of the powermac with 970.
[ 03-13-2003: Message edited by: gar ]</p>
Who would buy a G4/500 tower if a iMac G3/450 was sold for a fraction of the price? --Very few!
Now having dual in towers and servers and single in ther other line is a viable option.
Also, replacing a 1GHz G3 with a 1GHz G4 does little to the performance outside strict Altivec performance. Replacing a 1GHz G4/G3 with a 1GHz 970 will at least double the performance.
A iMac is better of with 1 GHz 970 than 1.6 GHz G4 if they can be inplemented at similar total costs. Not only will the computer be faster but Apple would by using 970 invest in a CPU that has a future as a desktop CPU. Buying G4 is investing in a dead end desktop CPU, the fact that is very good in embedded applications is a good thing for CISCO but not for Apple.
I can not imagine that Apple could replace all G4 and G3 computers overnight with 970 but I do not see any problem with doing it if it would have been possible, if you understand what I mean.
ASAP:First the towers, shortly there after the servers. The other ones can wait for their natrual replacement cycles to incorporate the 970.
I would guess that the G4 hold out the longest in the iBook. With 970 in both PB and iB the two lines will merge totaly or partly.
So no more G4 speed bumps in the towers ever and no speed bumps next year at all to the G4. So: 2003 in with the 970 in towers and servers and plain vanilla speed bumps for the rest, 2004 out with the G4. <img src="graemlins/lol.gif" border="0" alt="[Laughing]" />