<strong>What Apple needs to do is to *not* show incremental improvements now (they work better in a growing market when people will buy simply because they have budget), but rather to save them up and launch them all together when they can demonstrate an overall superior product.</strong><hr></blockquote>
Yeah, and most likely risk a few customers who need a machine now and say "hey, if apple isn't introducing nothing new now, then they wont later as well". That "save up performance for later" doesn't really work in computer markets.
[quote]Originally posted by johnsonwax:
<strong>Consider the creeping CPU speeds at Intel and AMD. How many people can articulate the benefits of 2.2 GHz over 1.6GHz? or 533MHz FSB vs. 333MHz? Oh, it's faster, sure, but faster enough to argue for funds to buy? How much faster? How much will we save with faster hardware? Those are the kinds of questions that come out when the money isn't there.</strong><hr></blockquote>
Most of those discussion I had the joy of listening too ended with "faster enough to save even more time". That's why former SGI shops are switching to PCs for 3D work.
[quote]Originally posted by johnsonwax:
<strong>A is 33% faster than B which is 33% faster than C which is 33% faster than D which is 33% faster than E. It doesn't sell as well in this market as if you said A is 400% faster than E. That requires that B-D never exist.
Keep in mind that Apple cannot grow market share by appeasing us. It does need to do that, but their user base is loyal and willing to be abused. They grow by converting, and people don't convert for marginal reasons.</strong><hr></blockquote>
But they are more likely to convert for marginal reasons than for no reasons at all. If someone considers converting _today_ and there were no 33% increases for the last 6 months he wont convert but buy another PC/Sun/SGI which means that he's done for 2 years. No converting here. And even if Apple has a 400% jump in 2 months it risks of beine even in two years time when it's upgrade time again.
[quote]Originally posted by johnsonwax:
<strong>From my view, Apple shouldn't sweat the incremental improvements. They should focus on demonstrating a substantial benefit to those outside of the current user base.</strong><hr></blockquote>
I am not saying you are completely wrong, just that incremental upgrades are nice too.
- I heard about further instructions to be added to the PPC instruction set some time ago, those instructions were neither BookE compliant nor part of IBM Fast Path. Maybe they're what we call Apple PI here - as far as I remember those instructions were ment to precode information for the pipeline, similar to some VLIW designs. This could be it if "PI" means "pipeline instructions".
- I already told you that I have some Motorola G5 inside info, and I even posted some of it. I'm really surprised that Apple is said to have "killed" the Motorola G5, as I find the specs pretty amazing. As far as I can say the Motorola desktop G5 should be 50 to 100% faster than the G4, and it should have shipped in 2002 or in early 2003. If Apple has killed the Motorola G5 there must have been good reasons to do so, and there must be some kind of replacement. I don't think Apple will be able to stay with the G4 until 2004, when the Power5 will ship.
- Just for your information, although maybe not relevant: "Trinity" was in fact the Cube codename, while "Trident" was the PowerPC 620, basically a 64 bit version of the PPC604.
- I'm trying to get some inside info about future IBM PPC designs, I'll be back as soon as I know more...
- I heard about further instructions to be added to the PPC instruction set some time ago, those instructions were neither BookE compliant nor part of IBM Fast Path. Maybe they're what we call Apple PI here - as far as I remember those instructions were ment to precode information for the pipeline, similar to some VLIW designs. This could be it if "PI" means "pipeline instructions".
- I already told you that I have some Motorola G5 inside info, and I even posted some of it. I'm really surprised that Apple is said to have "killed" the Motorola G5, as I find the specs pretty amazing. As far as I can say the Motorola desktop G5 should be 50 to 100% faster than the G4, and it should have shipped in 2002 or in early 2003. If Apple has killed the Motorola G5 there must have been good reasons to do so, and there must be some kind of replacement. I don't think Apple will be able to stay with the G4 until 2004, when the Power5 will ship.
- Just for your information, although maybe not relevant: "Trinity" was in fact the Cube codename, while "Trident" was the PowerPC 620, basically a 64 bit version of the PPC604.
- I'm trying to get some inside info about future IBM PPC designs, I'll be back as soon as I know more...</strong><hr></blockquote>
Here's your inside info on IBM PPC designs:
IBM has developed Power4. For the workstation.
On the low-end is a 1 chip, dual core design. High-end is a 1 chip, quad core... Shipping 4th quarter, 2002.
No word on any relation to these proc's and Apple. However, if this chip begins shipping in Q4, that would mean Sept., which would work out well if new Power Macs utilizing this chip were intro'd during MWNY... :cool:
On the low-end is a 1 chip, dual core design. High-end is a 1 chip, quad core... Shipping 4th quarter, 2002.
No word on any relation to these proc's and Apple. However, if this chip begins shipping in Q4, that would mean Sept., which would work out well if new Power Macs utilizing this chip were intro'd during MWNY... :cool: </strong><hr></blockquote>
Uh, that's not inside info -- that's public knowledge because the Q4 in question was last year. The POWER4 has been out since Oct 2001.
I don't think Motorola would balk at adding AltiVec, and Apple already has its own name for it (VelocityEngine) so they wouldn't call it Pi. Pipeline instructions is a good name, but it doesn't tell me what it means.
Uh, that's not inside info -- that's public knowledge because the Q4 in question was last year. The POWER4 has been out since Oct 2001.
I don't think Motorola would balk at adding AltiVec, and Apple already has its own name for it (VelocityEngine) so they wouldn't call it Pi. Pipeline instructions is a good name, but it doesn't tell me what it means.</strong><hr></blockquote>
For clarification purposes, what's been out since last year is the SERVER version. This is a WORKSTATION version - a derivative of the server chip... (this isn't out yet ;o)
<strong>I think you guys are too quick to accept this as the meaning of Trinity. I think it's clear. It refers to AIM.</strong><hr></blockquote>
This looks like fun.
Pi fits nicely with Pipeline Instructions.
Trinity makes the most sense as a reference to AIM, which means IBM is back in the game.
Dolphin is a tough one -- that IC company just doesn't look like something relevant, or at least their publicly announced products don't. Dolphin was the code name for the whole GameCube, not the processor in it.
IBM's G3 is just not a suitable replacement for Motorola's G4, even with SIMD added -- MPX bus might only be 166 MHz, but it is more efficient than the G3's 60x bus; the G4's FPU is faster; the G4 7455 has more functional units and longer pipes, allowing higher clock rates; the G4 has wider internal busses and better caches; etc. IBM would have to push the G3 a long way to catch up. More likely would be either buying the G4 from Motorola (probably not), or starting from a new design (not in the near term).
The GameCube's Gekko chip doesn't have "awesome memory bandwidth" as somebody said above... it is just a slightly more efficient G3, limited to the 60x bus @ 100 MHz (IIRC). The system has really good memory bandwidth, and most of it is used by the graphics chip without disturbing the processor's access to memory. Rather like the Xserve in that sense. There are a couple of processor extensions, but AltiVec provides equivalent or better capabilities already.
So I think "Dolphin" is probably a reference to a large cube-like Mac design. We know Steve likes his cubes.
For clarification purposes, what's been out since last year is the SERVER version. This is a WORKSTATION version - a derivative of the server chip... (this isn't out yet ;o)
</strong><hr></blockquote>
Aha, thank you. Is there any public reference to this at all, or is it really insider info?
If, by "after eating pie" references to "Pi" the number it to means that a nice improvement in floating point arithmetic will come around the corner. Which is good for workstation work.
Aha, thank you. Is there any public reference to this at all, or is it really insider info?</strong><hr></blockquote>
Unfortunately, no direct public reference. I posed a question to my acquantance in IBM, and was told that the answer to the specific question I asked was "IBM Confidential". Then, it was followed by, "but what I can tell you..." and that's what I posted here.
L3 cache is unlikely with high speed RAM, as it offer less of a performance advantage. It may just vbe tiding the G4+ over until a faster FSB along.
[quote]If I recall correctly "Pi" meant Pipeline Instructions. Apple PI was some sort of custom pipeline instructions that Apple wanted in the G5 chip<hr></blockquote>
Aren't pipelines (currently) invisible to code executing on them? Any examples of what a pipeline instruction would do with the pipeline?
[quote]Trinity makes the most sense as a reference to AIM, which means IBM is back in the game.<hr></blockquote>
At first I thought AOL == AOL Instant Messenger <img src="graemlins/embarrassed.gif" border="0" alt="[Embarrassed]" />
[quote]The GameCube's Gekko chip doesn't have "awesome memory bandwidth" as somebody said above... it is just a slightly more efficient G3, limited to the 60x bus @ 100 MHz (IIRC). <hr></blockquote>
From Nintendo's pages:
External Bus 1.3GB/second peak bandwidth (32-bit address space, 64-bit data bus 162 MHz clock)
The GameCube's Gekko chip doesn't have "awesome memory bandwidth" as somebody said above... it is just a slightly more efficient G3, limited to the 60x bus @ 100 MHz (IIRC.</strong><hr></blockquote>
Stoo is right, it runs at 162 Mhz. What's more interesting though is that IBM says it supports some sort of data compression, like the recent graphics chips, that results in an "effective" 5.2 GB/s.
I'm wondering if anyone can help out here with links. I can't seem to find much information on the Power4 on IBM's site. An index posted on has a link to PPC - Power is MIA. And when I use their search engine, the best I can find are press releases praising the Power4.
I need more information! Additionally, can anyone else confirm that Power5 and Power6 chips are being developed?
<strong>I'm wondering if anyone can help out here with links. I can't seem to find much information on the Power4 on IBM's site. An index posted on has a link to PPC - Power is MIA. And when I use their search engine, the best I can find are press releases praising the Power4.
<strong>Stoo is right, it runs at 162 Mhz. What's more interesting though is that IBM says it supports some sort of data compression, like the recent graphics chips, that results in an "effective" 5.2 GB/s.</strong><hr></blockquote>
Hmmm... you might be right about the clock rate. It is a modified 60x bus protocol, however, and doesn't sustain that peak throughput for very long (i.e. one cacheline burst at a time). The supposed "data compression" is simply fixed <-> float conversions in hardware together with a write combining write-thru pipe that was hacked onto the chip and has numerous issues, plus it must be explicitly coded for by the software developers. Very useful, to be sure, but it wouldn't be suited as-is in a Mac... besides which AltiVec can already do the fixed/float conversions better, and has cache streaming instructions and a higher clock rate.
The Apple Pi instructions could certainly be yet more improvements to the AltiVec cache controls -- that would be cool, but FSB bus speeds would need to increase to really make the most of it.
Oh -- one more thing about the GameCube... what I meant originally was that its memory is capable of more bandwidth than the processor can use. This is why the graphics engine running at full speed doesn't get in the way of the Gekkco. Hence the similar nature to the Xserve.
I seem to recall that Gastermating (the arch-rival of Mastergating) predicted the "Pi-Mac" months ago. MWSF 02 I think. So hey, maybe he was onto something. Maybe the next generation of Power Macs really will run at 3.14 GHz, boosted by the new Pipeline Instructions!
No, the credit goes to all the others here with real technical knowledge. This is a very interesting thread because of them, and I'm really diggin' the info/speculation about IBM and the G5.
[quote]Trinity is the code name for the Cube. <hr></blockquote>
I thought that Trinity was the name of the lead female character in "The Matrix." So, hopefully we'll be getting an Altivec-enhanced Matrix screensaver from Apple at MWNY '02.
Comments
<strong>What Apple needs to do is to *not* show incremental improvements now (they work better in a growing market when people will buy simply because they have budget), but rather to save them up and launch them all together when they can demonstrate an overall superior product.</strong><hr></blockquote>
Yeah, and most likely risk a few customers who need a machine now and say "hey, if apple isn't introducing nothing new now, then they wont later as well". That "save up performance for later" doesn't really work in computer markets.
[quote]Originally posted by johnsonwax:
<strong>Consider the creeping CPU speeds at Intel and AMD. How many people can articulate the benefits of 2.2 GHz over 1.6GHz? or 533MHz FSB vs. 333MHz? Oh, it's faster, sure, but faster enough to argue for funds to buy? How much faster? How much will we save with faster hardware? Those are the kinds of questions that come out when the money isn't there.</strong><hr></blockquote>
Most of those discussion I had the joy of listening too ended with "faster enough to save even more time". That's why former SGI shops are switching to PCs for 3D work.
[quote]Originally posted by johnsonwax:
<strong>A is 33% faster than B which is 33% faster than C which is 33% faster than D which is 33% faster than E. It doesn't sell as well in this market as if you said A is 400% faster than E. That requires that B-D never exist.
Keep in mind that Apple cannot grow market share by appeasing us. It does need to do that, but their user base is loyal and willing to be abused. They grow by converting, and people don't convert for marginal reasons.</strong><hr></blockquote>
But they are more likely to convert for marginal reasons than for no reasons at all. If someone considers converting _today_ and there were no 33% increases for the last 6 months he wont convert but buy another PC/Sun/SGI which means that he's done for 2 years. No converting here. And even if Apple has a 400% jump in 2 months it risks of beine even in two years time when it's upgrade time again.
[quote]Originally posted by johnsonwax:
<strong>From my view, Apple shouldn't sweat the incremental improvements. They should focus on demonstrating a substantial benefit to those outside of the current user base.</strong><hr></blockquote>
I am not saying you are completely wrong, just that incremental upgrades are nice too.
I think you guys are too quick to accept this as the meaning of Trinity. I think it's clear. It refers to AIM.
- I heard about further instructions to be added to the PPC instruction set some time ago, those instructions were neither BookE compliant nor part of IBM Fast Path. Maybe they're what we call Apple PI here - as far as I remember those instructions were ment to precode information for the pipeline, similar to some VLIW designs. This could be it if "PI" means "pipeline instructions".
- I already told you that I have some Motorola G5 inside info, and I even posted some of it. I'm really surprised that Apple is said to have "killed" the Motorola G5, as I find the specs pretty amazing. As far as I can say the Motorola desktop G5 should be 50 to 100% faster than the G4, and it should have shipped in 2002 or in early 2003. If Apple has killed the Motorola G5 there must have been good reasons to do so, and there must be some kind of replacement. I don't think Apple will be able to stay with the G4 until 2004, when the Power5 will ship.
- Just for your information, although maybe not relevant: "Trinity" was in fact the Cube codename, while "Trident" was the PowerPC 620, basically a 64 bit version of the PPC604.
- I'm trying to get some inside info about future IBM PPC designs, I'll be back as soon as I know more...
<strong>Just a few comments:
- I heard about further instructions to be added to the PPC instruction set some time ago, those instructions were neither BookE compliant nor part of IBM Fast Path. Maybe they're what we call Apple PI here - as far as I remember those instructions were ment to precode information for the pipeline, similar to some VLIW designs. This could be it if "PI" means "pipeline instructions".
- I already told you that I have some Motorola G5 inside info, and I even posted some of it. I'm really surprised that Apple is said to have "killed" the Motorola G5, as I find the specs pretty amazing. As far as I can say the Motorola desktop G5 should be 50 to 100% faster than the G4, and it should have shipped in 2002 or in early 2003. If Apple has killed the Motorola G5 there must have been good reasons to do so, and there must be some kind of replacement. I don't think Apple will be able to stay with the G4 until 2004, when the Power5 will ship.
- Just for your information, although maybe not relevant: "Trinity" was in fact the Cube codename, while "Trident" was the PowerPC 620, basically a 64 bit version of the PPC604.
- I'm trying to get some inside info about future IBM PPC designs, I'll be back as soon as I know more...</strong><hr></blockquote>
Here's your inside info on IBM PPC designs:
IBM has developed Power4. For the workstation.
On the low-end is a 1 chip, dual core design. High-end is a 1 chip, quad core... Shipping 4th quarter, 2002.
No word on any relation to these proc's and Apple. However, if this chip begins shipping in Q4, that would mean Sept., which would work out well if new Power Macs utilizing this chip were intro'd during MWNY... :cool:
<strong>
Here's your inside info on IBM PPC designs:
IBM has developed Power4. For the workstation.
On the low-end is a 1 chip, dual core design. High-end is a 1 chip, quad core... Shipping 4th quarter, 2002.
No word on any relation to these proc's and Apple. However, if this chip begins shipping in Q4, that would mean Sept., which would work out well if new Power Macs utilizing this chip were intro'd during MWNY... :cool: </strong><hr></blockquote>
Uh, that's not inside info -- that's public knowledge because the Q4 in question was last year. The POWER4 has been out since Oct 2001.
I don't think Motorola would balk at adding AltiVec, and Apple already has its own name for it (VelocityEngine) so they wouldn't call it Pi. Pipeline instructions is a good name, but it doesn't tell me what it means.
<strong>
Uh, that's not inside info -- that's public knowledge because the Q4 in question was last year. The POWER4 has been out since Oct 2001.
I don't think Motorola would balk at adding AltiVec, and Apple already has its own name for it (VelocityEngine) so they wouldn't call it Pi. Pipeline instructions is a good name, but it doesn't tell me what it means.</strong><hr></blockquote>
For clarification purposes, what's been out since last year is the SERVER version. This is a WORKSTATION version - a derivative of the server chip... (this isn't out yet ;o)
[ 06-12-2002: Message edited by: justafriend ]</p>
<strong>I think you guys are too quick to accept this as the meaning of Trinity. I think it's clear. It refers to AIM.</strong><hr></blockquote>
This looks like fun.
Pi fits nicely with Pipeline Instructions.
Trinity makes the most sense as a reference to AIM, which means IBM is back in the game.
Dolphin is a tough one -- that IC company just doesn't look like something relevant, or at least their publicly announced products don't. Dolphin was the code name for the whole GameCube, not the processor in it.
IBM's G3 is just not a suitable replacement for Motorola's G4, even with SIMD added -- MPX bus might only be 166 MHz, but it is more efficient than the G3's 60x bus; the G4's FPU is faster; the G4 7455 has more functional units and longer pipes, allowing higher clock rates; the G4 has wider internal busses and better caches; etc. IBM would have to push the G3 a long way to catch up. More likely would be either buying the G4 from Motorola (probably not), or starting from a new design (not in the near term).
The GameCube's Gekko chip doesn't have "awesome memory bandwidth" as somebody said above... it is just a slightly more efficient G3, limited to the 60x bus @ 100 MHz (IIRC). The system has really good memory bandwidth, and most of it is used by the graphics chip without disturbing the processor's access to memory. Rather like the Xserve in that sense. There are a couple of processor extensions, but AltiVec provides equivalent or better capabilities already.
So I think "Dolphin" is probably a reference to a large cube-like Mac design. We know Steve likes his cubes.
<strong>
For clarification purposes, what's been out since last year is the SERVER version. This is a WORKSTATION version - a derivative of the server chip... (this isn't out yet ;o)
</strong><hr></blockquote>
Aha, thank you. Is there any public reference to this at all, or is it really insider info?
<strong>My interpretation: the G3 gets Altivec</strong><hr></blockquote>
Isn't that a G4?
<strong>
Aha, thank you. Is there any public reference to this at all, or is it really insider info?</strong><hr></blockquote>
Unfortunately, no direct public reference. I posed a question to my acquantance in IBM, and was told that the answer to the specific question I asked was "IBM Confidential". Then, it was followed by, "but what I can tell you..." and that's what I posted here.
As an aside, Jamee Abdulhafiz is listed on this page ( <a href="http://www.research.ibm.com/journal/rd/461/luddeaut.html"; target="_blank">http://www.research.ibm.com/journal/rd/461/luddeaut.html</a>; ) as Senior Engineer and verification leader for a POWER4 derivative microprocessor...
So, there is some proof to the statement that Power4 derivatives are planned... Just no detail like what I was provided.
[ 06-12-2002: Message edited by: justafriend ]</p>
L3 cache is unlikely with high speed RAM, as it offer less of a performance advantage. It may just vbe tiding the G4+ over until a faster FSB along.
[quote]If I recall correctly "Pi" meant Pipeline Instructions. Apple PI was some sort of custom pipeline instructions that Apple wanted in the G5 chip<hr></blockquote>
Aren't pipelines (currently) invisible to code executing on them? Any examples of what a pipeline instruction would do with the pipeline?
[quote]Trinity makes the most sense as a reference to AIM, which means IBM is back in the game.<hr></blockquote>
At first I thought AOL == AOL Instant Messenger <img src="graemlins/embarrassed.gif" border="0" alt="[Embarrassed]" />
[quote]The GameCube's Gekko chip doesn't have "awesome memory bandwidth" as somebody said above... it is just a slightly more efficient G3, limited to the 60x bus @ 100 MHz (IIRC). <hr></blockquote>
From Nintendo's pages:
External Bus 1.3GB/second peak bandwidth (32-bit address space, 64-bit data bus 162 MHz clock)
Not awesome but better than 60x.
[ 06-12-2002: Message edited by: Stoo ]</p>
<strong>
The GameCube's Gekko chip doesn't have "awesome memory bandwidth" as somebody said above... it is just a slightly more efficient G3, limited to the 60x bus @ 100 MHz (IIRC.</strong><hr></blockquote>
Stoo is right, it runs at 162 Mhz. What's more interesting though is that IBM says it supports some sort of data compression, like the recent graphics chips, that results in an "effective" 5.2 GB/s.
I need more information! Additionally, can anyone else confirm that Power5 and Power6 chips are being developed?
[ 06-12-2002: Message edited by: Big Mac ]</p>
<strong>I'm wondering if anyone can help out here with links. I can't seem to find much information on the Power4 on IBM's site. An index posted on has a link to PPC - Power is MIA. And when I use their search engine, the best I can find are press releases praising the Power4.
</strong><hr></blockquote>
Here's the <a href="http://www-1.ibm.com/servers/eserver/pseries/hardware/whitepapers/power4.html"; target="_blank">white paper</a> on the Power4.
<strong>Stoo is right, it runs at 162 Mhz. What's more interesting though is that IBM says it supports some sort of data compression, like the recent graphics chips, that results in an "effective" 5.2 GB/s.</strong><hr></blockquote>
Hmmm... you might be right about the clock rate. It is a modified 60x bus protocol, however, and doesn't sustain that peak throughput for very long (i.e. one cacheline burst at a time). The supposed "data compression" is simply fixed <-> float conversions in hardware together with a write combining write-thru pipe that was hacked onto the chip and has numerous issues, plus it must be explicitly coded for by the software developers. Very useful, to be sure, but it wouldn't be suited as-is in a Mac... besides which AltiVec can already do the fixed/float conversions better, and has cache streaming instructions and a higher clock rate.
The Apple Pi instructions could certainly be yet more improvements to the AltiVec cache controls -- that would be cool, but FSB bus speeds would need to increase to really make the most of it.
Oh -- one more thing about the GameCube... what I meant originally was that its memory is capable of more bandwidth than the processor can use. This is why the graphics engine running at full speed doesn't get in the way of the Gekkco. Hence the similar nature to the Xserve.
[ 06-12-2002: Message edited by: Programmer ]</p>
Unbelievable. That's why I like Appleinsider fo much. :eek:
[quote]Trinity is the code name for the Cube. <hr></blockquote>
I thought that Trinity was the name of the lead female character in "The Matrix." So, hopefully we'll be getting an Altivec-enhanced Matrix screensaver from Apple at MWNY '02.