"The question is what Motorola can do with a 50% difference in clock speed"<hr></blockquote>
I know this is nitpicking, but wouldn't an Intel chip @ 2.0GHz be 100% difference than a 1.0GHz G4? And wouldn't the release of the 2.4GHz Intel chip would put it @ 140% difference in clock speed?
What's AMD's fastest chip...1.8GHZ? So Motorola is only 80% slower than AMD in clock speed.
Which only continues to amaze me that in Altivec capable operations the G4 more than keeps up. This really must burn the arse's of the engineers designing Intel's SIMD.
I don't need to be lectured on the relationship between MHz and performance. You're thinking in simple minded terms if you actually think that a G4 is faster than a Pentium 4 irrespective of MHz. Speed scales with MHz for any given architecture, and what it comes down to is that Intel has scaled the x86 architecture better than AIM has done with the PPC. Yes the PPC is a more efficient CPU and is capable of completing more instructions per clock cycle (ICC) under most circumstances, but the ICC is not a boundless number. ICC is constant for the PPC and even falls as Moto adds more stages to the pipeline. </strong><hr></blockquote>
Speed does scale with any architecture, you are correct. Do you account for the tradeoffs with deep pipelines though? Why can a PIII outperform a PIV clock cycle for cycle? When a BPU makes a mistake a PIV suffers a minimum of 19 wasted cycles and up to 30 whereas a G4 suffers only 7 to 13.
Now, I'm not saying the G4 outperforms the P4 straight up on any giving task, that would be foolish. The G4 does outperform in a lot of the tasks that most pro's use them for (where AltiVec is used) and that the gap by MHz looks big but the performance gap is not as bad as you make it look.
ICC is not constant as you say as the "newer" G4 actually has added to it's ICC in the AltiVec unit. This does not work across the board but does give a speed increase.
<strong>
Think of it this way: CPU performance can be expressed as instructions per clock cycle (ICC) times cycles per second (GHz). This is drastically oversimplified but it serves for explaining the idea to your feeble mind. Now, suppose for the sake of argument that the PPC can do 40% more instructions per clock cycle than the x86. But at the same time, the PPC runs at 1.2 GHz, and the x86 is at 3.0 GHz. So we have something like this:
x=instructions per clock cycle.
PPC: (1.40x)(1.2 GHz) = 1.68 x
x86: (1.00x)(3.0 GHz) = 3.00 x
So even though the PPC is 40% more efficient, the x86 completes ( 3.00/1.68 = 1.79) 79% more instructions per second by the use of brute GHz. </strong>
Your calculations don't take into account the larger penalties incurred for "bubbles" in the pipe or incorrect branch predictions. It doesn't take into account the larger micro-ops that also cause a slight penalty when they are called from an on chip ROM vs the L1 cache. Performance can't be measured in real world terms by a simple calculation. Just as GFLOPS alone can not determine that a chip is superior for any and all tasks.
<strong>
Does this help you understand the looming performance gap between Macs and Wintels? I hope so.
</strong>
No it does not, as only real world performance can judge the difference between different computers (which is more than just CPU design as well). You make it sound as though a P4 is 2.4 times faster than a G4 simply based on MHz gap which is untrue. Does the fastest single processor P4 outperform the fastest single processor G4 overall. YES, but is the gap so in real world so performance so large as to make a poor choice, NO.
Apple does need to improve there roll out of faster chips to at least maintain parity with performance, no question about it, so does AMD. But if Mac's were so far behind a lot of Pro's would have already switched since a significant difference in time is money.
In all reality Apple's move from 1 GHz to 2 GHz will give it a better performance boost than Intel's move from 2 GHz to 3 GHz, which if rolled out in the same time frame will actually change the performance gap.
<strong>Speed does scale with any architecture, you are correct. Do you account for the tradeoffs with deep pipelines though? Why can a PIII outperform a PIV clock cycle for cycle?</strong><hr></blockquote>
You are close to the truth, but fail to understand the developments in the various architectures. An architecture is a line of cores that support the same machinecode in a backwards compatible way. These cores are usually sufficiently different to make them very difficult to compare by clockspeed alone. The pentium II was a totally different core from the Pentium I, so you cannot compare the clockspeed of the two easily. The difference between the Pentium III and the Pentium IV is even worse. Caches can also have a big effect, the original Celeron without cache was a dog, even though the core is exactly the same as the Pentium II.
Similarly the 7450 has a different core from a 7400 (longer pipeline), so a 7450-1Gz will not be twice as fast as an 7400-500Mhz. The 7450-1Gz with 2MB of DDR cache will also outperform the same processor with less cache.
So in summary: You cannot even compare processors of the same architecture by clockspeed, let alone processors from totally different architectures. Only benchmarks can supply us with meaningfull data. The current benchmarks suggest that the G4 is highly superior for some tasks and highly inferior for other tasks. So your choice should depend on the things you do and not on Junkyard Dawg's trolls about his inability to brag to his friends.
Motorola's actually not so bad, depending on where you pick your end points.
Between New Years 2001 and the end of January 2002 Moto managed to nearly double the clock speed of the G4.
[quote] PPC: (1.40x)(1.2 GHz) = 1.68 x
x86: (1.00x)(3.0 GHz) = 3.00 x
<hr></blockquote>
Well, a couple of things: First, I'd disagree with your multiplier for the PPC, at least when compared with the P4. The multiplier should be at least 1.6, and probably 1.7. When compared with the Athlon or the P3, a multiplier of 1.2 or 1.3 would probably be more appropriate.
Second, you've choosen 3.0 GHz as the clock rate for the P4, which suggests you're comparing CPUs assumed to exist at year's end, as Intel has stated that they intend on reaching that speed at that time. By year's end, if Moto can keep up the same rate of improvement in clock speed they've been able to maintain for the past year, we'll probably be at 1.5GHz or 1.6GHz at the year's end. (Remember-- Moto has 1.1GHz G4s right now, even though Apple doesn't use them. I don't think it would be too big of a leap for them to get to 1.6 by the year's end.)
So with my numbers:
PPC: (1.60x)(1.5 GHz) = 2.40 x -to-
PPC: (1.70x)(1.6 GHz) = 2.72 x
x86: (1.00x)(3.0 GHz) = 3.00 x
Yeah, Apple'll still be behind, but it doesn't seem quite so dire...
<strong>I know this is nitpicking, but wouldn't an Intel chip @ 2.0GHz be 100% difference than a 1.0GHz G4? And wouldn't the release of the 2.4GHz Intel chip would put it @ 140% difference in clock speed?</strong><hr></blockquote>
Yes, you're quite right... I knew there was something odd with that when I posted it. To be more precise, then, the question is what Motorola can do when it's only AT 50% of clock speed.
<strong>Apple needs to move away from AltiVec but keep the speed the provide. IBMs SIMD engine works on all code not just the optimised bits so that might be the answer (there are few mentions of G4 at WWDC this year).
</strong><hr></blockquote>
Apple needs to move away from Altivec? Why? That would be a very dumb move. It's the one area where Macs enjoy a competitive advantage.
As for keeping the speed of Altivec, but removing it, yeah, that would be a nice trick. Of course, if AIM (Apple/Ibm/Motorola) were able to produce such a chip, imagine how much faster it would be if they left Altivec in...
What's this rubbish about IBM's SIMD engine working on all code? There's no magic bullet to make this happen. It certainly isn't a function of hardware. If anything, this will happen throuh auto-vectorizing compilers. To date, with all of Intel's research budget on this topic, they've enjoyed very little success with this.
I know this is nitpicking, but wouldn't an Intel chip @ 2.0GHz be 100% difference than a 1.0GHz G4? And wouldn't the release of the 2.4GHz Intel chip would put it @ 140% difference in clock speed?
</strong><hr></blockquote>
To nitpick a little more, 2.0GHz is 100% greater than 1.0GHz and 1.0GHz is 50% less. To be more accurate you should state the difference in these terms to prevent ambiguity.
You are close to the truth, but fail to understand the developments in the various architectures. An architecture is a line of cores that support the same machinecode in a backwards compatible way. These cores are usually sufficiently different to make them very difficult to compare by clockspeed alone. The pentium II was a totally different core from the Pentium I, so you cannot compare the clockspeed of the two easily. The difference between the Pentium III and the Pentium IV is even worse. Caches can also have a big effect, the original Celeron without cache was a dog, even though the core is exactly the same as the Pentium II.
Similarly the 7450 has a different core from a 7400 (longer pipeline), so a 7450-1Gz will not be twice as fast as an 7400-500Mhz. The 7450-1Gz with 2MB of DDR cache will also outperform the same processor with less cache.
So in summary: You cannot even compare processors of the same architecture by clockspeed, let alone processors from totally different architectures. Only benchmarks can supply us with meaningfull data. The current benchmarks suggest that the G4 is highly superior for some tasks and highly inferior for other tasks. So your choice should depend on the things you do and not on Junkyard Dawg's trolls about his inability to brag to his friends.</strong><hr></blockquote>
I know what you are saying and I worded my response poorly in that area, such is the case when at work trying to fire off a quick one. I wanted to make a point about the difference in design philosophy vs performance/MHz. Hmmm, maybe I'll wait til I get home before I get into this too much.
is it possible or feasible or even logical to have an MP system with one G4 and 2 G3 processors? Yes I know expensive as hell, maybe? Whenever an app needs AltiVec code it could then just send instructions to the G4?
As I see it, i think the G3 is ready to go beyond 1Ghz, way beyond . This way you get a fast PPC processor at high speeds and the power of AltiVec.
***********
Disclaimer: I know basically nothing about processor architecture, so any schooling you wanna give me is greatly appreciated :cool:
"This is the only version of Giants that can use multiple processors. We actually added multi-processing support while we were doing the port, as well as doing various optimizations on the graphics routines (and changing the graphics library from Direct3D to OpenGL). The end result is we're getting the same framerates on our Macintosh dual 500 MHz G4s with a GeForce 2 on an AGP 2x bus as we do on a brand-new, top of the line Wintel AMD 1.3 GHz Ahtlon with a GeForce 3 on an AGP 4x bus and DDR RAM. "
Said Omni Group of their conversion of Giants Kabuko (or whatever its called...)
Mhz. Pcs. PPCs. Theoretical. Perceptions and Real world performance...
<strong>"This is the only version of Giants that can use multiple processors. We actually added multi-processing support while we were doing the port, as well as doing various optimizations on the graphics routines (and changing the graphics library from Direct3D to OpenGL). The end result is we're getting the same framerates on our Macintosh dual 500 MHz G4s with a GeForce 2 on an AGP 2x bus as we do on a brand-new, top of the line Wintel AMD 1.3 GHz Ahtlon with a GeForce 3 on an AGP 4x bus and DDR RAM. "
Said Omni Group of their conversion of Giants Kabuko (or whatever its called...)
Mhz. Pcs. PPCs. Theoretical. Perceptions and Real world performance...
Lemon Bon Bon.
</strong><hr></blockquote>
This is a cool example, but I can show you a dozen cases that demonstrate the opposite. Its easy to be led astray by trying to draw conclusions from individual comparisons. This case no doubt had a poor Direct3D implementation and an excellent OpenGL/G4 implementation. The reverse is far more common -- developers optimize for the WIntel machines and then just bang it across to the Mac and sell whatever they end up with. Yet another reason for Motorola/IBM to build PowerPCs that make scalar code go really fast, and for Apple/Metrowerks to make their compilers much better (and vectorizing, if possible).
Comments
Check this out.
JD for Sheriff?
Lemon Bon Bon
<img src="graemlins/smokin.gif" border="0" alt="[Chilling]" />
"The question is what Motorola can do with a 50% difference in clock speed"<hr></blockquote>
I know this is nitpicking, but wouldn't an Intel chip @ 2.0GHz be 100% difference than a 1.0GHz G4? And wouldn't the release of the 2.4GHz Intel chip would put it @ 140% difference in clock speed?
What's AMD's fastest chip...1.8GHZ? So Motorola is only 80% slower than AMD in clock speed.
Which only continues to amaze me that in Altivec capable operations the G4 more than keeps up. This really must burn the arse's of the engineers designing Intel's SIMD.
<strong><a href="http://www.macedition.com/images/wanted/wanted.pdf"; target="_blank">http://www.macedition.com/images/wanted/wanted.pdf</a>;
Check this out.
JD for Sheriff?
Lemon Bon Bon
<img src="graemlins/smokin.gif" border="0" alt="[Chilling]" /> </strong><hr></blockquote>
Does a sheriff bark ?
..er...
'Deputy Junkyard Dawg!'
Lemon Bon Bon
<strong>
I don't need to be lectured on the relationship between MHz and performance. You're thinking in simple minded terms if you actually think that a G4 is faster than a Pentium 4 irrespective of MHz. Speed scales with MHz for any given architecture, and what it comes down to is that Intel has scaled the x86 architecture better than AIM has done with the PPC. Yes the PPC is a more efficient CPU and is capable of completing more instructions per clock cycle (ICC) under most circumstances, but the ICC is not a boundless number. ICC is constant for the PPC and even falls as Moto adds more stages to the pipeline. </strong><hr></blockquote>
Speed does scale with any architecture, you are correct. Do you account for the tradeoffs with deep pipelines though? Why can a PIII outperform a PIV clock cycle for cycle? When a BPU makes a mistake a PIV suffers a minimum of 19 wasted cycles and up to 30 whereas a G4 suffers only 7 to 13.
Now, I'm not saying the G4 outperforms the P4 straight up on any giving task, that would be foolish. The G4 does outperform in a lot of the tasks that most pro's use them for (where AltiVec is used) and that the gap by MHz looks big but the performance gap is not as bad as you make it look.
ICC is not constant as you say as the "newer" G4 actually has added to it's ICC in the AltiVec unit. This does not work across the board but does give a speed increase.
<strong>
Think of it this way: CPU performance can be expressed as instructions per clock cycle (ICC) times cycles per second (GHz). This is drastically oversimplified but it serves for explaining the idea to your feeble mind. Now, suppose for the sake of argument that the PPC can do 40% more instructions per clock cycle than the x86. But at the same time, the PPC runs at 1.2 GHz, and the x86 is at 3.0 GHz. So we have something like this:
x=instructions per clock cycle.
PPC: (1.40x)(1.2 GHz) = 1.68 x
x86: (1.00x)(3.0 GHz) = 3.00 x
So even though the PPC is 40% more efficient, the x86 completes ( 3.00/1.68 = 1.79) 79% more instructions per second by the use of brute GHz. </strong>
Your calculations don't take into account the larger penalties incurred for "bubbles" in the pipe or incorrect branch predictions. It doesn't take into account the larger micro-ops that also cause a slight penalty when they are called from an on chip ROM vs the L1 cache. Performance can't be measured in real world terms by a simple calculation. Just as GFLOPS alone can not determine that a chip is superior for any and all tasks.
<strong>
Does this help you understand the looming performance gap between Macs and Wintels? I hope so.
</strong>
No it does not, as only real world performance can judge the difference between different computers (which is more than just CPU design as well). You make it sound as though a P4 is 2.4 times faster than a G4 simply based on MHz gap which is untrue. Does the fastest single processor P4 outperform the fastest single processor G4 overall. YES, but is the gap so in real world so performance so large as to make a poor choice, NO.
Apple does need to improve there roll out of faster chips to at least maintain parity with performance, no question about it, so does AMD. But if Mac's were so far behind a lot of Pro's would have already switched since a significant difference in time is money.
In all reality Apple's move from 1 GHz to 2 GHz will give it a better performance boost than Intel's move from 2 GHz to 3 GHz, which if rolled out in the same time frame will actually change the performance gap.
<strong>Speed does scale with any architecture, you are correct. Do you account for the tradeoffs with deep pipelines though? Why can a PIII outperform a PIV clock cycle for cycle?</strong><hr></blockquote>
You are close to the truth, but fail to understand the developments in the various architectures. An architecture is a line of cores that support the same machinecode in a backwards compatible way. These cores are usually sufficiently different to make them very difficult to compare by clockspeed alone. The pentium II was a totally different core from the Pentium I, so you cannot compare the clockspeed of the two easily. The difference between the Pentium III and the Pentium IV is even worse. Caches can also have a big effect, the original Celeron without cache was a dog, even though the core is exactly the same as the Pentium II.
Similarly the 7450 has a different core from a 7400 (longer pipeline), so a 7450-1Gz will not be twice as fast as an 7400-500Mhz. The 7450-1Gz with 2MB of DDR cache will also outperform the same processor with less cache.
So in summary: You cannot even compare processors of the same architecture by clockspeed, let alone processors from totally different architectures. Only benchmarks can supply us with meaningfull data. The current benchmarks suggest that the G4 is highly superior for some tasks and highly inferior for other tasks. So your choice should depend on the things you do and not on Junkyard Dawg's trolls about his inability to brag to his friends.
Between New Years 2001 and the end of January 2002 Moto managed to nearly double the clock speed of the G4.
[quote] PPC: (1.40x)(1.2 GHz) = 1.68 x
x86: (1.00x)(3.0 GHz) = 3.00 x
<hr></blockquote>
Well, a couple of things: First, I'd disagree with your multiplier for the PPC, at least when compared with the P4. The multiplier should be at least 1.6, and probably 1.7. When compared with the Athlon or the P3, a multiplier of 1.2 or 1.3 would probably be more appropriate.
Second, you've choosen 3.0 GHz as the clock rate for the P4, which suggests you're comparing CPUs assumed to exist at year's end, as Intel has stated that they intend on reaching that speed at that time. By year's end, if Moto can keep up the same rate of improvement in clock speed they've been able to maintain for the past year, we'll probably be at 1.5GHz or 1.6GHz at the year's end. (Remember-- Moto has 1.1GHz G4s right now, even though Apple doesn't use them. I don't think it would be too big of a leap for them to get to 1.6 by the year's end.)
So with my numbers:
PPC: (1.60x)(1.5 GHz) = 2.40 x -to-
PPC: (1.70x)(1.6 GHz) = 2.72 x
x86: (1.00x)(3.0 GHz) = 3.00 x
Yeah, Apple'll still be behind, but it doesn't seem quite so dire...
<strong>I know this is nitpicking, but wouldn't an Intel chip @ 2.0GHz be 100% difference than a 1.0GHz G4? And wouldn't the release of the 2.4GHz Intel chip would put it @ 140% difference in clock speed?</strong><hr></blockquote>
Yes, you're quite right... I knew there was something odd with that when I posted it. To be more precise, then, the question is what Motorola can do when it's only AT 50% of clock speed.
<a href="http://membres.lycos.fr/applefr/Rumeurs.html"; target="_blank">http://membres.lycos.fr/applefr/Rumeurs.html</a>;
(English Translation)
<a href="http://216.239.35.120/translate_c?h...fr/Rumeurs.html"; target="_blank">http://216.239.35.120/translate_c?h...fr/Rumeurs.html</a>;
Most likely, its just more BS but it sure makes for good reading.
<strong>Apple needs to move away from AltiVec but keep the speed the provide. IBMs SIMD engine works on all code not just the optimised bits so that might be the answer (there are few mentions of G4 at WWDC this year).
</strong><hr></blockquote>
Apple needs to move away from Altivec? Why? That would be a very dumb move. It's the one area where Macs enjoy a competitive advantage.
As for keeping the speed of Altivec, but removing it, yeah, that would be a nice trick. Of course, if AIM (Apple/Ibm/Motorola) were able to produce such a chip, imagine how much faster it would be if they left Altivec in...
What's this rubbish about IBM's SIMD engine working on all code? There's no magic bullet to make this happen. It certainly isn't a function of hardware. If anything, this will happen throuh auto-vectorizing compilers. To date, with all of Intel's research budget on this topic, they've enjoyed very little success with this.
[ 04-04-2002: Message edited by: SteveS ]</p>
<strong>
I know this is nitpicking, but wouldn't an Intel chip @ 2.0GHz be 100% difference than a 1.0GHz G4? And wouldn't the release of the 2.4GHz Intel chip would put it @ 140% difference in clock speed?
</strong><hr></blockquote>
To nitpick a little more, 2.0GHz is 100% greater than 1.0GHz and 1.0GHz is 50% less. To be more accurate you should state the difference in these terms to prevent ambiguity.
<strong>Here is a French rumor site with some interesting configurations of future macs.
<a href="http://membres.lycos.fr/applefr/Rumeurs.html"; target="_blank">http://membres.lycos.fr/applefr/Rumeurs.html</a>;
(English Translation)
<a href="http://216.239.35.120/translate_c?h...fr/Rumeurs.html"; target="_blank">http://216.239.35.120/translate_c?h...fr/Rumeurs.html</a>;
Most likely, its just more BS but it sure makes for good reading.</strong><hr></blockquote>
this is for sure another BS : the speed of the powerbook are too fast : Tibook toaster, the new powerbook from Apple
There is lot of similar predictions here on AI.
[ 04-04-2002: Message edited by: powerdoc ]</p>
New iMac with CRT and G3 =Impossible!
new iBook with G3=possible but unlikely
Towers, who knows
<strong>
You are close to the truth, but fail to understand the developments in the various architectures. An architecture is a line of cores that support the same machinecode in a backwards compatible way. These cores are usually sufficiently different to make them very difficult to compare by clockspeed alone. The pentium II was a totally different core from the Pentium I, so you cannot compare the clockspeed of the two easily. The difference between the Pentium III and the Pentium IV is even worse. Caches can also have a big effect, the original Celeron without cache was a dog, even though the core is exactly the same as the Pentium II.
Similarly the 7450 has a different core from a 7400 (longer pipeline), so a 7450-1Gz will not be twice as fast as an 7400-500Mhz. The 7450-1Gz with 2MB of DDR cache will also outperform the same processor with less cache.
So in summary: You cannot even compare processors of the same architecture by clockspeed, let alone processors from totally different architectures. Only benchmarks can supply us with meaningfull data. The current benchmarks suggest that the G4 is highly superior for some tasks and highly inferior for other tasks. So your choice should depend on the things you do and not on Junkyard Dawg's trolls about his inability to brag to his friends.</strong><hr></blockquote>
I know what you are saying and I worded my response poorly in that area, such is the case when at work trying to fire off a quick one. I wanted to make a point about the difference in design philosophy vs performance/MHz. Hmmm, maybe I'll wait til I get home before I get into this too much.
As I see it, i think the G3 is ready to go beyond 1Ghz, way beyond
***********
Disclaimer: I know basically nothing about processor architecture, so any schooling you wanna give me is greatly appreciated :cool:
Said Omni Group of their conversion of Giants Kabuko (or whatever its called...)
Mhz. Pcs. PPCs. Theoretical. Perceptions and Real world performance...
Lemon Bon Bon.
<strong>"This is the only version of Giants that can use multiple processors. We actually added multi-processing support while we were doing the port, as well as doing various optimizations on the graphics routines (and changing the graphics library from Direct3D to OpenGL). The end result is we're getting the same framerates on our Macintosh dual 500 MHz G4s with a GeForce 2 on an AGP 2x bus as we do on a brand-new, top of the line Wintel AMD 1.3 GHz Ahtlon with a GeForce 3 on an AGP 4x bus and DDR RAM. "
Said Omni Group of their conversion of Giants Kabuko (or whatever its called...)
Mhz. Pcs. PPCs. Theoretical. Perceptions and Real world performance...
Lemon Bon Bon.
This is a cool example, but I can show you a dozen cases that demonstrate the opposite. Its easy to be led astray by trying to draw conclusions from individual comparisons. This case no doubt had a poor Direct3D implementation and an excellent OpenGL/G4 implementation. The reverse is far more common -- developers optimize for the WIntel machines and then just bang it across to the Mac and sell whatever they end up with. Yet another reason for Motorola/IBM to build PowerPCs that make scalar code go really fast, and for Apple/Metrowerks to make their compilers much better (and vectorizing, if possible).
"GIVE US THE GODDAMN G5...SHOW US THE MONEY...SHOW ME THE MONEEEEEEEEEEEEEEEEEEE!!!!!!"
lEmOn bOn BoN