Moki, I'm very curious are you implying that things will be relatively the same for the next year "see you next year". OR can we look forward to some great performance jumps in the next revision of the powermac?
With IBM possibly taking the reigns from Motorola, how soon might this happen, are we looking at the very short term or the relatively long term?
Thanks I'm really enjoying everyone's posts... did I mention EV Nova rocks .
See, now that's what I'm talking about. The main difference besides clock rate is the number of outstanding transactions on the bus at once because this allows them to be handled back-to-back. If they can increase this number and use them effectively (questionable, but who knows?) then they might be able to push the efficiency higher. Increasing cache line size would help as well, although I doubt they'll do that in the G4.
Hmmm... DDR might help just because it can let the memory controller get to the next transaction sooner, and thus keep the MPX busy more of the time.
I wish they'd hurry up and ship an Xserve so that we can see some real benchmarks from the thing.
<strong>Moki, I'm very curious are you implying that things will be relatively the same for the next year "see you next year". OR can we look forward to some great performance jumps in the next revision of the powermac?
With IBM possibly taking the reigns from Motorola, how soon might this happen, are we looking at the very short term or the relatively long term?</strong><hr></blockquote>
It seems pretty clear to me that Moki is saying 2003 before we see the product of IBM's work. The real reason to mention POWER4 is that it shows that IBM has an architecture that is basically equivalent to Intel's P4 in nature, the current implementation of which demonstrates an emphasis on floating point rather than integer calculations. Its also 64-bit.
If the next version of the G4 lack of DDR memory controller, don't you think they can get a larger L3 cache to improve the performance.
The L3 cache is clocked at quarter speed (if we speak of real clock speed, and not the equivalent in clock speed like do Apple). A 1,2 ghz G4 will have a 300 mhz DDR L3 cache 128 bit wide. This kind of memory still exist in high end graphic card. 32 MB L3 cache already exist in the power 4.
Don't you think that a 32 MB L3 cache (costing 100 $ more if i take in example the prize of graphics cards) will increase highly the performance of the G4 for intensive bandwitch applications ?
Don't forget, that while SPEC claims and wants to be the best cross platform CPu benchmark around, it's actually fairly (even very) poor at doing so.
That is due to some guidelines in SPEC testing, as well as some inevitable inequities in how SPEC benchmarks are produced.
Honestly, testing CPUs without being allowed to optimize and/or use the custom execution units is just outdated.
A real real-world comparison still says a lot more about how much faster different platforms are, especially as the quality of the software also is measured. And then, don't even start mentioning the fact that 50% of all SPEC benchmarks are fake or manipulated, ie not sicking to the rules.
So if you want to hear my personal humble opinion:
SPEC can blow my shoes,at least until they radically revamp their philosophy and software.
<strong>A real real-world comparison still says a lot more about how much faster different platforms are, especially as the quality of the software also is measured.</strong><hr></blockquote>
What would you suggest other than a Photoshop battery of tests?
What would you suggest other than a Photoshop battery of tests?
Mandricard
AppleOutsider</strong><hr></blockquote>
That depends on what you want to know about performance. If you are interested in how fast QuakeIII runs, then run a battery of QuakeIII tests. If you are interested in MPEG encoding, then run each platform's best MPEG encoder.
It is really hard to come up with a good benchmark because computing isn't about producing a single number (or two or four). People want a single number that makes it obvious which machine is better, but its rather silly actually. Do you buy a car based soley on how much horsepower it has at peak RPM? It has its place, but the real issue is "how fast does it get you where you are going?"
I would much prefer numbers from benchmarks, at least, rather than from theoretical numbers from tech papers.</strong><hr></blockquote>
Agreed, but I have seen a couple of sets of actual benchmarks done by different people -- none of which I have handy. They all showed numbers very close to the sustained values... or higher due to small data set sizes and the benefits of the cache (i.e. not valid test cases for measuring memory bandwidth). Note that these tests were streaming memory performance, random I/O is slower since cache hint is harder. I'm sure some time with Google could turn something up, but I've got work to do.
Anything you can get your hands on...Photosop, Quake, RC5, SETI, Maya, Cinema4D and all that stuff. I take it it will be the naked VGA version that you get, thus try to stick with benchmarks that don't use the graphics card too much, even if that means dropping all the games.
We're interested in memory bandwidth and sustained performance. Thus doing some SETI, RC5 and a STREAM suite for RAM throughput would be nice to have.
Agreed, but I have seen a couple of sets of actual benchmarks done by different people -- none of which I have handy. They all showed numbers very close to the sustained values... or higher due to small data set sizes and the benefits of the cache (i.e. not valid test cases for measuring memory bandwidth). Note that these tests were streaming memory performance, random I/O is slower since cache hint is harder. I'm sure some time with Google could turn something up, but I've got work to do.</strong>
We've talked about this before Programmer. And I've seen some of the benchmarks that show the memory performance improvements. The more solid proof I'm looking for is from non-AltiVec app benchmarks between the 750 and 7450.
<strong>The G5 will either debut this July as the next PowerMac or as the next iteration after the G4 bump this July.</strong><hr></blockquote>
So then if not July then MWSF?
[quote] I believe the next major chip Apple will use in its computers will not be from MOT -- and IBM is at the top of the list in terms of who the new dancing partner will be.
See you in a year. <hr></blockquote>
See, you did it again! So, then you keep throwing those smilies like nah, nah-nah, nah, nah! I know something you don't!
To clear up, are you saying one year from now like next MWNY or within a years time like possibly MWSF?
This is interesting especially coupled to what has been discussed on this thread. But one thing confuses me - why on earth would IBM develop a chip (that Apple may use), only to have them compete for the desktop/workstation space?
IBM still have workstations based on the 604 and Power3 (yet at an almighty cost) so what makes people think that Apple will be able to offer a more cost effective solution?
This is interesting especially coupled to what has been discussed on this thread. But one thing confuses me - why on earth would IBM develop a chip (that Apple may use), only to have them compete for the desktop/workstation space?
IBM still have workstations based on the 604 and Power3 (yet at an almighty cost) so what makes people think that Apple will be able to offer a more cost effective solution?
It would be remarkable if they did though...</strong><hr></blockquote>
I can't imagine IBM sell 50,000 POWER4 boxes per year, but Apple could potentially sell 250,000+. Economies of scale might push the price down, but whether it could be brought down to prosumer levels is anybodys guess - the cheapest pSeries 690 @ 1.1 Ghz x 8 is $450,000!
Comments
With IBM possibly taking the reigns from Motorola, how soon might this happen, are we looking at the very short term or the relatively long term?
Thanks I'm really enjoying everyone's posts... did I mention EV Nova rocks
<strong>[code]
Comparison of Bus Bandwidths in (Mbytes/sec.)
Device Bus Freq Peak Maximum Sustained
MPC750 100MHz 800 640 246
MPC74xx 100MHz 800 640 640
MPC74xx 133MHz 1064 851 851
</pre><hr></blockquote></strong><hr></blockquote>
See, now that's what I'm talking about. The main difference besides clock rate is the number of outstanding transactions on the bus at once because this allows them to be handled back-to-back. If they can increase this number and use them effectively (questionable, but who knows?) then they might be able to push the efficiency higher. Increasing cache line size would help as well, although I doubt they'll do that in the G4.
Hmmm... DDR might help just because it can let the memory controller get to the next transaction sooner, and thus keep the MPX busy more of the time.
I wish they'd hurry up and ship an Xserve so that we can see some real benchmarks from the thing.
<strong>Moki, I'm very curious are you implying that things will be relatively the same for the next year "see you next year". OR can we look forward to some great performance jumps in the next revision of the powermac?
With IBM possibly taking the reigns from Motorola, how soon might this happen, are we looking at the very short term or the relatively long term?</strong><hr></blockquote>
It seems pretty clear to me that Moki is saying 2003 before we see the product of IBM's work. The real reason to mention POWER4 is that it shows that IBM has an architecture that is basically equivalent to Intel's P4 in nature, the current implementation of which demonstrates an emphasis on floating point rather than integer calculations. Its also 64-bit.
The L3 cache is clocked at quarter speed (if we speak of real clock speed, and not the equivalent in clock speed like do Apple). A 1,2 ghz G4 will have a 300 mhz DDR L3 cache 128 bit wide. This kind of memory still exist in high end graphic card. 32 MB L3 cache already exist in the power 4.
Don't you think that a 32 MB L3 cache (costing 100 $ more if i take in example the prize of graphics cards) will increase highly the performance of the G4 for intensive bandwitch applications ?
That is due to some guidelines in SPEC testing, as well as some inevitable inequities in how SPEC benchmarks are produced.
Honestly, testing CPUs without being allowed to optimize and/or use the custom execution units is just outdated.
A real real-world comparison still says a lot more about how much faster different platforms are, especially as the quality of the software also is measured. And then, don't even start mentioning the fact that 50% of all SPEC benchmarks are fake or manipulated, ie not sicking to the rules.
So if you want to hear my personal humble opinion:
SPEC can blow my shoes,at least until they radically revamp their philosophy and software.
G-news
<strong>A real real-world comparison still says a lot more about how much faster different platforms are, especially as the quality of the software also is measured.</strong><hr></blockquote>
What would you suggest other than a Photoshop battery of tests?
Mandricard
AppleOutsider
From MOT's whitepaper on the G4:
<a href="http://e-www.motorola.com/brdata/PDFDB/docs/G4WP.pdf"; target="_blank">http://e-www.motorola.com/brdata/PDFDB/docs/G4WP.pdf</a>;
Comparison of Bus Bandwidths in (Mbytes/sec.)
[code]
Device Bus Freq Peak Maximum Sustained
MPC750 100MHz 800 640 246
MPC74xx 100MHz 800 640 640
MPC74xx 133MHz 1064 851 851
</pre><hr></blockquote><hr></blockquote>
I would much prefer numbers from benchmarks, at least, rather than from theoretical numbers from tech papers.
<strong>
What would you suggest other than a Photoshop battery of tests?
Mandricard
AppleOutsider</strong><hr></blockquote>
That depends on what you want to know about performance. If you are interested in how fast QuakeIII runs, then run a battery of QuakeIII tests. If you are interested in MPEG encoding, then run each platform's best MPEG encoder.
It is really hard to come up with a good benchmark because computing isn't about producing a single number (or two or four). People want a single number that makes it obvious which machine is better, but its rather silly actually. Do you buy a car based soley on how much horsepower it has at peak RPM? It has its place, but the real issue is "how fast does it get you where you are going?"
<strong>[code]
Device Bus Freq Peak Maximum Sustained
MPC750 100MHz 800 640 246
MPC74xx 100MHz 800 640 640
MPC74xx 133MHz 1064 851 851
</pre><hr></blockquote><hr></blockquote>
I would much prefer numbers from benchmarks, at least, rather than from theoretical numbers from tech papers.</strong><hr></blockquote>
Agreed, but I have seen a couple of sets of actual benchmarks done by different people -- none of which I have handy. They all showed numbers very close to the sustained values... or higher due to small data set sizes and the benefits of the cache (i.e. not valid test cases for measuring memory bandwidth). Note that these tests were streaming memory performance, random I/O is slower since cache hint is harder. I'm sure some time with Google could turn something up, but I've got work to do.
We're interested in memory bandwidth and sustained performance. Thus doing some SETI, RC5 and a STREAM suite for RAM throughput would be nice to have.
G-news
Agreed, but I have seen a couple of sets of actual benchmarks done by different people -- none of which I have handy. They all showed numbers very close to the sustained values... or higher due to small data set sizes and the benefits of the cache (i.e. not valid test cases for measuring memory bandwidth). Note that these tests were streaming memory performance, random I/O is slower since cache hint is harder. I'm sure some time with Google could turn something up, but I've got work to do.</strong>
We've talked about this before Programmer. And I've seen some of the benchmarks that show the memory performance improvements.
<strong>The G5 will either debut this July as the next PowerMac or as the next iteration after the G4 bump this July.</strong><hr></blockquote>
So then if not July then MWSF?
[quote] I believe the next major chip Apple will use in its computers will not be from MOT -- and IBM is at the top of the list in terms of who the new dancing partner will be.
See you in a year. <hr></blockquote>
See, you did it again! So, then you keep throwing those smilies like nah, nah-nah, nah, nah! I know something you don't!
To clear up, are you saying one year from now like next MWNY or within a years time like possibly MWSF?
<strong>The G5 will either debut this July as the next PowerMac or as the next iteration after the G4 bump this July.</strong><hr></blockquote>
Is this prediction based on intuition, something a friend told you, are you Steve Jobs or what?
<a href="http://www.theregister.co.uk/content/61/25722.html"; target="_blank">http://www.theregister.co.uk/content/61/25722.html</a>;
<strong>Hmmmm...:
<a href="http://www.theregister.co.uk/content/61/25722.html"; target="_blank">http://www.theregister.co.uk/content/61/25722.html</a></strong><hr></blockquote>;
Ya beat me to it!
Yes, very interesting indeed...
This is interesting especially coupled to what has been discussed on this thread. But one thing confuses me - why on earth would IBM develop a chip (that Apple may use), only to have them compete for the desktop/workstation space?
IBM still have workstations based on the 604 and Power3 (yet at an almighty cost) so what makes people think that Apple will be able to offer a more cost effective solution?
It would be remarkable if they did though...
<strong>Re; the Reg article about Power4 low end.
This is interesting especially coupled to what has been discussed on this thread. But one thing confuses me - why on earth would IBM develop a chip (that Apple may use), only to have them compete for the desktop/workstation space?
IBM still have workstations based on the 604 and Power3 (yet at an almighty cost) so what makes people think that Apple will be able to offer a more cost effective solution?
It would be remarkable if they did though...</strong><hr></blockquote>
I can't imagine IBM sell 50,000 POWER4 boxes per year, but Apple could potentially sell 250,000+. Economies of scale might push the price down, but whether it could be brought down to prosumer levels is anybodys guess - the cheapest pSeries 690 @ 1.1 Ghz x 8 is $450,000!
POWER4Mac, just $100,000 with a free iPod