<strong>Question: On the IBM PowerPC roadmap, there's still this .13, RIO, 1Ghz+ chip with SIMD. Is that the 970? Or is that some revision of the G3 with altivec that people are always talking about.</strong><hr></blockquote>
G3 + AltiVec? The Apple employee name "Michael Balas" comes to mind.
Michael Balas has delivered presentations and speeches all across the country. Audiences include regional and national advertising clubs, Fortune 500 business management consulting firms and government and educational institutions. He has also presented more than 100 installments of his highly acclaimed seminar to more than 20000 guests from every region of the country.<hr></blockquote>
His name pops up frequently in association with Final Cut Pro demonstrations and seminars.
Has this moved out of the realm of speculation?</strong><hr></blockquote>
*shrug* They had it as a confirmed date of arrival rather than expected so yeah I'd say it's likely sampling soon after or around the same time as 970 and in production late 2003.
[quote]Originally posted by BRussell:
<strong>Question: On the IBM PowerPC roadmap, there's still this .13, RIO, 1Ghz+ chip with SIMD. Is that the 970? Or is that some revision of the G3 with altivec that people are always talking about.</strong><hr></blockquote>
That chip = next G3. The 970 is different again as is the POWER series.
The only question is will there be a die shrink then an evolution or vice versa and which is IBM roadmapping for.
*shrug* They had it as a confirmed date of arrival rather than expected so yeah I'd say it's likely sampling soon after or around the same time as 970 and in production late 2003. </strong><hr></blockquote>
Thanks. I've followed this thread fairly closely but last I heard that was still less than speculation and just wishful thinking.
That chip = next G3. The 970 is different again as is the POWER series.
The only question is will there be a die shrink then an evolution or vice versa and which is IBM roadmapping for.</strong><hr></blockquote>
Unfortunately, the pdf shows an introduction date of 2004 for the next generation G3(re: the one with SIMID, multicore, Rapid I/O etc.). I was hoping it would be introduced before the 970. Oh well, hope the 970 beats the 2nd half of 2003 deadline.
SPEC INT isn't too bad, but there's nothing "semi real" about SPEC FP. For starters, 10 out of 14 tests are in Fortran!?!? That ought to raise a red flag in and of itself as to the relevance of the workloads.
I think it's fair to assume most Mac users workloads don't include the crap found in SPEC FP such as: Physics / Quantum Chromodynamics, Shallow Water Modeling, Computational Fluid Dynamics, High Energy Nuclear Physics Accelerator Design, etc.
In fact, I'd argue that only 1 of the 14 tests are relevant for Mac users's workloads: 3-D graphics library. Yet, even here, Mesa is an OpenGL work alike, yet it doesn't reflect how OpenGL would perform on a Mac. The Mac version of OpenGL is Simd Optimized, SPEC is not.
No, SPEC is fine for discussion, however, it's basically meaningless for cross platform comparison between PCs and Macs.
<strong>No, SPEC is fine for discussion, however, it's basically meaningless for cross platform comparison between PCs and Macs.</strong><hr></blockquote>
Really?
I mean, SPECFP does seem to yield a reasonable indication of the double precision floating performance of the box on which it is run. There aren't many C compilers out-optimizing fortran compilers for mathematically intense code -> one of the reasons there's so much fortran in science/engineering.
I'm not saying that a 20% SPEC FP advantage to mystery box B implies precisely a 20% advantage on a random piece of FP code of interest to me (or whoever), it never, ever, works that way. You want to know how _your_ code is going to work, you need to run your code on that box. It is just another piece of information, useful when used reasonably.
A fair number of the current cross-platform comparisons involve more complex issues than straight FP, for which you are correct. We need a SpecVP for the vector processors, and a consistent reporting on just how MP-aware a given app is.
<strong>Bigc, that line "Well then, here is a picture of the IBM 970 boxes being tested", and the pic was great. Thanks for the humor.
Outsider, interesting PDF. Aside from what you point out, I thought it was interesting on the last page, where it stated that the PPC 750CXe was currently shipping in 600MHz config and the PPC 750FX would be shipping in a 733MHz config in 11/02. The part that was interesting (as in confirming that Apple has a special deal with IBM, like they do with Moto) was that the iBook currently ships in 700MHz and 800MHz G3s. I'm not sure what version of the PPC 750 Apple uses though. That implies that Apple also has a special deal concerning the 9XX family. I say 9XX family because it is possible that Apple may not get the 970, but instead a 9XX (you take a guess as to what numerals the X's are, Me ... I'm not sure).</strong><hr></blockquote>
"I say 9XX family because it is possible that Apple may not get the 970, but instead a 9XX (you take a guess as to what numerals the X's are, Me ... I'm not sure)"
"I say 9XX family because it is possible that Apple may not get the 970, but instead a 9XX (you take a guess as to what numerals the X's are, Me ... I'm not sure)"
Ahh, yes and no. It's just that everything regarding the 970 has been so secretive, that it kinda made sense for IBM to release the 970, but have something else in mind for Apple.
Just as a general reminder, all the admins and moderators are emailed when we're PM'd, including me. So we don't let that feature go unchecked for months.
<strong>Just as a general reminder, all the admins and moderators are emailed when we're PM'd, including me. So we don't let that feature go unchecked for months. </strong><hr></blockquote>
There aren't many C compilers out-optimizing fortran compilers for mathematically intense code -> one of the reasons there's so much fortran in science/engineering.
</strong><hr></blockquote>
I don't know how true that is anymore, especially of C++ compilers. There's a lot of FORTRAN in those areas for primarily historical reasons (i.e. that's what the scientists & engineers know how to use, and they know how to use that because much of the existing code uses it -- a vicious circle). There isn't anything inherent about FORTRAN that makes it good at math... in fact there isn't really anything about it that makes it a good language. C/C++ certainly isn't perfect, but its a lot more appropriate for modern software engineering.
[quote]<strong>
A fair number of the current cross-platform comparisons involve more complex issues than straight FP, for which you are correct. We need a SpecVP for the vector processors, and a consistent reporting on just how MP-aware a given app is.</strong><hr></blockquote>
The problem with this is there aren't any standard tools that can take advantage of vector processors in a cross-platform manner. The SIMD units vary greatly in their capabilities so trying to build something that accurately reflects this and their performance is really tough.
MP-aware tests would be nice, and these days almost everybody supports the POSIX thread API so perhaps an MP-aware SPECmark suit isn't too far off. Certainly with all the hyperthreading, multithreading, and multicore processors that are coming this is going to be more and more important to reflect in benchmarks.
Always important to keep in mind that the workings of complex machines like these simply cannot be described by one or two numbers.
. . . everything regarding the 970 has been so secretive, that it kinda made sense for IBM to release the 970, but have something else in mind for Apple.
</strong><hr></blockquote>
I believe it is a big enough job to develop one new processor like this. Why slow it down by making two? Also, there is economy in having just one processor in higher volume production.
Regarding the 970 number as the first in a family of processors, this did cause me to wonder right away. The newer and better processors usually get higher numbers as they come along. Starting at 970 does not leave much room for big jumps up, to say 980 or 990. I don't know whether this struck anyone else as odd? What occurred to me is that IBM and Apple might be planning on lower performance chips in this family, for consumer applications. What would be common to the family? Could it be the bus, and/or 64 bits? Just a thought.
I don't know how true that is anymore, especially of C++ compilers. There's a lot of FORTRAN in those areas for primarily historical reasons (i.e. that's what the scientists & engineers know how to use, and they know how to use that because much of the existing code uses it -- a vicious circle). There isn't anything inherent about FORTRAN that makes it good at math... in fact there isn't really anything about it that makes it a good language. C/C++ certainly isn't perfect, but its a lot more appropriate for modern software engineering.
Hmm. I mostly agree. I was under the impression that a lot of the C-pointer math business and other shenanigans in C(and C++) make some of the math-oriented optimizations tougher. By allowing less flexibility in how you can frame the math in your program -> simpler for the optimizer.
In any case, when I've tried to compare the same algorithm in C vs fortran, the fortran has always come closer to the max theoretical FLOP count. Of course, that could be me
I believe it is a big enough job to develop one new processor like this. Why slow it down by making two? Also, there is economy in having just one processor in higher volume production.
You're correct, of course, and after yesterday, I know better. We'll all find out come next August. but I still think it's too long a wait. Hopefully we may catch a glimmer of hope in March.
It comes to something when we are desperately cheering a processor that appears to come second.
</strong><hr></blockquote>
The Itanium 2 is a big expensive server processor and not even targeted at the desktop market. It is Intel's top end powerhouse. Sure we are happy about the IBM 970, which may cost just a little more than a G4. Likely Apple will be able to buy a quad of 970s for less than one Itanium 2.
Comments
<strong>Question: On the IBM PowerPC roadmap, there's still this .13, RIO, 1Ghz+ chip with SIMD. Is that the 970? Or is that some revision of the G3 with altivec that people are always talking about.</strong><hr></blockquote>
G3 + AltiVec? The Apple employee name "Michael Balas" comes to mind.
<strong>Who is Michael Balas? </strong><hr></blockquote>
According to <a href="http://www.applesiders.com/avwhatis.html" target="_blank">this page</a> that I googled up:
[quote]Michael Balas,
Business Development Executive
Apple
[...]
Michael Balas has delivered presentations and speeches all across the country. Audiences include regional and national advertising clubs, Fortune 500 business management consulting firms and government and educational institutions. He has also presented more than 100 installments of his highly acclaimed seminar to more than 20000 guests from every region of the country.<hr></blockquote>
His name pops up frequently in association with Final Cut Pro demonstrations and seminars.
<strong>
Has this moved out of the realm of speculation?</strong><hr></blockquote>
*shrug* They had it as a confirmed date of arrival rather than expected so yeah I'd say it's likely sampling soon after or around the same time as 970 and in production late 2003.
[quote]Originally posted by BRussell:
<strong>Question: On the IBM PowerPC roadmap, there's still this .13, RIO, 1Ghz+ chip with SIMD. Is that the 970? Or is that some revision of the G3 with altivec that people are always talking about.</strong><hr></blockquote>
That chip = next G3. The 970 is different again as is the POWER series.
The only question is will there be a die shrink then an evolution or vice versa and which is IBM roadmapping for.
<strong>
*shrug* They had it as a confirmed date of arrival rather than expected so yeah I'd say it's likely sampling soon after or around the same time as 970 and in production late 2003. </strong><hr></blockquote>
Thanks. I've followed this thread fairly closely but last I heard that was still less than speculation and just wishful thinking.
<strong>
That chip = next G3. The 970 is different again as is the POWER series.
The only question is will there be a die shrink then an evolution or vice versa and which is IBM roadmapping for.</strong><hr></blockquote>
Unfortunately, the pdf shows an introduction date of 2004 for the next generation G3(re: the one with SIMID, multicore, Rapid I/O etc.). I was hoping it would be introduced before the 970. Oh well, hope the 970 beats the 2nd half of 2003 deadline.
<strong>
SPEC is a set of demi-real world tasks -- have a look:
<a href="http://www.spec.org/osg/cpu2000/papers/COMPUTER_200007-abstract.JLH.html" target="_blank">http://www.spec.org/osg/cpu2000/papers/COMPUTER_200007-abstract.JLH.html</a>
the specific benchmarks for SPEC INT: <a href="http://www.spec.org/osg/cpu2000/CINT2000/" target="_blank">http://www.spec.org/osg/cpu2000/CINT2000/</a>
the specific benchmarks for SPEC FP: <a href="http://www.spec.org/osg/cpu2000/CFP2000/" target="_blank">http://www.spec.org/osg/cpu2000/CFP2000/</a></strong><hr></blockquote>
SPEC INT isn't too bad, but there's nothing "semi real" about SPEC FP. For starters, 10 out of 14 tests are in Fortran!?!? That ought to raise a red flag in and of itself as to the relevance of the workloads.
I think it's fair to assume most Mac users workloads don't include the crap found in SPEC FP such as: Physics / Quantum Chromodynamics, Shallow Water Modeling, Computational Fluid Dynamics, High Energy Nuclear Physics Accelerator Design, etc.
In fact, I'd argue that only 1 of the 14 tests are relevant for Mac users's workloads: 3-D graphics library. Yet, even here, Mesa is an OpenGL work alike, yet it doesn't reflect how OpenGL would perform on a Mac. The Mac version of OpenGL is Simd Optimized, SPEC is not.
No, SPEC is fine for discussion, however, it's basically meaningless for cross platform comparison between PCs and Macs.
Steve
Or not...
<strong>No, SPEC is fine for discussion, however, it's basically meaningless for cross platform comparison between PCs and Macs.</strong><hr></blockquote>
Really?
I mean, SPECFP does seem to yield a reasonable indication of the double precision floating performance of the box on which it is run. There aren't many C compilers out-optimizing fortran compilers for mathematically intense code -> one of the reasons there's so much fortran in science/engineering.
I'm not saying that a 20% SPEC FP advantage to mystery box B implies precisely a 20% advantage on a random piece of FP code of interest to me (or whoever), it never, ever, works that way. You want to know how _your_ code is going to work, you need to run your code on that box. It is just another piece of information, useful when used reasonably.
A fair number of the current cross-platform comparisons involve more complex issues than straight FP, for which you are correct. We need a SpecVP for the vector processors, and a consistent reporting on just how MP-aware a given app is.
Thanks,
Jedai
<strong>Bigc, that line "Well then, here is a picture of the IBM 970 boxes being tested", and the pic was great. Thanks for the humor.
Outsider, interesting PDF. Aside from what you point out, I thought it was interesting on the last page, where it stated that the PPC 750CXe was currently shipping in 600MHz config and the PPC 750FX would be shipping in a 733MHz config in 11/02. The part that was interesting (as in confirming that Apple has a special deal with IBM, like they do with Moto) was that the iBook currently ships in 700MHz and 800MHz G3s. I'm not sure what version of the PPC 750 Apple uses though. That implies that Apple also has a special deal concerning the 9XX family. I say 9XX family because it is possible that Apple may not get the 970, but instead a 9XX (you take a guess as to what numerals the X's are, Me ... I'm not sure).</strong><hr></blockquote>
"I say 9XX family because it is possible that Apple may not get the 970, but instead a 9XX (you take a guess as to what numerals the X's are, Me ... I'm not sure)"
Well how does a 960 sound? ???
Is that the general idea you were going for?
D
[ 11-19-2002: Message edited by: DaveGee ]</p>
<strong>
"I say 9XX family because it is possible that Apple may not get the 970, but instead a 9XX (you take a guess as to what numerals the X's are, Me ... I'm not sure)"
Well how does a 960 sound? ???
Is that the general idea you were going for?
D
[ 11-19-2002: Message edited by: DaveGee ]</strong><hr></blockquote>
Ahh, yes and no. It's just that everything regarding the 970 has been so secretive, that it kinda made sense for IBM to release the 970, but have something else in mind for Apple.
[ 11-20-2002: Message edited by: MacJedai ]</p>
<strong>Just as a general reminder, all the admins and moderators are emailed when we're PM'd, including me. So we don't let that feature go unchecked for months. </strong><hr></blockquote>
Thanks, I wasn't sure
<strong>
There aren't many C compilers out-optimizing fortran compilers for mathematically intense code -> one of the reasons there's so much fortran in science/engineering.
</strong><hr></blockquote>
I don't know how true that is anymore, especially of C++ compilers. There's a lot of FORTRAN in those areas for primarily historical reasons (i.e. that's what the scientists & engineers know how to use, and they know how to use that because much of the existing code uses it -- a vicious circle). There isn't anything inherent about FORTRAN that makes it good at math... in fact there isn't really anything about it that makes it a good language. C/C++ certainly isn't perfect, but its a lot more appropriate for modern software engineering.
[quote]<strong>
A fair number of the current cross-platform comparisons involve more complex issues than straight FP, for which you are correct. We need a SpecVP for the vector processors, and a consistent reporting on just how MP-aware a given app is.</strong><hr></blockquote>
The problem with this is there aren't any standard tools that can take advantage of vector processors in a cross-platform manner. The SIMD units vary greatly in their capabilities so trying to build something that accurately reflects this and their performance is really tough.
MP-aware tests would be nice, and these days almost everybody supports the POSIX thread API so perhaps an MP-aware SPECmark suit isn't too far off. Certainly with all the hyperthreading, multithreading, and multicore processors that are coming this is going to be more and more important to reflect in benchmarks.
Always important to keep in mind that the workings of complex machines like these simply cannot be described by one or two numbers.
<strong>
. . . everything regarding the 970 has been so secretive, that it kinda made sense for IBM to release the 970, but have something else in mind for Apple.
</strong><hr></blockquote>
I believe it is a big enough job to develop one new processor like this. Why slow it down by making two? Also, there is economy in having just one processor in higher volume production.
Regarding the 970 number as the first in a family of processors, this did cause me to wonder right away. The newer and better processors usually get higher numbers as they come along. Starting at 970 does not leave much room for big jumps up, to say 980 or 990. I don't know whether this struck anyone else as odd? What occurred to me is that IBM and Apple might be planning on lower performance chips in this family, for consumer applications. What would be common to the family? Could it be the bus, and/or 64 bits? Just a thought.
[ 11-20-2002: Message edited by: snoopy ]</p>
Originally posted by Nevyn:
...so much fortran in science/engineering.
<hr></blockquote>
I don't know how true that is anymore, especially of C++ compilers. There's a lot of FORTRAN in those areas for primarily historical reasons (i.e. that's what the scientists & engineers know how to use, and they know how to use that because much of the existing code uses it -- a vicious circle). There isn't anything inherent about FORTRAN that makes it good at math... in fact there isn't really anything about it that makes it a good language. C/C++ certainly isn't perfect, but its a lot more appropriate for modern software engineering.
Hmm. I mostly agree. I was under the impression that a lot of the C-pointer math business and other shenanigans in C(and C++) make some of the math-oriented optimizations tougher. By allowing less flexibility in how you can frame the math in your program -> simpler for the optimizer.
In any case, when I've tried to compare the same algorithm in C vs fortran, the fortran has always come closer to the max theoretical FLOP count. Of course, that could be me
<strong>I think, Eugene, that it convincingly dusts a Pentium 4, especially in fpu, and its just, estimated behind an Itanic 2 on fpu. Impressive.
[ 10-15-2002: Message edited by: Lemon Bon Bon ]</strong><hr></blockquote>
It comes to something when we are desperately cheering a processor that appears to come second.
<strong>
I believe it is a big enough job to develop one new processor like this. Why slow it down by making two? Also, there is economy in having just one processor in higher volume production.
<snip>
[ 11-20-2002: Message edited by: snoopy ]</strong><hr></blockquote>
You're correct, of course, and after yesterday, I know better. We'll all find out come next August. but I still think it's too long a wait. Hopefully we may catch a glimmer of hope in March.
<strong>
It comes to something when we are desperately cheering a processor that appears to come second.
</strong><hr></blockquote>
The Itanium 2 is a big expensive server processor and not even targeted at the desktop market. It is Intel's top end powerhouse. Sure we are happy about the IBM 970, which may cost just a little more than a G4. Likely Apple will be able to buy a quad of 970s for less than one Itanium 2.