Itanium2 vs. POWER4 benchmarks
Some disclaimers: the Itanium 2 isn't actually available in these configurations for another month or so, benchmarks aren't necessarily everything, and we have zero evidence so far that the POWER4 (or derivative) will be in future Apple computers.
All that being said, these new SPEC results from the Itanium 2 are interesting and a little depressing:
<a href="http://www.queru.com/articles/Benchmarks3.html"; target="_blank">http://www.queru.com/articles/Benchmarks3.html</A>;
Why depressing? Because the 1300 MHz POWER4, with 128MB (!!) of L3 [off-chip] cache, gets beaten by a 1000 MHz Itanium 2 with 3MB of L3 [on-chip] cache. The Itanium 2 is a much cheaper chip.
And a POWER4 (or derivative/successor) is a favorite among us Great Unwashed for a high-performance Macintosh CPU.
This isn't meant to be a doom-and-gloom post. I realize that extrapolating the performance of a family of chips from a single example is an exercise in futility. However, it's unfortunate to realize that even if Apple switched to POWER4 immediately, they'd still be behind the performance offered by the competitor.
[sigh]
All that being said, these new SPEC results from the Itanium 2 are interesting and a little depressing:
<a href="http://www.queru.com/articles/Benchmarks3.html"; target="_blank">http://www.queru.com/articles/Benchmarks3.html</A>;
Why depressing? Because the 1300 MHz POWER4, with 128MB (!!) of L3 [off-chip] cache, gets beaten by a 1000 MHz Itanium 2 with 3MB of L3 [on-chip] cache. The Itanium 2 is a much cheaper chip.
And a POWER4 (or derivative/successor) is a favorite among us Great Unwashed for a high-performance Macintosh CPU.
This isn't meant to be a doom-and-gloom post. I realize that extrapolating the performance of a family of chips from a single example is an exercise in futility. However, it's unfortunate to realize that even if Apple switched to POWER4 immediately, they'd still be behind the performance offered by the competitor.
[sigh]
Comments
<strong>Why depressing? Because the 1300 MHz POWER4, with 128MB (!!) of L3 [off-chip] cache, gets beaten by a 1000 MHz Itanium 2 with 3MB of L3 [on-chip] cache. The Itanium 2 is a much cheaper chip.
</strong><hr></blockquote>
The Itanium2 may be cheaper than a Power4 (most processors are, no?
Bye,
RazzFazz
Don't worry. There are a couple of issues here. The first is that the POWER4 is due for a die shrink right about NOW. The second is that the manner in which AIX manages memory on that system is nowhere NEAR optimized. People (engineers) are speculating that this alone will result in a huge improvement. Third, a POWER4 could run Photoshop, Quark, Shake, FCP3, without a rewrite, whereas things are ALOT different for IA-64.
[quote]
And a POWER4 (or derivative/successor) is a favorite among us Great Unwashed for a high-performance Macintosh CPU.<hr></blockquote>
High-performance != top-of-the-line
I must have washed, it ain't my favorite for a "high perfomance Macintosh CPU", derivative or not.
But, I will say I am ignorant of CPU design, but I am washed - at least weekly
I hear it's pending a die shrink and mhz hike. And maybe it'll have some of it's own 'bottlenecks' to performance removed as well.
I await the next Power benchmarks with great interest.
The Itanic 2 scrapes by in Integer and Fpu.
Worried. Don't think so.
Itanic aint gonna be used in the 'desktop' market as we know for some while if ever. Different market. While the Power is wayyyyy ahead of the mhz 'kings'.
I'd take a Power variant for the next POWERMAC anyday!!!
Lemon Bon Bon :cool:
[ 08-02-2002: Message edited by: Lemon Bon Bon ]</p>
I just purchased a Dell 2650 with 2 Xeon 2.4's with HyperThreading. According to those specs my machine is slower than a PIV 2.5 which we know is not true because of the testing we have done.
There is much more to the equation than how fast ONE process can run. The other issue when you start dealing with server CPU's is the fact that they can handle so much more load than a standard desktop CPU. They may only be about 10% faster at running one task at a time but throw a database on both machines and then you will see HUGE difference. Especially if the database is thread safe running on multiple CPU's.
EDIT:
One more thing...is anyone out there actually using the original Itanium in a server? We looked at using it but were concerned about our applications being 100% compatible with it. At this point we would actually purchase an 8 way Xeon machine before I would purchase a dual or quad Itanium 2 machine.
[ 08-02-2002: Message edited by: Mike ]</p>
Aw heck, say it ain't so. You mean all those companies out there can't scrap plans for their Power 4 purchases, save a buck and replace them with itanic, er um, I mean titanic oh oh I meam Itanium 2. Maybe they could bolt a couple of itanics together to replace the Power 4. And here I thought based on those benchmarks that 128MB cache just became obsolete.
I guess I don't have to kill myself now.
<strong>"but benchmarks like that don't mean jack!"
Aw heck, say it ain't so. You mean all those companies out there can't scrap plans for their Power 4 purchases, save a buck and replace them with itanic, er um, I mean titanic oh oh I meam Itanium 2. Maybe they could bolt a couple of itanics together to replace the Power 4. And here I thought based on those benchmarks that 128MB cache just became obsolete.
I guess I don't have to kill myself now.</strong><hr></blockquote>
What did happen to the titanic?
Have you noticed that Dell decided to NOT release a system based on the Itanium 2!
The bench looked like it was using just one core of the single chip version....
The other reason people like the POWER4 is it has an insane amount of interchip memory bandwidth - precisely where the current powermacs are deficient.
<strong>Some disclaimers: the Itanium 2 isn't actually available in these configurations for another month or so, benchmarks aren't necessarily everything, and we have zero evidence so far that the POWER4 (or derivative) will be in future Apple computers.
All that being said, these new SPEC results from the Itanium 2 are interesting and a little depressing:
<a href="http://www.queru.com/articles/Benchmarks3.html"; target="_blank">http://www.queru.com/articles/Benchmarks3.html</a>;
Why depressing? Because the 1300 MHz POWER4, with 128MB (!!) of L3 [off-chip] cache, gets beaten by a 1000 MHz Itanium 2 with 3MB of L3 [on-chip] cache. The Itanium 2 is a much cheaper chip.
And a POWER4 (or derivative/successor) is a favorite among us Great Unwashed for a high-performance Macintosh CPU.
This isn't meant to be a doom-and-gloom post. I realize that extrapolating the performance of a family of chips from a single example is an exercise in futility. However, it's unfortunate to realize that even if Apple switched to POWER4 immediately, they'd still be behind the performance offered by the competitor.
[sigh]</strong><hr></blockquote>
Do you want cheese with that whine?
Currently, IBM uses the POWER4 only in servers. The least expensive of these is $12,500.
Even in top of the line [$48,000] RS/6000 44P Model 270 workstation, Blue Blue runs the 375 MHz Power3-II. The monitor is extra. Big Blue stills uses the PPC 604e in some lesser RS/6000 models.
What this all gets down is that the chips in currently shippinng Macs have awesome power--and reliability. The chips in the Macs introduced later this month will be more so. Life as a Mac buyer is good. It is only getting better.
<strong>
What this all gets down is that the chips in currently shippinng Macs have awesome power--</strong><hr></blockquote>
... perhaps ... or more likely it means IBM has certain customers by the balls and locked into cheesy legacy solutions.
Hang on, didn't IBM just buy PwC consulting?
Oooooh boy - then it's gone way past the scrotum folks, at this point, it's throat, jewels, blood supply, and needless to say, bank account.
[sudden insight]
[ 08-02-2002: Message edited by: OverToasty ]</p>
Also, it looks like Intel has learned Sun's art trick that inflates the FP scores.
<strong>I don't think SPEC FP is that relevant to most Mac users, and it's a dead heat in INT.
Also, it looks like Intel has learned Sun's art trick that inflates the FP scores.</strong><hr></blockquote>
Its hardly a "trick" and the scores aren't "inflated". The IA64 floating point unit is highly optimized and superscalar, and it turns in excellent FPU performance for the same reason that Sparc, Alpha, and POWER chips have had excellent FPU performance.
That doesn't change the fact that the Itanium is late, overly hot, get poor yields, and requires better compilers than Intel can currently create. Can't stand the whole EPIC design philosophy myself, I hope it continues to fall flat on its face.
<strong>
Its hardly a "trick" and the scores aren't "inflated". The IA64 floating point unit is highly optimized and superscalar, and it turns in excellent FPU performance for the same reason that Sparc, Alpha, and POWER chips have had excellent FPU performance.
</strong><hr></blockquote>
The trick he was referring to is clearly visible <a href="http://www.spec.org/osg/cpu2000/results/res2002q3/cpu2000-20020607-01412.html"; target="_blank">here</a>. Note how Sun managed to super optimize their compiler for that one specific floting point test compared to all other FP related tests.
<strong>
... perhaps ... or more likely it means IBM has certain customers by the balls and locked into cheesy legacy solutions.
Hang on, didn't IBM just buy PwC consulting?
Oooooh boy - then it's gone way past the scrotum folks, at this point, it's throat, jewels, blood supply, and needless to say, bank account.
[sudden insight]
[ 08-02-2002: Message edited by: OverToasty ]</strong><hr></blockquote>
This is not the mainframe market where VS COBOL rules. This is the market of UNIX-based servers and engineering workstations. Here, IBM has competition from Sun, HP, SGI, and now scads of Intel boxes running Linux. This is a market where you can get fired if you buy IBM and it doesn't work.
The light you saw was in a mirror.
<strong>
Here, IBM has competition from Sun, HP, SGI, and now scads of Intel boxes running Linux. This is a market where you can get fired if you buy IBM and it doesn't work.
The light you saw was in a mirror.</strong><hr></blockquote>
Perhaps ... but only if we forget one very minor point.
"Migration"
Short answer: people pay a lot of money for certainty - too much - especially in a world where you can get fired even if you go with IBM ... so if you're already with IBM, and it works, why not pay stoopid amounts of cash to avoid a painful - and potentially career "modifying" migration to sun. Besides, upper management doesn't know a 604e from a Moto Guzzi ... IBM knows this, it's their bread and butter.
So it is agreed on the golf course, so it shall be done.
Like I said: by the short 'n curlies.
<strong>The trick he was referring to is clearly visible <a href="http://www.spec.org/osg/cpu2000/results/res2002q3/cpu2000-20020607-01412.html"; target="_blank">here</a>. Note how Sun managed to super optimize their compiler for that one specific floting point test compared to all other FP related tests.</strong><hr></blockquote>
I agree that Sun pulled a fast one there, but do you have a similar breakdown for the Itanium scores?
<strong>
I agree that Sun pulled a fast one there, but do you have a similar breakdown for the Itanium scores?</strong><hr></blockquote>
<a href="http://www.spec.org/osg/cpu2000/results/"; target="_blank">http://www.spec.org/osg/cpu2000/results/</a>; has all the breakdowns you could want. Just click on the HTML version of the results. I did not agree with the poster that Intel used a similar compiler optimization as Sun. I was just clarifying what "trick" the poster was referring to.
To say the truth - I never liked SPEC at all. But it was considered a perfect benchmark for the big ones. "unlike the tools you have, Mike". Ah, see.
Now I know its a compiler benchmark. Thats why the Macs are so slow compared. Yeah I know how it sounds - wherever the Mac is slow, we just flame the benchmark.
The question anyway is - what is a platform benchmark? Does the CPU win that has the most power on common code? Or does the one win, that has the better SIMD engine? Does a good or a bad compiler belong the the platform's advantage/disadvantage, or should we compare 1:1 Assembler commands?
In the end, we could say every CPU and its platform should do what it can to win a benchmark. no matter what they use - if a system has an additional DSP, then this chip can be used for many things - also in "real world applications". So why not take it into the benchmark?
If you don't use the second CPU, and if you forget about the SIMD unit and if you don't optimize for the specific CPU, the tests are no longer accurate, since with the example of AltiVec, this unit is used where they can, and not just within RC5 Benchmarks. So I think its integral part of the G4 and needs to be tested. (Besides its the G4's best part).
All this shows me how different the systems can be. But they all have one thing in common: One or more High Performance CPU sitting and Idle'ing ;-)