Centrino twice as fast as PowerBook in Java development?!

[Joonas Lehtinen]

As a naive benchmark I packaged Millstone UI Library sources together with Apache Ant and a simple test-script to recompile Millstone with Ant. Even though this benchmark is naive, I think it gives a pretty good picture of the performance from a developer point of view.

The most interesting founding is that Pentium-M based notebook seems to be twice as fast as PowerBook - clock for clock. The situation is even worse because one can currently get 2.0Ghz Pentium-M notebook, while the fastest PowerBook is equipped with 1.67Ghz G4.

Please tell me that I am wrong!

More details can be found here. You can test it yourself by using my simple benchmark.

There is already 16 different machines tested including PC:s and Macs. Two examples: Thinkpad T42 1.6Ghz does the compilation in 13 secs, while my 1.5Ghz PowerBook G4 uses 27 secs.

 

[Capt Underpants]

Well the Pentium-M stomps all over the G4 processor. No surprise here, man...

 

[khammack]

That's not surprising, but I'd bet that it's due to the difference in front side bus speed rather than operations per cycle.

-kev

 

[Joonas Lehtinen]

Well the Pentium-M stomps all over the G4 processor. No surprise here, man...

The same applies also for G5: A modest 1.6Ghz Pentium-M laptop outperforms PowerMac G5 dual 2.0Ghz by a large margin in Java compilation. In fact it might be even be faster than dual 2.5Ghz G5, but I haven't tested that really - just extrapolated the results...

 

[varmit]

So that is seconds to complete, and how many cycles it too. Ok, it looks like the PowerMacs are faster than the Xenon machine, but the single Pentium M is faster than both. Why is the Athlon XP 1.9 faster than the Athlon 2700+?

 

[Joonas Lehtinen]

um, tell me why when two computers have the same gigacycles or the same amout of seconds, but then don't have the same of the other? Take the two on the left side of your graph, same in seconds, but not gigacycles. Explain.

Gigacycles are probably not used anywhere, but I calculated them to chart by multiplying gigaherts with runtime. So this answers the question: how many cycles the computer used for compiling the source. This somewhat reflects the efficiency of the computer.

Anyway - the quick answer would have been: their CPU:s run at different clocks.

Try this link for more detailed specs on the machines. For example: java version seems to make a difference on PC:s: 1.4.2 is significantly faster than 1.5.0 in this benchmark. Strange...

 

[Joonas Lehtinen]

So that is seconds to complete, and how many cycles it too. Ok, it looks like the PowerMacs are faster than the Xenon machine, but the single Pentium M is faster than both. Why is the Athlon XP 1.9 faster than the Athlon 2700+?

That is a complete mystery. I have the Athlon 2700+ sitting next to me and the other result is submitted by one tester who complited the test correctly. Neither of us can explain the results. Machines are running the same java version, same Linux kernel, same chipset, same processor. That 2700+ machine even has more RAM than the other. The only explanation I can think of is that there is something wrong with the settings of my Athlon or kernel (APIC mode, memory bus speed, ...).

 

[saabmp3]

That is a complete mystery. I have the Athlon 2700+ sitting next to me and the other result is submitted by one tester who complited the test correctly. Neither of us can explain the results. Machines are running the same java version, same Linux kernel, same chipset, same processor. That 2700+ machine even has more RAM than the other. The only explanation I can think of is that there is something wrong with the settings of my Athlon or kernel (APIC mode, memory bus speed, ...).

What kind of optimization level are you using in compiling your kernel? I've found that this can make a vast difference in the AMD processors. Much more so than an Intel processor. It has alot to do with the location of the memory controller.

BEN

 

[tech4all]

Did anyone else notice in the middle the "PowerBook G5 1.5Ghz"?

 

[plinden]

As a naive benchmark I packaged Millstone UI Library sources together with Apache Ant and a simple test-script to recompile Millstone with Ant. Even though this benchmark is naive, I think it gives a pretty good picture of the performance from a developer point of view.

The most interesting founding is that Pentium-M based notebook seems to be twice as fast as PowerBook - clock for clock. The situation is even worse because one can currently get 2.0Ghz Pentium-M notebook, while the fastest PowerBook is equipped with 1.67Ghz G4.

Please tell me that I am wrong!

More details can be found here. You can test it yourself by using my simple benchmark.

There is already 16 different machines tested including PC:s and Macs. Two examples: Thinkpad T42 1.6Ghz does the compilation in 13 secs, while my 1.5Ghz PowerBook G4 uses 27 secs.

26 seconds was my best time, and 35 seconds was the average time over 10 runs, with a Thinkpad T40p, 1.6GHz Pentium-M, Windows XP, 1.5GByte RAM, using cygwin and the following JDK:
Java(TM) 2 Runtime Environment, Standard Edition (build 1.4.2_04-b05)
Java HotSpot(TM) Client VM (build 1.4.2_04-b05, mixed mode)

I'll try using a Windows cmd shell if I can remember how to edit .bat files.

Looks like your linux kernal has been optimized in some way, not that I know a llot about that, I admit.

 

[plinden]

26 seconds was my best time, and 35 seconds was the average time over 10 runs, with a Thinkpad T40p, 1.6GHz, 1.5GByte RAM, using cygwin and the following JDK:
Java(TM) 2 Runtime Environment, Standard Edition (build 1.4.2_04-b05)
Java HotSpot(TM) Client VM (build 1.4.2_04-b05, mixed mode)

I'll try using a Windows cmd shell if I can remember how to edit .bat files.

Looks like your linux kernal has been optimized in some way, not that I know a llot about that, I admit.

Is it bad form to quote yourself? Anyway, looks like, under WinXP, my laptop is on a par with a Powerbook 1.5GHz.

 

[Joonas Lehtinen]

Is it bad form to quote yourself? Anyway, looks like, under WinXP, my laptop is on a par with a Powerbook 1.5GHz.

Try to run it directly without the cygwin in the middle.. I would guess that it creates a lot of overhead in the process.

Try to do the steps in the test.sh by hand. I would expect the windows xp to be roughly in the same ballpark as the Linux.

- Joonas

 

[plinden]

Try to run it directly without the cygwin in the middle.. I would guess that it creates a lot of overhead in the process.

Try to do the steps in the test.sh by hand. I would expect the windows xp to be roughly in the same ballpark as the Linux.

- Joonas

No, it's still in the same range of about 30 seconds.

I wouldn't expect cygwin to have a large overhead - it has a very small memory and processer footprint.

You can't make any conclusions from the data you have now. You need more WinXP, Mac OSX and linux compilations, but perhaps compiler performance is highly dependent on optimization. It may be that the linux javac compiler is better optimized than the WinXP or Mac OSX compiler.

Anyway, assuming that all these results hold up, the only conclusion you can really make is that Linux is the best platform for Java development, and that the G4 and Pentium-M are equivalent for this task.

That would be fine for me, since I'm a Java developer, and I was planning on installing Linux along side WinXP on my laptop, and to get a PowerBook when I can persuade the wife that it's a good idea.

 

[plinden]

Couldn't find anything similar to this after a quick google (don't have time now to do it properly) but I found this, which tested Java processes performance on linux and WinXP http://www.numlock.ch/variouspapers/mono_bench/

It shows WinXP runs Java processes at half the speed of linux.