Xbench benchmarks...

kybernaut · Post by **kybernaut** » Fri Jun 18, 2004 12:54 am

Since Xbench does no longer crash since 06/17 nightly, I'm posting my benchmarks now.

I ONLY benched the CPU (unchecked all other options). I think this way results should be fairly comparable. Get yourself Xbench 1.1.3 from http://xbench.com/ and post your results!

----------------------

Host: Athlon 700, Win XP
Guest: 10.3.1, 256 MB

-----------------

For comparison here the same benchmark (CPU only) of my G3 600 MHz iBook.

kybernaut · Post by **kybernaut** » Fri Jun 18, 2004 1:08 am

sorry, my free provider has some problems today. The images should appear earlier or later

Anyway, here are the results: PPC 0.13, iBook 39.93. But the 'subjective' performance of PearPC is better than these figure suggest

--kybernaut

HDC · Post by **HDC** » Fri Jun 18, 2004 9:08 pm

And this suggests that benchmarking of emulators with a synthetic software benchmarks is irrelevant, especially if you're trying to benchmark some build with a broken timer which tends to run at 10x speed. Use a stopwatch

robojam · Post by **robojam** » Fri Jun 18, 2004 9:08 pm

I am using a sun dial...

kybernaut · Post by **kybernaut** » Fri Jun 18, 2004 9:15 pm

HDC wrote:And this suggests that benchmarking of emulators with a synthetic software benchmarks is irrelevant, especially if you're trying to benchmark some build with a broken timer which tends to run at 10x speed. Use a stopwatch

Who says that the timer is still broken?

robojam · Post by **robojam** » Fri Jun 18, 2004 9:19 pm

I don't think it's broken, but from the point of view of benchtesting software, I don't think you will get accurate results if it's not a realtime clock.

It will probably give you a good approximation, but I don't think it will be entirely accurate.

HDC · Post by **HDC** » Fri Jun 18, 2004 9:24 pm

I know it was fixed some time ago. But CVS builds from, say, week ago were still broken, and I wasn't aware of current situation in this field

Anyway, your result seems to be very far from reality

Fox Mulder · Post by **Fox Mulder** » Mon Jun 21, 2004 11:52 pm

Hi All,
I've been trying to benchmark PPC for some time. Found that X-bench is not a reliable benchmarking tool, and... it crashes.
SpeedX seems more reliable. It only calculates common instructions and FP instructions. the results are given in MIPS, which is quite a useful benchmark for an emulator. i have been able to benchmark UAE, basilik II and Mandrake 10 in the same way.
You can find SpeedX 1.2 here: http://www.macupdate.com/info.php/id/12102

Since the 0.12 release I have seen a real interesting speed increase and the 0.3pre CVS and SDL builds from 16 june have shown the highest benchmarks so far.
On my P4 3Ghz, 1024-786 x 32 refresh at 40, CVS shows 87.1 MIPS (889 Mhz. in profiler screen, but that could be a miscalculation) and the SDL built shows even 91.4 MIPS. The speed has increased since early June from 83.6 MIPS. And indeed it feels a little faster. Great job so far!
The SDL-build of 16-6 feels rather smooth and even word-processing runs at acceptable speed, but quicktime movies are stil jerky.
Keep it up!

F.M.

[/img]

gb · Post by gb » Tue Jun 22, 2004 7:54 am

Try to use nbench/bytemark, I have uploaded OSX binaries here:
<http://gwenole.beauchesne.free.fr/benchmarks/>

I also put Linux/PPC and Linux/X86 statically linked binaries I use for performance evaluation. I will try to provide MacOS FAT binaries too but that will use CodeWarrior instead of gcc in other builds.

gb · Post by gb » Tue Jun 22, 2004 8:01 am

Note: you can't check timings with nbench. It would probably be better to use ssbench with alternate args to get reasonable stopwatchable execution time (around 10 minutes at most). So, I had only a build for Linux/PPC on integer part.

CaptainValor · Post by **CaptainValor** » Tue Jun 22, 2004 2:22 pm

Thanks for the info, Fox Mulder. I'm gonna try out SpeedX now thanks to your recommendation.