How fast is PearPC: Benchmarks

[Ribtorus]

The latest PearPC Athlon Build from RichardGoodwin.com ran Strata3D CX without difficulty (meaning the app started up OK).



The time to complete the render was 7 hours:10 minutes:35 seconds

The time to complete the render on the host was 8 minutes :35 seconds.

That's a slowdown factor of [b:fbdcc6ca38]50.6X[/b:fbdcc6ca38], which is in good agreement with the slowdown factor found with Cinebench 2003 rendering.

[Ribtorus]

Checking out other Strata 3D render times for a variety of machines, the average time to render for a 1GHz G4 mac (the approximate speed given for my emulated mac) running OS X is 37 minutes.

That give me a slowdown factor of about [b:0102d1d7fb]11.5X[/b:0102d1d7fb] from the reported speed of my emulated mac.

There was one mac dual G4 with 1GB ram running Win98 under emulation (the reverse of what I'm doing), reporting a time of over [b:0102d1d7fb]30 hours[/b:0102d1d7fb]. That's more than four times slower than my emulation, and factoring in processor speed, may be about [b:0102d1d7fb]2X[/b:0102d1d7fb] slower than my pearPC installation.

[Ribtorus]

Another benchmark.

This time it's Quicktime 6.5 Pro.


I downloaded this quicktime movie from Apple and save it to the hard drive of both the Host PC as well as the emulated mac.

http://www.apple.com/imac/video/480.html

This is the large version of the new G5 iMac ad that is 7.2 MB

I then opened it in Quicktime and exported it to an AVI file. I used Quicktime's default video and audio settings.

Time to export in the host WinXP, 4minutes 39 seconds.

Time to export in the Guest OS X 10.3.5, 42 minutes 16 seconds.

That's a slowdown of [b:e66b4ad793]9.1X[/b:e66b4ad793] over the host machine.

I have a time for the same test run in an actual 1GHz G4. The time to export was 10 minutes. That's a slowdown of only [b:e66b4ad793]4.2X[/b:e66b4ad793] over a similarly clocked G4.

[Ribtorus]

Yet another Quicktime benchmark:

This time I installed the Ogg Vorbis plug-in into Quicktime 6.5 Pro, and converted an .aif music file to a compressed Ogg file.

Time to convert a 3 minute .aif song to ogg in the host machine: 18 seconds.

Time to convert the 3 minute .aif to Ogg in the guest OS: 920 seconds.

That's a slowdown of [b:22bf41c806]51X[/b:22bf41c806] over the host machine.

I tried converting a .wav to .mp3 in iTunes, but after a few minutes of converting, iTunes in the emulated PearPC system would crash.

Without knowing the particular strengths and optimisations inherent in particular applications and plug-ins, it's very difficult to predict how much of a performance hit can be expected with PearPC.

The range in the performance slowdown is very large, from more than [b:22bf41c806]50X[/b:22bf41c806] to under [b:22bf41c806]3X[/b:22bf41c806] in the apps I've tested so far.

[prasys]

Well I somewhat Love Circles (mathz crazy already)

So I will try to run http://sourceforge.net/projects/projectpi/ (Project Pi)

in MacOS X and on Windows (compile it) and try to run....Well , I will run it on my Real Mac and on PearPC...Well see the result...

[Ribtorus]

Circles are cool. I often feel like I'm running around inside one.

I found a small Pi calculator that uses a random number generator to demonstrate the Monte Carlo method. It's just for demonstration purposes, and it's not cross platform, but I imagine it demonstrates an important aspect of the emulated CPU numerical power.

http://homepage.mac.com/alk/picalculator.html

At the bottom of the page are three versions, one for 68k processors, one for PowerPC with OS 7 to OS 9, and the third for OSX.

I ran both the OSX version in PearPC/OSX 10.3.5 and the 68K version in Basilisk OS 8.1.



I set the counter to 2000 iterations, with the check boxes as shown in the screenshot. The time to complete was 76 seconds.



The same run in Basilisk/OS 8.1 (2000 iterations) took only 8 seconds.

Unfortunately there's no exact windows equivalent that I could find, so I can't tell how much of a slowdown there is compared to the host.

I have asked some people to give me an idea of where PearPC falls in terms of performance relative to other macs. So far I have one result, a G3 450 that turned in a time of 7.2 seconds for a 2000 iteration run.

That gives my PearPC about a [b:3698ddd7a6]10.5X[/b:3698ddd7a6] slowdown compared to a G3 450.

[User]

I am using other benchmarking utility it is name is 'speedrun'. http://mac.megalink.com/files/speedrun.sit. I think it shows for me much more realable results


My Duron CPU results are 11

On PowerPC G4 500 they are 170

I thinks my system emulate at 1/13 speed

My specs : duron 850. 512 MB Ram[/url]

[Ribtorus]

Thanks for the link.



The score shows my PearPC emulated G3 1.06GHz running about [b:492205cbc0]4[/b:492205cbc0] to [b:492205cbc0]4.5X[/b:492205cbc0] slower than a real 1000MHz G4 and about [b:492205cbc0]8X[/b:492205cbc0] slower than a Dual G5 2GHz.
That is not really reflected in the benchmarking with Photoshop or Bryce5 3D or Strata 3D CX.

However, the overall score of [b:492205cbc0]90[/b:492205cbc0] does seem reasonable in terms of placing my emulated G3 just below a real G3 266 when I consider the Photoshop benchmark.

So on the one hand the score looks reasonable, yet on the other it looks far too high.

[Ribtorus]

Here's a brief summary of the benchmarking I have done so far. All tests were performed with a 0.4pre build of pearPC, with only the XBench test using an emulated G4, the others with a G3. All were done with PearPC given 512MB ram.

For Photoshop 7, considering a variety of filtres and functions, my emulated 1GHz G3 performance is about [b:82596d5991]5X[/b:82596d5991] slower than the CPU speed given by Panther/PearPC in the system profiler.

For Bryce 5, the 1GHz emulated system performs about [b:82596d5991]9.5X[/b:82596d5991] slower than the CPU speed given in the system profiler.

For Xbench, the emulated G4 (not G3 this time) performs about [b:82596d5991]16X[/b:82596d5991] slower than the CPU speed given in the System profiler.

For CineBench 2003, the emulated G3 performs about [b:82596d5991]16X[/b:82596d5991] slower than the CPU speed given in the system profiler.

For Strata3D CX, the emulated G3 performs about [b:82596d5991]16X[/b:82596d5991] slower than the CPU given in the system profiler.

For Pi calculator, the emulated G3 performs about [b:82596d5991]16X[/b:82596d5991] slower than the CPU given in the system profiler.

For SpeedRun, the emulated G3 performs about [b:82596d5991]5X[/b:82596d5991] slower than the CPU given in the system profiler.

Both Photoshop and SpeedRun place my PearPC system somewhere around a G3 200MHz to 300MHz.

The tests where I see a 16X slowdown places my emulated G3 in the realm of a 66.25MHz CPU. That's interesting, because 67MHz is the bus speed reported by the system profiler. [i:82596d5991]I wonder if that is a coincidence[/i:82596d5991].

Bryce5's slowdown, off from the 5X slowdown of Photoshop and SpeedRun might be attributable to Bryce's relatively poor implimentation for the mac???? Just a guess.

Some of the slowdown times are bit different from what I reported earlier in this thread because I have had the opportunity to run more benchmark trials as well as review the benchmark results of many more [i:82596d5991]actual[/i:82596d5991] macs.

[Guest]

Watching all specs i understood that pearpc have speed improvements with last videocards how ATI RADEON XT and other and it works at 1/5 of the host speed.