How fast is PearPC: Benchmarks

[Ribtorus]

Photoshop 7.01 is running reasonably reliably, so I ran PS7Bench, a Photoshop benchmarking action script that creates a 50MB file and performs some typical and not so typical filtres and functions. I ran 12 of the 21 PS7bench scripts because it was just taking too long, and the pattern was well established anyway.

Using the following build: ppc-g4-athlonxp-sdl-20040828-1800-tekmaven
with Panther on a dual opteron 246.

The System Profiler shows my setup as a 1.06GHz G4, so I compared my results to an actual 1GHz G4.

Times are in seconds.

PearPC/G4 1000

Rotate 90, [b:72e42ab7eb]7.4/1.7[/b:72e42ab7eb]

Rotate 9 , [b:72e42ab7eb]94.2/8.2[/b:72e42ab7eb]

Rotate 0.9, [b:72e42ab7eb]90.8/7.6[/b:72e42ab7eb]

gaussian blur 1, [b:72e42ab7eb]38.2/3.8[/b:72e42ab7eb]

gaussian blur 3.7, [b:72e42ab7eb]75.8/7.6[/b:72e42ab7eb]

gaussian blur 85, [b:72e42ab7eb]116.0/9.6[/b:72e42ab7eb]

unsharp 50-1-0, [b:72e42ab7eb]41.2/6.2[/b:72e42ab7eb]

unsharp 50-3-7, [b:72e42ab7eb]79.0/9.8[/b:72e42ab7eb]

unsharp 50-10-5, [b:72e42ab7eb]109/10.2[/b:72e42ab7eb]

despeckle, [b:72e42ab7eb]95.4/4.3[/b:72e42ab7eb]

RGB>CMYK, [b:72e42ab7eb]57.5/10.4[/b:72e42ab7eb]

reduce size 60%, [b:72e42ab7eb]50.6/3.3[/b:72e42ab7eb]

In this test, the emulated G4 is as little as 4X slower, and as much as 22X slower than an actual 1GHz G4.

The average slowdown is [b:72e42ab7eb]10.5X[/b:72e42ab7eb]

The slowdown of PS7Bench vs the [i:72e42ab7eb]host[/i:72e42ab7eb] machine (dual Opteron)ranges from 24X to 130X, with an average slowdown of [b:72e42ab7eb]68X[/b:72e42ab7eb].

I think this small test is consistent with other speed claims.

[prasys]

Good job


Well , Thanks for taking your time and posting your benchmark result

For those who wish to benchmark thier PearPC Emulated OSX Computer , just download XBench from www.versiontracker.com

[Guest]

Thanks for the link.

I ran XBench and compared my result to a 1GHz G4.

My overall score is 5.70

Where my performance really lags, and brings the score way down is in;

CPU Test, 99.67 vs 1.59
Floating point, 99.07 vs 2.47
Altivec basic, 98.38 vs 0.73
Vector Library FFT, 101.63 vs 0.77

The other system and processor tests showed my emulated G4 running mostly better than 50% to 80% the speed of the actual G4

In the hard drive tests, it got interesting;

Disk test, 99.79 vs 53.84
Sequential, 99.22 vs 30.17
U/C write, 98.4 vs 11
U/C write, 99.66 vs 61.02
U/C Read, 100.12 vs 47.17
U/C Read, 98.72 vs 243.33
Random , 100.37 vs 250.09
U/C Write, 99.71 vs 350.47
U/C Write, 103.28 vs 99.07
U/C Read, 99.07 vs 1007.12 :shock:
U.C Read, 99.52 vs 486.75

Either there may be a flaw in the timing of the disk tests, or the Ultra320 SCSI drive I use makes a big difference.

[Guru3]

[quote="Anonymous":31035a71b2]
U/C Read, 99.07 vs 1007.12 :shock:
[/quote:31035a71b2]Yeah... There might still be some timing flaws lurking in the corners...

[Ribtorus]

I re-ran Xbench four times and each time it was in the high 900's and low 1000s

Each run showed pretty consistent results.

[Ribtorus]

Here's the results of Xbench, it shows my emulated G4 in Red compared to an eMac G4 with 512MB ram, in Blue.

Host: Dual Opteron 246, 1GB ram, Windows XP
PearPC build: ppc-g4-athlonxp-sdl-20040901-1200-tekmaven

Guest: 1GHz G4 with 512MB ram, Panther 10.3.5

[Ribtorus]

I've run XBench a few times now, and I have noticed something interesting.

When I set the cpu to the recommended setting for emulating a G4, the XBench scores [i:4da205d2c8]drop[/i:4da205d2c8] by 9%.

My system profiler still says it's a 1.06GHz G4, but I get a bit better performance by leaving it to the default value.

Placing the mouse speed at either 19 or 270 made no real difference at all between runs.

[prasys]

Yeah , try to run quartz test and xbench will give a good report on the video card ,..it rocks , PearPC has the one of the best PCI Card I have ever seen with 0 MB memory onboard with 12Mhz PCI Bus-Speed

[Guru3]

Well, I guess that got Disk I/O nailed .

[Ribtorus]

I ran the complete PS7bench script of 21 actions on a 50MB tiff file, using Photoshop 7.01 .

This particular benchmark runs a series of 21 photoshop actions and filters on a file, and the operator times the results using photoshop's built-in timer, or a good stopwatch. I used both timing nethods to ensure there was no clock/timing isues with pearPC. There were none; Photoshop's built-in clock and the stopwatch agreed with eachother.

I took the total time to complete all the filters as the benchmark.

It's bad, but not as bad as I thought it would be. Clearly there is some accelleration going on in some of the filtres.

The Host is a Dual opteron 246 with 1GB ram.

The guest is the latest "ppc_trunk_sdl_nightly" set to use 896MB ram with the system Profiler showing a 1.06GHz G4 running Panther 10.3.5

The completion time for the Host machine with Windows is 89 seconds and for the guest G4 is 2097 seconds, giving a slowdown of [b:c7e44dd970]23.6X[/b:c7e44dd970]

Comparing the guest machine to some comparably clocked macs running the same benchmark gives the following slowdowns:

Dual 1GHz G4, a slowdown of [b:c7e44dd970]8.6X[/b:c7e44dd970]

Quicksilver 733, a slowdown of [b:c7e44dd970]4.5X[/b:c7e44dd970]

Another Dual 1GHz G4, a slowdown of [b:c7e44dd970]11.4X[/b:c7e44dd970]

Another G4 733, a slowdown of [b:c7e44dd970]5.6X[/b:c7e44dd970]

G4 800, a slowdown of [b:c7e44dd970]5.8X[/b:c7e44dd970]

G3 700, a slowdown of [b:c7e44dd970]4.4X[/b:c7e44dd970]

I could not find a complete PS7Bench score for a 1GHz G4 unfortunately.

Overall performance improved when I completed the full PS7Bench test, instead of just the first twelve actions, probably because of optimisations with Altivec. I also had significant improvements after I changed from one build to another, and re-installed OS X on its own hard drive.