How fast is PearPC: Benchmarks

[Ribtorus]

I applied Shadowkiller and Tinkertool to shut down font smoothing.

I don't think Shadowkiller works on my OSX install, because the shadows are still there.

Tinkertool [i:284026fd5a]did[/i:284026fd5a] eliminate font smoothing though.

None of my benchmarks showed any improvement whatsoever. I would have thought that LetWindowsBloom and perhaps PiCalculator (where I enable the video refresh setting) would have shown some improvement. But they didn't. Neither did SpeedX, SpeedRun or Xbench show any improvement at all.

All that I saw was that my screen fonts looked crappier.

Animated features in the Finder were turned off as well, but it didn't result in windows displaying any faster, it just meant they popped up without the cool effect, but with the amount of time remaining the same.

What's up with that?

[Guest]

I have tried the speedup tip of using the task manager to set a higher CPU priority for PPC.exe, but unfortunately I saw no measurable speed improvement at all for XBench, SpeedX, letWindowsBloom, Picalculator or PhotoshopCS.

[akamai]

definite speed improvement: RAMDisks... create a partition in your system ram, mount it, copy (dont cut) pearpc to your RAMDisk, and run it - be sure to edit the config. Everything in RAM will be LOST after reboot/power failure, so be careful...

btw: some benchmarks done on my ibook testing a ramdisk vs. my normal hdd:

http://tequiladrinker.homeunix.org/RAMDisk.pdf
http://tequiladrinker.homeunix.org/HDD.pdf

have fun

[Ribtorus]

Do you mean copy only PearPC to a small ramdisk, or copy the entire PearPC [i:221ca884af]and[/i:221ca884af] OSX drive to ramdisk? There's no way I can copy over my entire 6GB OSX disk to a ramdisk; I don't have enough ram for that.

But I have run PearPC/OSX off an IDE drive, a SCSI-2 Hard drive and an Unltra SCSI 320 drive with no appreciable change in performance. My experience so far has been that PearPC/OSX performance is not significantly related to hard drive performance; the bottleneck being the CPU emulation first and foremost.

[Guest]

Just a couple notes on disparancies in benchmarking. I think the benchmark "slowdown" differences we are seeing are dependant on exactly what is being benchmarked. Obviously the graphics subsystem in PearPC leaves alot to be desired, so anything benching this is likely to suffer from some rather severe slowdowns. But anything that can be easy compiled by the JIT core, then ran repeatedly will probably show little slowdown from the host system. I believe it is cases like this where we see performance on par with 3-400mhz G4 systems. In cases where there is heavy IO, and Graphics system usage, I think that number drops signifigantyl, to the 100mhz g3/4 speed range.

Just something I was thinking of...

[FunkyRider]

The graphics sub-system is tooooo slow.

[dirtydog]

Here's my results for anyone interested..

On a P4 2.66B @ 3.0, 150fsb, 1GB PC2700 @ 360, Windows XP SP2

pearpc reports a G3 1.14GHz. Using latest 0.4pre P4 build. OSX 10.3.7

[b:b5c3da1b51]SpeedX[/b:b5c3da1b51]
42.92 seconds, 233 MIPS

[b:b5c3da1b51]XBench[/b:b5c3da1b51]
Results 2.38
System Info
Xbench Version 1.1.3
System Version 10.3.7 (7S215)
Physical RAM 1024 MB
Model PowerPC (PearPC 0.4pre)
Processor PowerPC G4 @ 1.15 GHz
L1 Cache 32K (instruction), 32K (data)
Bus Frequency 58 MHz
Video Card PearPC,display
Drive Type EIN GEBUESCH!
CPU Test 0.47
GCD Loop 33.20 1.30 Mops/sec
Floating Point Basic 2.42 8.75 Mflop/sec
vecLib FFT 0.12 1.94 Mflop/sec
Floating Point Library 18.56 742.92 Kops/sec
Thread Test 23.50
Computation 50.72 684.77 Kops/sec, 4 threads
Lock Contention 15.29 191.97 Klocks/sec, 4 threads
Memory Test 41.32
System 47.34
Allocate 50.86 33.18 Kalloc/sec
Fill 47.65 379.32 MB/sec
Copy 44.00 220.02 MB/sec
Stream 36.66
Copy 42.01 307.10 MB/sec
Scale 28.63 211.31 MB/sec
Add 38.83 248.48 MB/sec
Triad 40.61 248.11 MB/sec
Quartz Graphics Test 17.39
Line 19.42 494.40 lines/sec [50% alpha]
Rectangle 13.41 943.26 rects/sec [50% alpha]
Circle 16.43 378.73 circles/sec [50% alpha]
Bezier 19.84 215.58 beziers/sec [50% alpha]
Text 19.92 324.72 chars/sec
OpenGL Graphics Test 1.68
Spinning Squares 1.68 1.18 frames/sec
User Interface Test 18.22
Elements 18.22 5.86 refresh/sec
Disk Test 23.20
Sequential 12.54
Uncached Write 4.04 1.68 MB/sec [4K blocks]
Uncached Write 60.84 24.91 MB/sec [256K blocks]
Uncached Read 21.03 3.33 MB/sec [4K blocks]
Uncached Read 135.78 54.86 MB/sec [256K blocks]
Random 154.96
Uncached Write 170.49 2.56 MB/sec [4K blocks]
Uncached Write 76.05 17.15 MB/sec [256K blocks]
Uncached Read 360.49 2.38 MB/sec [4K blocks]
Uncached Read 248.49 51.14 MB/sec [256K blocks]

[thewer]

Has anyone been able to run gprof on pearpc to see where things get slowed down? I cant get it to run in cygwin. If anyone has stats from it, i'd love to see them...

thewer ;)

[prasys]

I do not know...Try over emaculation..I heard someone managed to ...get it to work...Again , I could be wrong

[thewer]

I looked at emaculation and couldn't fund any topics to do with gprof or profiling in the forum :(
Sar lar V I guess I will keep trying...

thewer ;)