Benchmarking Dolphin

[JADS]

(04-19-2010, 11:45 AM)NaturalViolence Wrote:

Quote:Yeah, that was my laptop. What are you getting in that area with your rig?

Back in the day I was getting 20 fps in that area with build 5016. Unfortunately my rig is currently setup as a linux server and must remain this way for another month or so, one of the many reasons why I wanted to postpone the benchmark. I don't remember the exact framerate but with framelimit off I remember JAD getting over 60 fps, I might not be right about this but it was definitely insanely high. JAD, care to post your fps in faron woods (remember to use the listed settings)?

60 FPS in some areas in Faron Woods, the benchmark is a bit lower

OC on, 3.5 GHz


Stock Settings


Things to keep in mind:

1. Turbo automatically overclocked the processor to a cool 3.2 Ghz on stock settings. Measured using the Intel Turbo Boost Technology desktop widget for Windows 7.
2. Disabling or enabling Hypertreading did not affect the speed in anyway.
3. Even though dolphin can only use two threads, dropping to two threads shows me framerates below 30 FPS on stock speeds. Enabling a third thread showed the same speeds as under stock settings.
4. Overclocking my ATI Radeon 5850 to 5870 speeds did not increase FPS in anyway. GPU usage, as measured by MSI Afterburner, was in the high 20ths.
   

[NaturalViolence]

Try it with framelimit set to auto and see if you still get 30 fps.

[JADS]

(04-20-2010, 05:50 AM)NaturalViolence Wrote: Try it with framelimit set to auto and see if you still get 30 fps.

Yeah, I still get 30FPS with OC on or on stock settings.

[ExtremeYoshiFan]

I may benchmark with an Athlon II X2 240 Regor, and I have pretty much barebone everything else for requirements(read my specs in sig), and I feel you also need a Gamecube game benchmark as well, That would be a help to most of those who run GC games. And do you want GC Twilight Princess or Wii, I can use NTSC or PAL, but I need to know which ones you want, also, what is the IRC you are using for this project, I may go join it.

[NaturalViolence]

JADS did some benchmarking for me today. Still analyzing the results to try and figure out where his insane speedup is coming from.

Normal settings, 4 cores 8 threads: 42 fps

2 cores disabled (ht off): 25 fps

4 cores with lock threads to cores (HT on and ht threads are being used): 33 fps

Audio throttle on: 42 fps (no difference whatsoever)

Two cores is 25 frames, 3 cores is the frames that you see in the screenshots, four cores is no change.

100% CPU load under two threads, 50 ish under four cores and 25 with HT and 4 cores

1920x1080P 4xSSA: 38 FPS GPU usage at around 75%

As you can see dolphin is balancing the load perfectly between 2, 4, and 8 threads and is fully utilizing 2 threads. However it seems that a large speedup occurs (50% speedup) if a third thread is availible, this probably has something to do with the load balancing but it is odd that this speedup does not occur on phenom II or core 2 quad cpus that also have a third and fourth thread available. I'm going to talk to a dev either this afternoon or next to try and get some more research to use.

Quote:I may benchmark with an Athlon II X2 240 Regor, and I have pretty much barebone everything else for requirements(read my specs in sig), and I feel you also need a Gamecube game benchmark as well, That would be a help to most of those who run GC games. And do you want GC Twilight Princess or Wii, I can use NTSC or PAL, but I need to know which ones you want, also, what is the IRC you are using for this project, I may go join it.

We'll take all the help we can get. You need the GC NTSC version of twilight princess. And here is the irc info:

irc.unvanquished.net
#dolphinbenchmark
Port 6667

[Starscream]

Fighting the first boss in the underwater dungeon seems to run very well for my PC. This entire dungeon runs extremely well. Of course, this is with an older version that works well for me.

[NaturalViolence]

We can't be using older versions or a spot that runs well. I'm still working on understanding why JADS system is handling this so well, simple put his results don't make sense, his framerate is way higher than it should be with that setup. I talked to one of the devs and he's making a test to see how fast a few mutexes are between threads to compare them between CPUs. It shouldn't take more than a day or 2, truth be told this is exactly what I hoped would come of this project, this unexplained performance gain is exactly the kind of discovery I was hoping to find.

[Starscream]

Obviously I would not want to test a spot in a game that runs well for me seeing how you guys have better PC's than I do. If I can get good frame rates, there is no doubt you would also be running the game in that spot as good or better. I was just posting a comment and a picture that is unrelated to the actual test that shows my PC can get full speed in areas.

[NaturalViolence]

Ah, ok, the mutex test is almost done.

[Xtreme2damax]

Saving myself from typing, captures from CPU-Z and GPU-Z..