0
0

Hey guys, hey Brett!

I am working with fmod quite a while but just now I had the time to do some benchmarks for myself and for a upcoming paper I am writing right now. Now I have a strange behaviour, I cannot explain myself.
Probably you can ?

I added fmod to a scenegraph software and wrote a small benchmark that adds 3D-Sounds (simple white noise) to a scene with 64 spheres. I wanted to see in numbers how much CPU is used and how the FPS changes with the number of 3D Sounds. I have newest NVidea driver and the polygon count in the scene is only some thousand, no deal at all.

This is what I have:
PC1: Dual 1Gig, GF2MX, 1024MB Ram, Win2k/RedHat, onBoard Via
PC2: Dual 1Gig, GF3, 512 MB Ram, Win2k, Terratec Sixpack

Now I get those results on the first PC:

[code:3rzfhlz4] Linux PC1 Build DBG
Sound FMOD-CPU FPS Total Frames in 20 Sec.
1 0.27 185.29 3555
8 0.767 178.80 3431
16 1.64 177.5 3402
24 2.35 167.87 3217
32 3.13 165.4 3167
40 2.93 159.0 3041
48 3.12 155 2974
56 3.36 152.7 2924
64 3.17 148.2 2845[/code:3rzfhlz4]

[code:3rzfhlz4] Linux PC1 Build OPT
Sound FMOD-CPU FPS Total Frames in 20 Sec.
1 0.42 317 6094
8 0.876 317.55 6091
16 1.91 317.61 6095
24 2.72 317.6 6090
32 3.46 318 6098
40 3.55 318 6081
48 3.50 318 6069
56 3.25 318.32 6077
64 3.47 319 6061[/code:3rzfhlz4]

[code:3rzfhlz4] Win2k PC1 Build OPT
Sound FMOD-CPU FPS Total Frames in 20 Sec.
0 0 272.74 5355
1 .38 252 4950
4 0.46 249 4907
8 0.71 247 4873
16 1.80 246.42 4851
32 2.79 239.95 4733
48 2.64 235.52 4639
56 8.65 232 4569
57 2.35 229 4548
64 2.69 230 4525[/code:3rzfhlz4]
See that strange behavior at 56 Sounds? I tested it 4 times!

Ok Now the other PC:
[code:3rzfhlz4] Win2k PC2 Build OPT
Sound FMOD-CPU FPS Total Frames in 20 Sec.
0 0 278.85 5473
16 0.541 202.965 3932
32 0.548 165.922 3326
48 0.5815 152.769 3012
64 0.587 127.128 2482
[/code:3rzfhlz4]

[code:3rzfhlz4] Win2k PC2 Build DBG
Sound FMOD-CPU FPS Total Frames in 20 Sec.
0 0 79.16 1555
16 0.454 62.143 1220
32 0.444 54.91 1076
48 0.441 49.55 965
64 0.493 46.76 916[/code:3rzfhlz4]

Now this is more or less what I expected.
On PC1 the CPU load is pretty linear to the number of 32 3D-Sounds,
then its more or less constant. In DBG also the FPS drops linear.

Then on PC2 the CPU load is lower and constant, I think thats because of the better soundboard. But then again the FPS are much lower and drop more than on PC1, even with a GF3.

Things I dont understand:
on PC1 in Linux/OPT the FPS is so high and constant, it even gains a frame when I have 64 3D Sounds. Why the **** is that?
I asume PC1 totally lacks hardware support so, the CPU Load between the OSs should be similar, actually Win acts a little bit better. But still the FPS tells me Linux runs faster even with a bit more load.

I know both are Dual machines, but how and what can this effect the FSOUND_GetCPUUsage()? Any comment is highly welcome!

Brett, could you give me some insight details (probably per mail) if there is a difference between Linux and Win threading design or 3D sound calculation in FMOD implementation?

I couldnt measure a real difference between different output modes, eg. STEREO or DOLBYDIGITAL. Is that correct, that fmod needs similar calculations even if the physical output channels (2/6) is 3 times higher?

Really guys, I am grateful for some help because deadline for submission for that paper is in a few days.

Thanks!

MVD

  • You must to post comments
Showing 0 results
Your Answer

Please first to submit.