boring: YU12 dvdripping bottlenecks on p3 machine

andy

Hello everyone.

I am back to my old trusted p3 pc and it brought me that old quot;is it supposed to be _that_ slow?quot; question again.

I decided to benchmark that it with several clips and see

Here is the setup:

intel P3-800something
BX@120mhz mainboard and PC133 ram.
win2k sp2.

dvd2avi 1.84
avisynth 2.5
vdumbod

The script just has mpeg2dec3 line with PP disabled. No crop/resize filters.

I seem to be getting 7-10 fps, depending on source vob file. changing codecs to huffy or xvid with carious settings didnt help/hurt speed noticeably.

But using a same clip encoded to 720x xvid at quant2 gives me about 20 fps.

Is mpeg2 decoding acting as a bottleneck or it's me missing something here? Can you guys do a small check with similar avs script and post an approx speed value and hardware setup here?

Cosidering this it seems very slow.

What's your script?
Try using avs2avi using the XVID null-filter, and paste what the screen says after encoding.

Those speeds sound about right depending on your script. At 640 resolution my dual PIII 850 system  gives me about 20FPS but I've never really done any serious benchmarking.

I have a one-CPU P3-850, 192 RAM running XP SP1.

And I get the same average speed (6-10 fps) depending on script.
I'm using Avisynth 2.07 with trbarry's mpeg2dec2, and avs2avi.

I've noticed that when script syntax errors occur it encodes @20 fps, but I didn't check the resolution.

A script with SimpleResize and mpeg2dec2 only encodes @11.3 fps.

I'm curious about the bottlenecks here will make some more tests tomorrow night.

Yep, i had very low fps on my p3-800 too. Around 7 i think...

avs2avi with null xvid:

Source: * Filename: gg.avs * Bit depth: 12 * FOURCC: YV12 * Frames: 154 * Resolution: 720x480
Compressor: * Name: XviD MPEG-4 Codec * FOURCC: XVID
Destination: * Filename: gg.avi * Frames: 154 * Key frames: 154 * Size: 0.000MB
Pass 1/1, compression time 00:00:03.114 (49.45fps).

Hmm... It seems i was wrong and avisynth itself gives me excellent speed (thank you sh0dan for showing me a way to quickly benchmark avs stuff).

Now a quick check on codec speed (having fun with benchmarking): Code:
xvid ME:0 quant2 h263: 18.85fps - 8.368M file
xvid ME:1 quant2 h263: 14.08fps - 6.179M file
xvid ME:2 quant2 h263: 13.98fps - 6.179M file
xvid ME:3 quant2 h263: 13.87fps - 6.179M file
xvid ME:4 quant2 h263: 13.03fps - 4.825M file
xvid ME:5 quant2 h263: 12.32fps - 4.767M file
xvid ME:6 quant2 h263:  9.30fps - 4.711M file
There. Divx3 gave me about similar speeds so i am sure the xvid build i am using isnt broken (XviD-09122002-1). All other tests were done with ME=6, as these are the ones i use for dvdripping. I tried to resize it to a resolutions of 640x352(typical 1:1.85 encode) and 640x272 (a typical 1:2.35 encode) and see if it would speed up things by much.Code:
640x352 bilinear: 12.26fps 2.186M
640x352 lanczos3: 10.81fps 2.980M
640x272 bilinear: 14.51fps 1.736M
640x272 lanczos3: 12.46fps 2.352M
The source quot;FILM efffectquot; type noise, hence the huge difference in a filesize.

As we see the resolution has significate impact on perfomance (Surprise.), but still, that's nowhere near 31fps that quot;mixanobios
quot; guy reported.[edit] Doh, i didnt notice he was using null-pass for xvid there. 28.32 fps this time, with sharp bicubic resize to 640x272. I guess, that's just about as fast as it gets.

P.S. I never thought resizing affects speeds that much, the drop from ~50fps down to ~29 is evil :-(