LAME benchmarks on Core 2 Duo CPUs? (3.97 -V5 --vbr-new), Also, C2D benchmarks of 3.98 beta 1? Options

[ShowsOn]

I'm wondering roughly how fast LAME 3.97 runs using -V presets and --vbr-new on Intel Core 2 Duo CPUs?

I currently have an Athlon X2 4600 CPU (2 X 2.4 GHz, 512 cache). It encodes using LAME 3.97 (-V5 --vbr-new) at around 20 X real time (per core). I'm wondering roughly which Intel Core 2 Duo CPU would be significantly faster (say at least 15 - 20%) which would mean at least 25 x real time using 3.97 -V5 --vbr-new?

Unfortunatley all the top PC hardware review sites don't use the -V presets! They often just encode to CBR, which is annoying, because surely VBR is a better test of CPU power than CBR?

Here are results I got encoding Under the Bridge by the Red Hot Chili Peppers. First with 3.97 then with 3.98 beta 1. Can anyone with a Core 2 Duo CPU tell me how fast the command line encoder is on their PC? I got my encoder compiles from Rarewares.org

C:\TOOLS\AUDIO\LAME>lame -V5 --vbr-new underthebridge.wav
LAME 3.97 32bits (http://www.mp3dev.org/)
CPU features: MMX (ASM used), 3DNow! (ASM used), SSE (ASM used), SSE2
Using polyphase lowpass filter, transition band: 15826 Hz - 16360 Hz
Encoding underthebridge.wav to underthebridge.wav.mp3
Encoding as 44.1 kHz VBR(q=5) j-stereo MPEG-1 Layer III (ca. 11.9x) qval=3
Frame | CPU time/estim | REAL time/estim | play/CPU | ETA
10118/10118 (100%)| 0:13/ 0:13| 0:13/ 0:13| 19.647x| 0:00

Curiously, for 3.98 beta 1 I only get ~16 X. Perhaps it was compiled differently?

C:\TOOLS\AUDIO\LAME398b1>lame -V5 underthebridge.wav
LAME 3.98 (beta 1, May 16 2007) 32bits (http://www.mp3dev.org/)
CPU features: MMX (ASM used), 3DNow! (ASM used), SSE (ASM used), SSE2
Using polyphase lowpass filter, transition band: 15826 Hz - 16360 Hz
Encoding underthebridge.wav to underthebridge.wav.mp3
Encoding as 44.1 kHz VBR(q=5) j-stereo MPEG-1 Layer III (ca. 11.9x) qval=0
Frame | CPU time/estim | REAL time/estim | play/CPU | ETA
10118/10118 (100%)| 0:16/ 0:16| 0:16/ 0:16| 15.989x| 0:00

[kanak]

Hi

I've got a Core 2 Duo @ 2GHz on my laptop with a 5400 rpm hard drive and 1 gig ram. When i use foobar, i usually get around 36X (for 2 songs so roughly 18X per thread). Note that this is:

a. on a laptop
b. on a 5400 rpm hard drive
c. on a 1 gig @DDR-533

bumping it up to 7200 rpm and DDR@667 should yield better results.

[ShowsOn]

Hi

I've got a Core 2 Duo @ 2GHz on my laptop with a 5400 rpm hard drive and 1 gig ram. When i use foobar, i usually get around 36X (for 2 songs so roughly 18X per thread). Note that this is:

a. on a laptop
b. on a 5400 rpm hard drive
c. on a 1 gig @DDR-533

bumping it up to 7200 rpm and DDR@667 should yield better results.

Thanks for this info. Are their major differences between laptop and desktop core 2 duos? I realise it is a different core, but are there certain sacrifices made in favour of power efficiency that means you can't directly compare same MHz C2Ds across both platforms?

Anyone with a desktop Core 2 Duo?

[le_canz]

I ran the test

C2D E6600 @ 420 x 8 = 3.36 GHz

I used the same song, ripped with EAC from original CD.

With Lame 3.97 :

C:\TOOLS\AUDIO\LAME>lame -V5 --vbr-new under_the_bridge.wav
LAME 3.97 32bits (http://www.mp3dev.org/)
CPU features: MMX (ASM used), 3DNow! (ASM used), SSE (ASM used), SSE2
Using polyphase lowpass filter, transition band: 15826 Hz - 16360 Hz
Encoding under_the_bridge.wav to under_the_bridge.wav.mp3
Encoding as 44.1 kHz VBR(q=5) j-stereo MPEG-1 Layer III (ca. 11.9x) qval=3
Frame | CPU time/estim | REAL time/estim | play/CPU | ETA
10119/10119 (100%)| 0:07/ 0:07| 0:07/ 0:07| 35.842x| 0:00

And then with 3.98 b1 :

C:\TOOLS\AUDIO\LAME398b1>lame -V5 under_the_bridge.wav
LAME 3.98 (beta 1, May 16 2007) 32bits (http://www.mp3dev.org/)
CPU features: MMX (ASM used), 3DNow! (ASM used), SSE (ASM used), SSE2
Using polyphase lowpass filter, transition band: 15826 Hz - 16360 Hz
Encoding under_the_bridge.wav to under_the_bridge.wav.mp3
Encoding as 44.1 kHz VBR(q=5) j-stereo MPEG-1 Layer III (ca. 11.9x) qval=0
Frame | CPU time/estim | REAL time/estim | play/CPU | ETA
10119/10119 (100%)| 0:07/ 0:07| 0:07/ 0:07| 35.025x| 0:00

Almost same speed, 3.98 just a little bit slower.

Both compiles are made with ICL 9.1, it's strange that 3.98 is so slower on your comp.

Edit : Only the cache size changes between portable and desktop versions of C2D.

This post has been edited by le_canz: May 21 2007, 10:24

[tycho]

My laptop has the same specs as kanak, and I get about the same results, as expected. I don't think a faster HD would improve results much, if at all. Note it is single channel ram (1 x 1Gb).

My desktop Pentim D840 3.20GHz, on Win64, is surprisingly also slower than my laptop (get about 32x with foobar). It has 4 x 1Gb 2.66 ram, and I suppose Dual channel is therefore enabled.

[gib]

b. on a 5400 rpm hard drive
c. on a 1 gig @DDR-533

bumping it up to 7200 rpm and DDR@667 should yield better results.

MP3 encoding is CPU bound even on very fast systems, so neither RAM nor disk speed will have much of an affect on things.

[Alexxander]

On my E6600 not overclocked (9*266 MHz) and WinXP SP2:

CODE

F:\Progs\lame3.97>lame -V5 --vbr-new pvd.wav
LAME 3.97 32bits (http://www.mp3dev.org/)
CPU features: MMX (ASM used), SSE (ASM used), SSE2
Using polyphase lowpass filter, transition band: 15826 Hz - 16360 Hz
Encoding pvd.wav to pvd.wav.mp3
Encoding as 44.1 kHz VBR(q=5) j-stereo MPEG-1 Layer III (ca. 11.9x) qval=3
    Frame          |  CPU time/estim | REAL time/estim | play/CPU |    ETA
23935/23935 (100%)|    0:23/    0:23|    0:23/    0:23|   26.378x|    0:00

Overclocking by increasing the FSB speed increases the encoding speed proportionally.

[ShowsOn]

On my E6600 not overclocked (9*266 MHz) and WinXP SP2:

CODE

F:\Progs\lame3.97>lame -V5 --vbr-new pvd.wav
LAME 3.97 32bits (http://www.mp3dev.org/)
CPU features: MMX (ASM used), SSE (ASM used), SSE2
Using polyphase lowpass filter, transition band: 15826 Hz - 16360 Hz
Encoding pvd.wav to pvd.wav.mp3
Encoding as 44.1 kHz VBR(q=5) j-stereo MPEG-1 Layer III (ca. 11.9x) qval=3
    Frame          |  CPU time/estim | REAL time/estim | play/CPU |    ETA
23935/23935 (100%)|    0:23/    0:23|    0:23/    0:23|   26.378x|    0:00

Overclocking by increasing the FSB speed increases the encoding speed proportionally.

Thanks for that. I used Le Canz data to estimate that the E6600 at stock speed would be about 25X, but it looks like it is slightly better than my prediction. Thanks!

[R.A.F.]

I ran the test
C2D E6600 @ 420 x 8 = 3.36 GHz

Doesn't the E6600 have a maximum multiplier of 9, instead of 8?

[le_canz]

Doesn't the E6600 have a maximum multiplier of 9, instead of 8?

You're right, I use 8 for OC.

With my motherboard (Asus P5B Deluxe), FSB freq from +/- 370 to 399 MHz makes the system unstable (because of chipset's memory strap), and the CPU cannot handle 400 x 9.

So 420 x 8 is a good compromise

This post has been edited by le_canz: May 21 2007, 14:59

[xmixahlx]

it's about 8% slower here on linux, too.

perhaps the new fft/3dnow/assembler stuff is the culprit.


later

This post has been edited by xmixahlx: May 21 2007, 16:41