Hi all.
First of all, let me say that, as much as I'd love to, I don't get to stay abreast of every thing that is discussed here at HA 24/7.
So, if someone else has covered this, my apologies.

I decided to run a little test between 3.96, 3.96.1 and 3.97b1.
Using a 44.1kHz 16bit stereo wav file of 59min 53sec duration, I encoded 3 versions, all at CBR 320 using All2Lame as the frontend.

3.96
Time taken: 3:29
File size on completion: 140,244 KB

3.96.1
Time taken: 3:27
File size on completion: 140,365 KB

3.97b1
Time taken: 4:07
File size on completion: 140,365 KB

I thought 3.97b1 was supposed to be quicker, but obviously not. 4:07 represents an approximate 20% longer encode time over the 3:27 taken by 3.96.1.
I haven't tested at lower bitrates yet, and I'm not making any claims on audio quality. I was merely interested in how the different builds compared for CPU time.
Discuss as you will.

i think the speed up was with VBR

what command line settings were you using? the biggest advantage that 3.97b1 has is that it's --vbr-new (which is markedly faster) is considered to be better quality that the standard, --vbr-old. so the better test would be 3.97b1 w/ --vbr-new vs. 3.96.1 (default setting).

What settings did you use to encode?

320cbr is not a good parameter to make a comparision.
Just compare them using V0 / V2 with the new vbr and the old one.

Here are the results from my tests using two computers:
Lame 3.97b1 at -V 4 --vbr-new:
One song that is 17munites 35 seconds long
Computer 1 (Pentium 4 2GHz non-HT processor, 256MB RAM, 7200RPM 80GB HD)
Encoding speed = ~11x
Time = 1 minute 35 seconds
Computer 2 (Pentium M 2GHz, 512MD RAM, 5400RPM 80GB HD):
Encoding speed = ~20x
Time = 45 seconds

Lame 3.96.1:
Same song
Computer 1:
Encoding speed = ~10x
Time = 1 minute 45 seconds
Computer 2:
Encoding speed = ~14x
Time = 1 minute

I didn't bother with Lame 3.90.3. I can tell you that on Computer 1, Lame 3.90.3 would encode at an average speed of 4x. These are my non-technical results. Family Guy is on and it is hard trying to pay attention.

Edit: Oops. I should have mentioned that LAme 3.96.1 was utilized to encode the two songs using the -V 4 command only, no --vbr-new.

VCSkier, I was using "-b320 -ms"
Following Kornchild's lead, I redid the 3 encodes, using "-V4--vbr-new"
The results were, as Bubka suggested, different:

3.96
Time: 4:48
Filesize: 68,996 KB

3.96.1
Time: 4:39
Filesize: 69,021 KB

3.97b1
Time: 4:25
Filesize: 57,653 KB

cool. its good to know that 3.97b1 is even faster than older versions w/ --vbr-new. even more revealing would be to compare 3.96.1's highest quality setting at that bitrate range (-V4) to 3.97b1's highest quality setting at that bitrate range (-V4 --vbr-new).

Who says that -V4 is better than -V4 --vbr-new in 3.96.1? There hasn't been any test about this AFAIK.

a bigger question would be "why has cbr encoding speed declined since 3.96.1?"

that doesn't make sense to me... unless it is ICL vs MSVC (compiler settings)


later

You kids worry too much about too little.