TMPGEnc 2.0 & Pentium 4

[jwhitlow]

I am very curious if any of the P4 users (I have 1Ghz PIII) see any benefits from the SSE-2 optimization added to the version 2 of TMPGEnc. If you do, how significant would you say it is?
I would like to ask any willing P4 users if they could encode a video with & then without the SSE-2 option turned on (Option > Environmental Setting... > CPU > Enable SSE-2) & post the times in reply to this message.

[caish5]

I'll get right on to that. I'm kinda interested myself. The new TMPGenc does seem faster.

[caish5]

Ok here's the results....
I captured one minute of PAL 352*288 video with virtual dub with huffyuv compression and uncompressed audio.
Went to TMPGenc and loaded the standard videocd pal template. Quality set to normal. Times Were.....
Without SSE-2 - 2:00
With SSE-2 - 1:49

My computer is a 1.3ghz pentium 4 with 128mb RDRAM.

[jwhitlow]

Not too bad! 11 seconds off of 2 minutes (120 seconds) is a nice little speed boost (~10%).

Thanks for the quick response!

[PolarBear]

That's interesting. Any chance you could run a test with 12j as a "control"?

Matt

[caish5]

Same file with 12j
3:46

I guess 2.00 really is FFAASSTT!!!!

[BennyT]

2.0 is quite fast, about 2X as fast on the highest quality video setting, then 12E.

For a SVCD High Motion video, I frame served with VDub using the following filters (courtesy of LABDV.com):

deinterlace Discard field 1
smart resize NTSC: 480x480, Precise bicubic
Smart smoother D=5, T=25
Color correction histogram equilize Str=34, Inten=224

Along with TEMPGEnc's noise reduction filter on the default setting.

Using same with CCE2.5 I did not see much diferrence between to 2 videos, except CCE finished about 40% quicker. I have to say TMPGEnc is getting real close and for the price it is the best encoder out there right now in my opinion.

Regards,

BennyT


Forgot to mention, I was running TEMPEng with SSE2 enabled for my P4 1.7GHz system

<font size=-1>[ This Message was edited by: BennyT on 2001-10-03 17:52:26 ]</font>

[PolarBear]

I made a test as well for MPEG2. P4/1.4/256MB/40GB.

Using a 1 minute, 352x576 video, used svcd pal template for v 12j, 2.0 and cce 2.5 just for good measure. No filters, just 8 MM capture.

12I w/SSE2: 2min 50sec
12I w/o SSE2: 3:00
2.0 w/SSE2: 2:45
2.0 w/o SSE2: 3:00
CCE: 0:40

So CCE still kills, but its limitations compared to TMPGEnc are its falling. Of course that's why there's vfapi!

Matt

[ktnwin]

After reading this post, I made a try on my system

Athlon 1.2GHz, 256MB

I use the exact same clip (25min video).
Conversion time using standard VCD template
12F: 1 hour 01 min
2.0: 1 hour 04 min

I chose 12F since it was noticeably faster than 12a.

[Jeomite]

There are rumors floating around that TMPGEnc beta 12a (or just plain beta 12) is faster than 12j or whatever. Can someone squash or prove these rumors by doing some prelimary tests? Preferably with atleast 2.00, beta 12, beta 12a, and beta 12j... ?

(I like the speedy CCE best =)

[PolarBear]

Per my earlier post on this thread, here's the time for 12a:

12a: 4:30

No SSE2 option on that ver. Don't have original v12.

Matt

[KingJohn]

Mmmm... Thought I would have a try with a tuff DivX

The file is a
352 x 240, 24 Bits, 84491 Frames, 20.000 Frames/Sec, 62 KB/Sec, MS-MPEG4 V3... These MS mpeg 4, V3 are highly compressed, not the normal run of the mill DivX files. With part 1 and 2, it makes 3 VCD's

Previously On my Pentium 3, 450 I converted the 264Mb file to a PAL VCD and it took "2 hours 15 minuits." (Part 1)


I have since built a Pentium 4, 1.6 and just tried an experiment on the same AVI

With SSE-2 turned on it says 1 hour 06 minuets
With SSE-2 turned off it says 1 hour 10 minuets
With SSE and SSE-2 turned off it says 1 hour 27 minuets

Not much in it for the first 2

But a huge improvement over my P3

Bing, Bong, Bing Bing...