Nero AVC/H.264 Speed Test

[Toast]

Lately, I've been contemplating building a new PC. It's primary purpose would be to capture, encode, and playback HDTV. As Nero AVC/H.264 is my preferred codec at the moment, I'd like to compare it's performance with various processor(s) and setups. Therefore, I'd appreciate it if you could take a few minutes to do the following :D

Download this HDTV clip(32MB): (Mods, I hope this is ok. It's a 26 sec clip of the opening credits of the season opener of Alias)
http://sliceoftoast.net/Alias/Alias.m2v

Next, index the .m2v with DGIndex(no field operations), and create the following AVS file:

# LOAD PLUGINS
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\DGDecode.dll")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\Decomb521.dll")

# SOURCE
mpeg2source("E:\Alias\Alias.d2v")

# DECIMATE
SelectEven()
Decimate(cycle=5)

Load the AVS file into Recode2 and record how long it takes to complete a 2 pass encoding. I've included screenshots of the settings to be used at the end of this post. Please report the results as:

CPU: Pentium 4c 3.2GHz
RAM: 1GB
Encoding Time: 14min 58sec
Able to playback ok?: Yes
CPU usage during playback: 35-40%

I know it's a lot to ask so thanks to anyone who can provide some feedback. :) Also, I've uploaded the audio to accompany this clip in case anyone wants it as well(618KB):
http://sliceoftoast.net/Alias/Alias.ac3


Settings for Nero Recode2 (please disable deinterlace, crop, and resizing):
http://sliceoftoast.net/Alias/Settings1.jpg
http://sliceoftoast.net/Alias/Settings2.jpg
http://sliceoftoast.net/Alias/Settings3.jpg
http://sliceoftoast.net/Alias/Settings4.jpg
http://sliceoftoast.net/Alias/Settings5.jpg


Eagerly looking forward to your results :)

[kappa]

System configuration:

Athlon 64 3200+ 1.5GB DDR400, VIA KT800 Socket 754 Gigabyte MB
Time: 12:56@DDR400
12:32@DDR420
11:52@DDR440

AMD wins!
The DDR420 measurement was the first taken, the other
two after consecutive reboots, so I guess that is the reason
for the non-linearity.
I'm waiting for the builds that doom9 used for his comparison,
because this is really slow. It takes 6 hours for a movie.
Could you explain your choices at the expert features dialog boxes?

[babayaga]

Originally posted by kappa
I'm waiting for the builds that doom9 used for his comparison,
because this is really slow. It takes 6 hours for a movie.
Could you explain your choices at the expert features dialog boxes?
With the Doom9's settings used for the test, on his "super-strong" PC (Amd64 3500+ if I recall correctly), he reaches about 30fps per frame on the average for each pass.
This gives about 3h 20mn to encode a 2h movie.

The differences with your configuration are :
- he uses only 2 references instead of 5
- he uses "Best" quality level instead of "Extra"
- his PC is slightly more powerfull
- the tested codec is about 40% faster on the 1st pass
- the tested codec is about 20% faster on the 2nd pass

Besides, as Doom9 wrote it in the 1st page of his test, the 1st pass is much faster than the 2nd pass only on a long sequences (typically whole movies) but quite slow on short sequences (less than about 10 000 frames) which is the case for a trailer.

Just a last note : with AVC it's quite easy to have extremely slow encoder settings. Those slow settings leads usually to marginally visual quality improvements.
It's about the same problem with other codecs (look at DivX 6 "insane" mode more instance).
Some users love to have access to these ultra-slow settings, that's why they are provided.

During the test, several codec-makers were wise enough to provide "reasonable" settings to avoid exhausting the patience of the reviewer. This was the case for XviD for instance who suggested only VHQ1 instead of its most powerfull mode which is VHQ4.

Personally, I prefer the settings given to Doom9 but with deblocking strengh set to 0 instead of -2. It's a matter of "taste" :)

[Toast]

Originally posted by kappa
System configuration:

Athlon 64 3200+ 1.5GB DDR400, VIA KT800 Socket 754 Gigabyte MB
Time: 12:56@DDR400
12:32@DDR420
11:52@DDR440

AMD wins!
The DDR420 measurement was the first taken, the other
two after consecutive reboots, so I guess that is the reason
for the non-linearity.
I'm waiting for the builds that doom9 used for his comparison,
because this is really slow. It takes 6 hours for a movie.
Could you explain your choices at the expert features dialog boxes?

Wow, those are some impressive results! I didn't expect A64 to be faster, especially by that much of a margin. I'm curious, were you increasing core frequency along with raising the mem speed to DDR 420 and 440 or were you maintaining the same core speed by reducing the multiplier while raising the mem speed?

Anyhow, I chose those settings as they are what looks best to my eyes when I'm trying to encode a half hour HDTV show down to 746MB (1/6 DVD) size. I could have asked for much faster settings but I'd question the accuracy of the results if say the whole process only took 30 seconds :D In which case I'd have to ask testers to DL a much larger sample so it was a bit of a trade off.

To be honest, I only use those settings for half hour shows. In my encodes I also do some pre-processing with Avisynth so it takes me about 11 hours to fully encode a half hour 1080i cap and a little bit less for a 720P stream. For hour or 2 hour long caps, I drop ref frames down to 1 and eliminate any pre-processing as it takes too long otherwise!

[Toast]

Originally posted by babayaga
The differences with your configuration are :
- the tested codec is about 40% faster on the 1st pass
- the tested codec is about 20% faster on the 2nd pass


This is welcome news as I have a tremendous backlog of HDTV caps waiting to be converted. Looking forward to the next release :)

[kappa]

Originally posted by Toast
I'm curious, were you increasing core frequency along with raising the mem speed to DDR 420 and 440 or were you maintaining the same core speed by reducing the multiplier while raising the mem speed?

The cpu is multiplier locked, so DDR420 means an FSB of 210 MHz (5% overclock) and DDR440 means 220 MHz (10% overclock). Plus I'm using a not-so-fast socket 754 and a not-so-fast chipset.

@babayaga

Thank you very much for the info. When I wrote 6 hours, I was refering indeed to a full encode (2 hour movies). My (HTPC) Ahtlon XP 2400+ needs 9.5h!!
Do the faster settings refer to a specific AVC profile currently in Recode, e.g. "Cinema - AVC", or do we wait for the "Normal Profile" of the next release? I too am not a fun of marginal "insane" modes, what you get for 100%-300% more processing time you can probably get with a 5% bitrate increase. IMHO the difference between "transparent" re-encodings (not yet achieved in 1CD encodes) and "very good - mp3" ones lies in general codec performance and not in a setting or two.
Just my 0.02€.

Edit: I RTFM'd (Doom9's guide on Recode) so everything is clear to me now.


PS I know this is not Ahead support :-), I would just like to mention that even with the preview in Recode disabled, I can't start the TV of my AIW 9800. When it takes 9.5h for an encode, that can be a problem!

[DaveId]

having used an AIW previously i know the troubles with the overlay being active upon launch of the tv program, or capturing programs accessing the tuner, try launching the aiw tv program first, have it run in the background, see if then you still have video when you launch subsequent programs like recode.
hope this helps.
dave

[dvd_maniac]

PS I know this is not Ahead support :-), I would just like to mention that even with the preview in Recode disabled, I can't start the TV of my AIW 9800. When it takes 9.5h for an encode, that can be a problem!

I have had nothing but AIW's in my PC since it's inception of the 2MB version. I'm now running the AIW 9800 Pro and waiting for AIW X800. I have never used anything else. I did however get rid of MMC and that was one of the reasons. I have since moved to SageTV with hardware encode cards. Much better cap quality. When I do my encodes even in Overlay mode it doesn't interfere with TV. BetondTV would also work and can use the AIW to cap. MUCH better EPG and features.

[kappa]

Thanks a lot for the (admitably off-topic) advice, but I really like MMC 9.03, it captures perfectly (as seen on a 130" display) and the TV-on-demand works flawlessly. Sage TV nees a hardware encoder and when I tested BeyondTV I found it premature. Maybe I'll give it a shot again. But this still is a Recode problem above all.

[The Edge]

WinXP SP2
CPU: Pentium 4M 3.06GHzH/T
RAM: 512mb PC2700, Intel 845GL

Encoding Time:
Recode 2.2.6.1 15min 47sec
Recode 2.2.6.4 15min 56sec ?

Able to playback ok?:
Yes

CPU usage during playback:
47-47% (FFDshow 05/01/2005) No Hyperthreading
46-50% (Nero AVC Dec 2.0.2.30) Looks like H/T used during playback....lot smoother!

Playback in MPC. Filesize was 16,119Kb on both encodes.;)
Anyone else got time to test this on their rigs?

[woah!]

just tried this on my setup:

A64 3000+ @ 2300mhz (10x230fsb)
1gig ballistix pc4000 ram 230@2.5-2-2-5 timings
msi neo plat 754 mobo
seagate 200gig HD 7200rpm
ati X800Pro Vivo.

Encoding Time:
Recode 2.2.6.1 12min 08sec
size after encode: 15.7meg (16,505,093 bytes)

Able to playback ok?:
Yes

CPU usage during playback:
between 60%-70% using nero showtime 2
between 50%-65% using powerDVD 6 Deluxe

[eb]

A little OT acc.to TIME IS MONEY
Above sample was processed with Athlon (at 2000MHz) and Xvid at defoult settings:
time: 90 seconds
size: 5.4MB with ac3 as it is
Please watch this sample not using magnifing glass with normal distance.
Please make some comments.
ftp://www.eb.enterpol.pl:eb@www.eb.enterpol.pl/Alias+.avi

eb

[woah!]

Originally posted by eb
A little OT acc.to TIME IS MONEY
Above sample was processed with Athlon (at 2000MHz) and Xvid at defoult settings:
time: 90 seconds
size: 5.4MB with ac3 as it is
Please watch this sample not using magnifing glass with normal distance.
Please make some comments.
ftp://www.eb.enterpol.pl:eb@www.eb.enterpol.pl/Alias+.avi

eb


slight difference is yours is at 640x360 AR and the test results we show are 1280x720. might be why yours is faster, but if i expand yours to the same size it looks... well terrible.

[eb]

I rather expected comments about "original" sample that is probably from some material upscaled to 1280x720. As for HDTV quality is not so good as should be.

eb

[kappa]

Originally posted by eb
I rather expected comments about "original" sample that is probably from some material upscaled to 1280x720. As for HDTV quality is not so good as should be.
If the source is analogue, one can easily produce true 1280x720 out of it. Even if that isn't true for the particular sample (I am not saying that it is and I won't re-download it just to see), this doesn't change the merit of this particular discussion nor the undeniable fact that HDTV is a breathtaking experience compared to DVD (D1 I think) resolution, especially if viewed on a projector.
Preserving the resolution of original HDTV content is not moot, although IMO I wouldn't do that for a TV series :) . Of course, I imagine that Toast supllied ths particular intro as a non disturbing "non-copyrighted" sample. Maybe we could use a WMV9 sample for the test. We could also use the settings used in Doom9's test for a better speed/quality ratio.

[Toast]

Thanks for all the results so far The A64s are looking good (I'm leaning in this direction). Anyone got a fast (4+ Ghz) P4 or a dually to test this on?

eb:
I'm not sure where you're trying to go with your example. You're comparing apples to oranges here. If you're asking if I can see a noticeable quality difference, the answer is yes. I don't need my glasses to see the quality difference both on my LCD monitor or my HDTV. The speed comparison also makes no sense. You're only processing 1/4 resolution at less than optimal settings so you'll obviously see a speed reduction with any codec you use. If you encode the stream in Xvid without resizing using MSP6, VHQ4, GMC, and QPel then the time savings are not so great. And the quality is still inferior which is my primary concern. :D

kappa:
Yes, I put that sample up because I thought it would be fairly "safe" as opposed to say putting up some HD clips of The Matrix :p I will look into putting up some better samples shortly. Both 1080i and 720P.

[DeeGee]

Seems like good old Athlon XP (barton) can still kick some but :)
CPU: 2800+ @ 200Mhz x 11 = 2200Mhz
Mem: 1GB
Time: 14:42

I'm not sure if it changes anything, but I used decomb.dll (version 5.00?) instead of Decomb521.dll. (Lazy, so I didn't bother to get it)

Seems like pleyback in Nero Showtime takes something like 99% of cpu time :) Still no dropped frames or anything.

[dark_eye]

WinXP SP2
CPU: AMD Athlon XP 2700+ (2.13 GHz)
RAM: 2x256MB DDR-2700, Nforce2

Encoding Time:
15:51 sed

Playback:
100% CPU usage
but not sure about dropped frames

after all, only 25% behind, no need to buy a faster machine :)

[hoyas79]

Ok. Very n00b on this subject. Alert.

CPU: Mobile Athlon64 2000+
RAM: 2x256MB DDR? (dunno which)
HDD: Toshiba 60GB 7200rpm
GPU: ATI Mobility Radeon 9700 128MB
OS: WinXP SP2 Pro
RECODE: 2.2.6.9
decomb521.dll
dgdecode.dll 1.01

1st Time: 15'07'' 16.495.xxx bytes
2nd Time: 15'27'' 16.491.653 bytes (why?)

In no way my system is optimized, meaning, no latest drivers, just standard install. I did close some programs like, Samurize, Google desktop and internet connection, leaving Recode barely as the only application running.

Because I forgot one setting, I put 3 maximum reference frames in the first encode and also forgot to remove deinterlacing, resizing and cropping? Weird... second time was 20'' slower

Playback for both files: Ok 99% CPU load average (other 1% was something else)
- In MPC 6.4.8.3 it played smoothly (19 to 20fps is this ok?)and did not report missing frames.
- In Showtime (do not recall the version though) it did not play smoothly, plus I had a warning about high definition content!? wtf??

Several questions to ask if you have the time to answer:
1) Why Alias? The intro is good but,... I do not like her face.
2) If a 26' clip takes 15' to encode in MD AVC that is very good ratio that will take 3 days to encode 1 hour of HDTV. Considering when I used Cinema AVC with regular settings on a DVD the ratio was much closer to 2.5-3 hours per hour of movie. (And sound/subtitle encoding was present). Why go there?
3) In the clip just rendered when I playback I can see "pixels" moving on her face, like some noise plugin was being applied at random on each frame, did this occur in your encodings?
4) Why are subtitles so ugly in ND encodes (at least Cinema AVC)? (no relation to this post I know)
5) How do I encode/import the .ac3 track to ND?
6) It the log says there are before each pass two lines saying it did not recognize the two plugins/filters (I assume decomb and dgdecode). Any ideas?
7) Just out of curiosity (probably an ignorant question), an MPEG2 DVD movie could be seen as a HDTV movie with higher compression and cropped to 720xWhatever resolution?
8) Unrelated to this post as well, I recently encoded The Secretary, in Cinema AVC, regular settings, Fit 2 CDs, and 1st CD had higher bitrate than 2nd CD, like 500kbits difference. Any suggestions? On another movie this difference was lower, around 15 kbits.

Thank you
h79