CoreCodec/H.264 Codec "CoreAVC"

[soresu]

Does it still count as worthwhile SMP if you have a Hyperthreading P4?

[Sharktooth]

sorry but HT is not multicore/smp... it even gives weird results in multithreaded apps...

[BetaBoy]

You will see improvments with HT as well... our testing is showing a 15% improvement.

[BetaBoy]

Some early test results;

- HT gets about 15-20% boost
- Dual Core gets about a 90% increase in playback performance
- Load balancing needs to be optimized (cpu usage differs)
- 'Field only' has some bugs

On a side note... TS format is now fully supported.

[Gnerma]

The SMP support is looking great BetaBoy. Forgive me for asking this but will you be going into detail about GPU support once version 1 is out the door?

Regardless I've got my credit card in hand :D

[Selur]

@betaboy: got a dual athlon 1800+ MP system so dropped you a mail ;)

[sasam]

That is great stuff...

[Eretria-chan]

I'll be interested to participate as well. I already sent an email.

[BetaBoy]

I am emailing everyone now... you might as well post your results to this thread so the community can see the progress we are making.

Specifically I will send those people TCPMP win32 with SMP enabled... I will also send the pre SMP 'avc.plg' to compare it too.

Post the differences between the two.... but even if you dont have SMP and have HT available. You will see a nice boost in performace ;-)

Anyone say 64 bit version?

[Kostarum Rex Persia]

Well, yes, 64-bit version would be very nice to see.

[Eretria-chan]

Yeah, since I have an Athlon64 and will be installing XP 64-bit, that would also be interesting to see! Send a 64-bit build when you can too! Will be interesting to see the difference.

[Sirber]

WinXP64 is not worth it.

[tomos]

got my mail (thanks Dan :) )

started some little tests and so far:

BUGS

player sees avi's as folders and gives error on trying to open "private test version!, CPU 0003 2", misreports their size. opens and plays fine afterwards tho - apart from seeking wont work (all my clips are in avi so may just be due to that) no probs at all with codec itself so far. :D

CPU usage

with clips that just cant play well on a single core alone, it plays flawlessly so far. havent had a clip go above 55% overall yet.

Speed gain

seems 1 core cpu % is lower than the prev version i was using. playing same clip in mpc versus this one. MPC goes about 45% but with this codec it hits about 30-35% with same vid clip.

CPU info - 4400+ X2 @ stock

will have to make a new clip to mess around with at different bitrates for more info.

[3ngel]

There will be some price discount for the retail version, for us "alpha tester" doom9 registered? :D

[Eretria-chan]

WinXP64 is not worth it.
Why so? I still need to reinstall this stupid Windows installation that seems to have become corrupted after a few days use.

[Sirber]

I tryed it a month and I came back to 32bit. Too much incompatibility with current softs and no speed gain.

[tomos]

Vista x64 should do it better i would hope. dont see why MS are doing it in 32 bit as well.

[Sirber]

I won't move to vista. I don't like their DRM stuff.

[nashse]

- Bugs

- CPU load
30-50 % running maria_taylor_m1080p.mov

- Speed gain
pre smp version Average Speed 99.84% running benchmark of maria_taylor_m1080p.mov

smp Average Speed 185.44% running benchmark of maria_taylor_m1080p.mov


- How many CPU's (2-4)
Dual Core Intel Pentium D 805

[tomos]

I won't move to vista. I don't like their DRM stuff.

i dont like that either, but in time, games will be vista only i would think, or something more in line with this forum - dxva 2 will be vista only

will see how it goes for a few months before moving across tho

[Sirber]

Then I'll put linux and use a console for games :)

[tomos]

consoles! for FPS? that's like gaming heresy man.

:p

[niknik]

Aren't you getting a bit off topic?
(unless games start streaming realtime video output in x264... ehehe :)

[Selur]

here are my first benchmarks using two trailers and a commercial on my dual athlon 1800+ MP system with the TCMP_SMP_version.

X-Men 2 Trailer:

SMP:
average: 29.190%
maximum: 53.906%
benchmark Average Speed: 120.01%
(see: xmen2_bench_20060316_171446_smp.txt)

non SMP:
average: 30.777%
maximum: 42.969%
benchmark average speed: 71.29%
(see: xmen2_bench_20060316_171903_normal.txt)

bravia commercial:

SMP:
average: 2.683%
maximum: 15.625%
benchmark average speed: 340.60%
(see: bravia_bench_20060316_171137_smp.txt)

non SMP:
average: 2.375%
maximum: 17.969%
benchmark average speed: 288.83%
(see: bravia_bench_20060316_165546_normal.txt)


Ice Age 2 Trailer:

SMP:
average: 37.985%
maximum: 71.875%
benchmark Average Speed: 86.00%
(see: iceage2_bench_20060316_164324_smp.txt)

non SMP:
average: 28.224%
maximum: 37.500%
benchmark average speed: 56.18%
(see: iceage2_bench_20060316_164911_normal.txt)

sidenode:
- used perfmon to monitor cpu usage
- used benchmark incorparated in TCMP
- attached the mentioned txt files to post

Cu Selur

[BetaBoy]

Selur... thx for the post... it seems dual AMD MP's are a little slower then I had expected.

[tomos]

some more results:

made a 1080p clip and a 10mbit bitrate which was pretty much unplayable on a single core - played flawless on the new SMP coreavc. definetely more efficient than the prev releases too. shows when locking affinity to just one core.

anyway, some benchies.

x-men - last stand - 1080p quicktime clip on my 4400+ X2

TCPMP Version 0.72RC1 Benchmark Results

Average Speed 381.48%
Video Frames 3590
Audio Samples 7207147
Amount of Data 179001 KB

Bench. Time 0:39.250
Bench. Frame Rate 91.46
Bench. Sample Rate 183623
Bench. Data Rate 37.3 Mbit/s

Original Time 2:29.733
Original Frame Rate 23.98
Original Sample Rate 44100
Original Data Rate 9.7 Mbit/s

URL c:\xmenthelaststand_h1080p.mov
Size 183297384
Platform Windows
OS Version 5.01
Clock speed 2140 Mhz
Video output DirectDraw 1920x1200 32bits Lookup
Video zoom 1920x800 -> 1920x800
Audio output Wave Output 44100Hz 16Bits 2Ch.

x-men - last stand - 1080p quicktime clip on my 3800+ X2 (O/C)

TCPMP Version 0.72RC1 Benchmark Results

Average Speed 249.20%
Video Frames 3599
Audio Samples 7187659
Amount of Data 179001 KB

Bench. Time 1:00.234
Bench. Frame Rate 59.75
Bench. Sample Rate 119330
Bench. Data Rate 24.3 Mbit/s

Original Time 2:30.108
Original Frame Rate 23.98
Original Sample Rate 44100
Original Data Rate 9.7 Mbit/s

URL c:\xmenthelaststand_h1080p.mov
Size 183297384
Platform Windows
OS Version 5.01
Video output DirectDraw 1920x1200 32bits Lookup
Video zoom 1920x800 -> 1920x800
Audio output Wave Output 44100Hz 16Bits 2Ch.

****************
for some reason, this is slower than my 4400+??? i mean this is clocked to 2.8 gig per core as opposed to 2.2 for the other. everything else is faster on this PC apart from this? weird. also the clock speed is missing from the results. its not in the txt file or on the popup after finishing the benchmark.
****************

Ultraviolet 1080p quicktime trailer

3800+ X2

TCPMP Version 0.72RC1 Benchmark Results

Average Speed 249.40%
Video Frames 3286
Audio Samples 6558374
Amount of Data 165204 KB

Bench. Time 0:54.952
Bench. Frame Rate 59.80
Bench. Sample Rate 119350
Bench. Data Rate 24.6 Mbit/s

Original Time 2:17.053
Original Frame Rate 23.98
Original Sample Rate 44100
Original Data Rate 9.8 Mbit/s

URL c:\ultraviolet-tlr1_h1080p.mov
Size 169169668
Platform Windows
OS Version 5.01
Video output DirectDraw 1920x1200 32bits Lookup
Video zoom 1920x1056 -> 1920x1056
Audio output Wave Output 44100Hz 16Bits 2Ch.


4400+ X2

TCPMP Version 0.72RC1 Benchmark Results

Average Speed 348.04%
Video Frames 3281
Audio Samples 6573737
Amount of Data 165204 KB

Bench. Time 0:39.318
Bench. Frame Rate 83.45
Bench. Sample Rate 167196
Bench. Data Rate 34.4 Mbit/s

Original Time 2:16.845
Original Frame Rate 23.98
Original Sample Rate 44100
Original Data Rate 9.8 Mbit/s

URL c:\ultraviolet-tlr1_h1080p.mov
Size 169169668
Platform Windows
OS Version 5.01
Clock speed 2170 Mhz
Video output DirectDraw 1920x1200 32bits Lookup
Video zoom 1920x1056 -> 1920x1056
Audio output Wave Output 44100Hz 16Bits 2Ch.

***********
same again. playing these back normally definetely uses more CPU time on the 4400 than the 3800 so why does the benchie not show this?
***********

[Selur]

Did another test and caused a crash while scolling through a file.
get it here: http://rapidshare.de/files/15668430/lemony-muxed.mp4.html

lemony

SMP:
average: 65.574%
maximum: 80.469%
benchmark Average Speed: 80.20%
(see: lemony_bench_20060316_190607_smp.txt)

Single:
average: 38.164%
maximum: 49.219%
benchmark average speed: 42.67%
(see: lemony_bench_20060316_190607_normal.txt)

attached the txtfiles and the crash.txt
(if needed I can also provide the source and the avs script that I used to create the file)

Cu Selur

[Eretria-chan]

EDIT: Looking for HD trailers...
Can't seem to find any...

[Sirber]

Is it me or a new buid is avalible and everyone is testing it except me? ;)

[tomos]

we mailed betaboy as he req if we wanted to test the new version :)

[PicardGK]

some more results:
3800+ X2
Bench. Frame Rate 59.75
Bench. Frame Rate 59.80

same again. playing these back normally definetely uses more CPU time on the 4400 than the 3800 so why does the benchie not show this?

It seems the 3800+ has a video card which always does vertical sync (and limits benchmark to 60fps). Try comparing the speeds after selecting option/video/null menu.

[tomos]

It seems the 3800+ has a video card which always does vertical sync (and limits benchmark to 60fps). Try comparing the speeds after selecting option/video/null menu.

ok cool. cant see the video playing anymore but seems to go faster :thanks:

TCPMP Version 0.72RC1 Benchmark Results

Average Speed 642.81%
Video Frames 3593
Audio Samples 7227914
Amount of Data 179001 KB

Bench. Time 0:23.312
Bench. Frame Rate 154.12
Bench. Sample Rate 310059
Bench. Data Rate 62.9 Mbit/s

Original Time 2:29.858
Original Frame Rate 23.98
Original Sample Rate 44100
Original Data Rate 9.7 Mbit/s

URL c:\xmenthelaststand_h1080p.mov
Size 183297384
Platform Windows
OS Version 5.01
Video output Null Video 1920x800 0bits Lookup
Video zoom 1920x800 -> 0x0
Audio output Wave Output 44100Hz 16Bits 2Ch.

and

TCPMP Version 0.72RC1 Benchmark Results

Average Speed 648.94%
Video Frames 3282
Audio Samples 6595727
Amount of Data 165204 KB

Bench. Time 0:21.093
Bench. Frame Rate 155.59
Bench. Sample Rate 312723
Bench. Data Rate 64.1 Mbit/s

Original Time 2:16.886
Original Frame Rate 23.98
Original Sample Rate 44100
Original Data Rate 9.8 Mbit/s

URL c:\ultraviolet-tlr1_h1080p.mov
Size 169169668
Platform Windows
OS Version 5.01
Video output Null Video 1920x1056 0bits Lookup
Video zoom 1920x1056 -> 0x0
Audio output Wave Output 44100Hz 16Bits 2Ch.

[travisbell]

Wish I had a SMP system to test with *sniff-sniff*

[MuTeK]

TCPMP Version 0.72RC1 Benchmark Results

harry_potter_goblet-tlr2_h720p.mov

Average Speed 1766.44%
Video Frames 3342
Amount of Data 114390 KB

Bench. Time 0:07.890
Bench. Frame Rate 423.52
Bench. Data Rate 118.7 Mbit/s

Original Time 2:19.389
Original Frame Rate 23.98
Original Data Rate 6.7 Mbit/s


x-men_3-pre_teaser_h720p.mov

Average Speed 2096.96%
Video Frames 2362
Amount of Data 70145 KB

Bench. Time 0:04.697
Bench. Frame Rate 502.77
Bench. Data Rate 122.3 Mbit/s

Original Time 1:38.515
Original Frame Rate 23.98
Original Data Rate 5.8 Mbit/s

[foxyshadis]

All are audio disabled, ddraw interface.

Silent Hill 1080p:

Average Speed 409.45%
Video Frames 3594
Amount of Data 187462 KB

Bench. Time 0:36.609
Bench. Frame Rate 98.17
Bench. Data Rate 41.9 Mbit/s


Original Time 2:29.774
Original Frame Rate 23.98
Original Sample Rate 44100
Original Data Rate 10.2 Mbit/s

non-smp:

Average Speed 224.43%
Video Frames 3601
Amount of Data 187462 KB

Bench. Time 1:06.921
Bench. Frame Rate 53.81
Bench. Data Rate 22.9 Mbit/s


Da Vinci Code 1080p:

Average Speed 324.73%
Video Frames 3594
Amount of Data 161894 KB

Bench. Time 0:46.160
Bench. Frame Rate 77.86
Bench. Data Rate 28.7 Mbit/s

Original Time 2:29.899
Original Frame Rate 23.98
Original Data Rate 8.8 Mbit/s

non-smp:

Average Speed 177.34%
Video Frames 3601
Amount of Data 161894 KB

Bench. Time 1:24.690
Bench. Frame Rate 42.52
Bench. Data Rate 15.6 Mbit/s


So far all of the trailers play fine, no hiccups or random bugs noticed. Normally uses up around 30% cpu. Oh, and audio drops the bench speed by about 7% in each case.

I'll try testing non-apple trailers when I get more time to make one this weekend.

Specs:
1.66 GHz Core Duo (2mb shared cache)
2GB 677 DDR2
X1300 mobility

It'll probably be too late to update for the 4-way servers coming in next month, but I can give the dual xeons w/ hyperthreading a shot.

[Selur]

V

SMP:
average: 26.255%
maximum: 48.438%
benchmark Average Speed: 120.64%
(see: v_bench_20060317_100704_smp.txt)

Single:
average: 28.622%
maximum: 36.719%
benchmark average speed: 75.11%
(see: v_bench_20060317_101814_normal.txt)

ipod_itunes-1000_songs_h1080p

SMP:
average: 30.957%
maximum: 48.438%
benchmark Average Speed: 109.79%
(see: ipod_bench_20060317_104045_smp.txt)

Single:
average: 28.784%
maximum: 39.063%
benchmark average speed: 67.64%
(see: ipod_bench_20060317_103741_normal.txt)

basic_instinct_2_h1080p

SMP:
average: 48.958%
maximum: 67.969%
benchmark Average Speed: 91.39%
(see: basic2_bench_20060317_104701_smp.txt)

Single:
average: 31.116%
maximum: 13.750%
benchmark average speed: 55.05%
(see: basic2_bench_20060317_105525_normal.txt)

textfiles attached as usual ;)
Cu Selur

[billou2000]

I done some tests too with 2 encoded avc streams on a dual xeon 3.2 Ghz.
Two things to note:
The video card is a bit old and slows everything down if the windows gets too large, so the bench were processed in NULL renderer, and the CPU usage in real time playback were obtained with DirectX in the default window size of TCPMP, which is ok.
I also put a quick comp with nero decoder.

clip1: D1 25fps @ 2mbps (High Profile)
------------------------------------
coreAVC 1CPU
bench 563.37% ( 11.2 Mbps)
playb. 11-19 % CPU (avg 13%) / 25 fps

coreAVC 2CPU
bench 1076.42 % (21.5 Mbps)
playb. 9-19% CPU (avg 13%) / 25 fps

Nero
playb. 16-22 % CPU (avg 19%) / 25 fps

clip2: 720p 50fps @ 4mbps (High Profile)
---------------------------------------
coreAVC 1CPU
bench 129.83 % (5.1 Mbps)
playb. avg 40% / 30 fps

coreAVC 2CPU
bench 245.32 % (9.8 Mbps)
playb. avg 38% / 50 fps

Nero
playb. 50% CPU / 20 fps

[sasam]

Did you delayed the release of coreavc because you wanted to add SMP support to it before the release?
Can someone test how much faster is it with HT turned ON compared to OFF on 1core...

[Sharktooth]

betaboy said 15% improvment:
You will see improvments with HT as well... our testing is showing a 15% improvement.

[tomos]

there's improvement on a single core PC as well (tried on my old athlon xp) V prev beta release

[SeeMoreDigital]

there's improvement on a single core PC as well (tried on my old athlon xp) V prev beta releaseThat's very nice to know :D

Wish I could try it!

[sasam]

That's great progress. By the time its finished if it continues to develop at this rate it will be able to play HD AVC movies on PII :D

BetaBoy, will this new version be on your site or some older stable version on March xx ?

[Eretria-chan]

Argh, really sorry to haven't been able to test, but I've had problems with the computer.
It seems the windows installation is corrupted :( I will format it and reinstall, at least until tomorrow and hope that I will be able to test by then.

[BetaBoy]

Our plan atm is just for the Directshow filter to be available for the initial launch (Standard and Professional). We maybe offering a download as well for both CorePlayer and BetaPlayer till we officially launch them.

[Eretria-chan]

Okay, first results are in! I tried the XMen 3 demo with directshow, and unfortunaly, the computer was unable to properly display it (it jumped to 100% and down, then up, then down, etc) and the picture was lagging. This might be due to that I have underclocked the system (lowered the hypertransport and the ram speed), so I will test again later.

With TCPMP:
Okay so using TCPMP SMP I got cpu load about 40% - 60% flawless playback! Great work!
With TCPMP Non-SMP I got practically the same... is this normal? Both seems to use both the processors really. Unless I did something wrong...

[Selur]

@betaboy: did you check the crash.txt ?
@Eretria-chan: you did excahnge the avc.plg didn't you?

[jblade]

Anyway we can get an updated CoreAVC codec before release :) still using .0.0.4

[Eretria-chan]

@betaboy: did you check the crash.txt ?
@Eretria-chan: you did excahnge the avc.plg didn't you?
Yes, I did. I moved the smp one avc.plg to a sub-folder called plugins and copied the old one from the mail into the folder.

[BetaBoy]

Selur... yes, thx. I'll probally have one more test for you guys before we slap 1.0 on it. What we really need atm is some more people with 'true' 4 CPU's to test loads on each CPU.

Anyone?

[GhengisKhan]

People have 4 CPU's! I want one of those computers - getting a 2.16 Dual Core MacBook Pro (soon)... I guess that would be two CPU's...

Betaboy, how does the current, updated decoder compare to the latest public beta test?