Microsoft Expression Encoder 2 GPGPU?

[FoxRacR17]

So I'm starting to build a new i7 rig, and I was looking at video cards. I have pretty much decided on he ati 4870. While I was doing my research on different video cards I came across THIS (www.amd.com/us-en/assets/content_type/DownloadableAssets/ATIStreamUpdate_Press_Deck_11.13.08_Final.pdf) .pdf. On page 20 it says that microsoft expression encoder will use ATi stream to encode things faster. Is this true? Does this mean that if I get the 4870 and use expression encoder 2 to encode my hd wmv movies it will be much faster because it will be using the 4870? If thats the case then I will buy expression encoder 2 and the 4870 in a heartbeat. Anyone have any experience with this?

[ACrowley]

I dont think so

Look here
http://social.expression.microsoft.com/Forums/en-US/encoder/thread/7ff293e9-794d-46be-a8c7-f46a18720bc4/

The Ati Stream Capability was released with Catalyst 8.12...and there was no new Expression Encoder 2 Update so far...

I think MS i talking about ATI AVIVO GPU Accelartion/DXVA for the Source Videos

[FoxRacR17]

Hmm, maybe there will be an update soon then, because on page 20 of that .pdf is says" Accelerated ENCODING of contect" Thats what made me think it helps encode, rather then decode, ala DXVA.

[benwaggoner]

Hmm, maybe there will be an update soon then, because on page 20 of that .pdf is says" Accelerated ENCODING of contect" Thats what made me think it helps encode, rather then decode, ala DXVA.
Expression Encoder 2 uses the VC-1 Encoder SDK, and can be hardware accelerated by any product that uses the associated acceleration API.

The only shipping products that use that API are the Tarari boards from LSI. I haven't heard about ATI announcing anything there, but it should would be cool if they did!

EE will use the GPU to decode source if it's a supported format.

[FoxRacR17]

EE will use the GPU to decode source if it's a supported format.

Oh I think I understand. So say that I am encoding to VC-1 from a bluray rip that is h.264. EE will use the GPU to decode the source h.264 leaving the CPU free to do all the encoding to VC-1?

[Furiousflea]

Oh I think I understand. So say that I am encoding to VC-1 from a bluray rip that is h.264. EE will use the GPU to decode the source h.264 leaving the CPU free to do all the encoding to VC-1?

Sortof, theoretically, yes. In practice hardly makes any difference at all in Expression Encoder.

The meat of the processing is encoding, off-loading the decoding saves hardly any time at all.

[Sharktooth]

well, not properly.
decoding HD stuff (expecially h.264 videos) require a LOT of CPU cycles.

[Dark Shikari]

well, not properly.
decoding HD stuff (expecially h.264 videos) require a LOT of CPU cycles.But by comparison, have you seen the speed of the Expression Encoder, or perhaps should I say the lack thereof? :p

[Sharktooth]

lol, yes. that's why you need every CPU cycle you can spare...

[FoxRacR17]

So what would probably be the best program to use to encode vc1 wmv video files to playback on the 360? What are people using now instead of the modified wmcmd.vbs?

[Dark Eiri]

I'm using the Expression Encoder 2 SP1. Yes it is INSANELY "heavy" and encodes absurdly slow compared to x264 encoding, but it's the fastest VC-1 encoder out there. TMPGEnc is about two times slower to encode.

[Sharktooth]

@FoxRacR17: there's no best as per forum rules.

[Sagittaire]

I'm using the Expression Encoder 2 SP1. Yes it is INSANELY "heavy" and encodes absurdly slow compared to x264 encoding, but it's the fastest VC-1 encoder out there. TMPGEnc is about two times slower to encode.

Well slow speed is always relative to setting. You have slow encoding because you use slow setting. VC1-SDK encoder can be really fast if you use fast setting.

Moreover Mainconcept have a really fast VC1 encoder with default setting.

[Dark Eiri]

^ I'm using minimally reasonable settings. Everything is at the default, except for B-Frames, increased to 3. And still is absurdly slow. If I set it to something that could actually be on-par with x264, it would take about 2 months to encode a 3min 720p video.

[Sagittaire]

1) bframes are not adaptative with VC1 SDK. Use only 1 or 2 bframes.

2) Simply don't use default setting.
-complexity 1 -motionsearchlevel 4 -mesearchmethod 2 -mbcost 0 -mvcost 1 -mvrange 4
If you make high bitrate encoding use high complexity level is useless.

[Dark Shikari]

Well slow speed is always relative to setting. You have slow encoding because you use slow setting. VC1-SDK encoder can be really fast if you use fast setting.Really fast? It's like 3 times slower than Mainconcept's H.264 and 5x+ slower than x264, last I tried :confused:

[Sagittaire]

Really fast? It's like 3 times slower than Mainconcept's H.264 and 5x+ slower than x264, last I tried :confused:

Well speed comparison is relative here ...

If you make comparison at same "quality" (aka PSNR or SSIM) it's simply impossible to compare x264 and VC1 because in my last test (very old) highest quality VC1 encoding (slowest) is not able to have higher metric score than lowest x264 encoding (fastest). I don't know if it's always the case.

http://jfl1974.free.fr/Sample/Speed.PNG

[Sagittaire]

Really fast? It's like 3 times slower than Mainconcept's H.264 and 5x+ slower than x264, last I tried :confused:

I remake little test here with my core2 duo and Mainconcept SDK is on par with x264 for speed with balanced profil.
I can make more serious test but I am sure that x264 is never 2x faster than Mainconcept for same quality at same bitrate for usual profil.


D:\Mes dossiers\Codec\x264>x264 --crf 24 --threads auto --thread-input --bframes 3 --b-adapt 1 --b-p
yramid --weightb --ref 3 --mixed-refs --direct auto --deblock -2:-2 --psy-rd 0.0:0.0 --aq-mode 0 --a
q-strength 0.0 --progress --8x8dct --subme 6 --me "hex" --partitions "all" --trellis 1 --qcomp 0.75
--keyint 240 --mixed-refs -o x264HP-896.mp4 test.avs
avis [info]: 720x480 @ 23.98 fps (4123 frames)
x264 [info]: using cpu capabilities: MMX2 SSE2Fast SSSE3 Cache64
x264 [info]: profile High, level 3.0
mp4 [info]: initial delay 2002 (scale 24000)
x264 [info]: slice I:110 Avg QP:22.42 size: 19021 PSNR Mean Y:47.92 U:52.70 V:53.26 Avg:48.89 Gl
obal:46.16
x264 [info]: slice P:2083 Avg QP:24.37 size: 6986 PSNR Mean Y:44.12 U:48.77 V:49.69 Avg:45.17 Gl
obal:43.36
x264 [info]: slice B:1930 Avg QP:25.65 size: 1466 PSNR Mean Y:44.56 U:49.96 V:50.52 Avg:45.61 Gl
obal:42.78
x264 [info]: consecutive B-frames: 23.8% 32.0% 12.3% 31.9%
x264 [info]: mb I I16..4: 36.5% 54.3% 9.2%
x264 [info]: mb P I16..4: 11.4% 15.0% 2.0% P16..4: 29.0% 8.0% 4.2% 0.6% 0.5% skip:29.2%
x264 [info]: mb B I16..4: 0.6% 0.0% 0.0% B16..8: 13.8% 1.8% 1.2% direct: 3.0% skip:79.7% L
0:27.3% L1:47.0% BI:25.7%
x264 [info]: 8x8 transform intra:52.1% inter:61.5%
x264 [info]: direct mvs spatial:98.0% temporal:2.0%
x264 [info]: ref P L0 88.0% 8.0% 4.0%
x264 [info]: ref B L0 92.1% 7.9%
x264 [info]: ref B L1 96.7% 3.3%
x264 [info]: SSIM Mean Y:0.9758153
x264 [info]: PSNR Mean Y:44.430 U:49.431 V:50.175 Avg:45.472 Global:43.130 kb/s:905.94

encoded 4123 frames, 32.21 fps, 906.14 kb/s

D:\Mes dossiers\Codec\Elecard2>mch264ve.exe config22.cfg SourceFileName = test.avs TargetFileName =
"ElecardAVC_900.264" AvgBitrate = 906000
MainConcept H.264/AVC encoder (build 2.1.6907)
Copyright (c) 2006 MainConcept AG

THIS SOFTWARE IS FOR EVALUATION PURPOSES ONLY!

[time: 0:01:40] [left: 0:00:00] [speed: 41.0 fps] [queue: 0.000 ms]]

Summary information:
Number of coded frames 4123
Total encoding time 100593 ms
Average time per frame 24.398 ms
Average speed achieved 41.0 fps
Average bitrate 905.44 kbit/sec @ 23.98 Hz

Overall PSNR (Y) 41.881 dB
Overall PSNR (U) 47.187 dB
Overall PSNR (V) 48.192 dB
Overall PSNR (A) 43.103 dB

[Dark Shikari]

I remake little test here with my core2 duo and Mainconcept SDK is on par with x264 for speed with balanced profil.
I can make more serious test but I am sure that x264 is never 2x faster than Mainconcept for same quality at same bitrate for usual profil.Of course, I was doing a very barebones test designed to test the maximum speed of the encoders (something that would matter when doing, for example, streaming). Comparing quality vs speed is completely absurd when comparing an H.264 encoder to a VC-1 encoder, since we all know exactly what the results of that will be.

I'm not going to comment on your intentionally bad choice of settings for x264, since we all know what happened last time you attempted to pull off this troll (you got your alt account banned from doom9).

[Sagittaire]

Of course, I was doing a very barebones test designed to test the maximum speed of the encoders (something that would matter when doing, for example, streaming). Comparing quality vs speed is completely absurd when comparing an H.264 encoder to a VC-1 encoder, since we all know exactly what the results of that will be.

I'm not going to comment on your intentionally bad choice of settings for x264, since we all know what happened last time you attempted to pull off this troll (you got your alt account banned from doom9).

lol ... little test in area with your setting.

[Sagittaire]

I have already make test and even at max speed x264 is never 2x faster than Mainconcept SDK. Really simple for me to prove that.

You never say "I'am wrong" ... lol

stop trolling about Mainconcept SDK ... please ... and make serious test!


I'm not going to comment on your intentionally bad choice of settings for x264, since we all know what happened last time you attempted to pull off this troll (you got your alt account banned from doom9).

1) It's not absurd setting
2) I have already make test with your setting and result is the same

http://forum.doom9.org/showthread.php?p=1212296#post1212296

[shon3i]

stop trolling about Mainconcept SDK ... please ... and make serious test!I am second on that. For same quality, MC AVC is lot faster than x264.

[benwaggoner]

I'm using the Expression Encoder 2 SP1. Yes it is INSANELY "heavy" and encodes absurdly slow compared to x264 encoding, but it's the fastest VC-1 encoder out there. TMPGEnc is about two times slower to encode.
Note that Expression Encoder defaults to some pretty slow preprocessing, particularly in the use of Super Sampling for scaling. The deinterlacer's no speed demon either in the current version, although it's got pretty good quality for a non motion adaptive one.

EE can take .avs natively. Set Size=Source and Deinterlacing=Force Off to do an apples-to-apples compare.

[benwaggoner]

Of course, I was doing a very barebones test designed to test the maximum speed of the encoders (something that would matter when doing, for example, streaming).
Note that live streaming would typically be done using Lookahed Rate Control and Dynamic Complexity, which will have pretty different performance characteristics than you might estimate based on a fixed complexity 1-pass CBR encode.

Quality wise, LRC handily outperforms 2-pass CBR in the current SDK (granted, that may say more about the limitations of 2-pass CBR, which as definitley been lower priority than 1-pass CBR and 2-pass VBR in recent years).

Hard to believe it's been a whole year since we RTM'ed the SDK. Hopefully we'll be able to share some of the improvements since then soon.

[Dark Eiri]

Note that Expression Encoder defaults to some pretty slow preprocessing, particularly in the use of Super Sampling for scaling. The deinterlacer's no speed demon either in the current version, although it's got pretty good quality for a non motion adaptive one.

EE can take .avs natively. Set Size=Source and Deinterlacing=Force Off to do an apples-to-apples compare.

I am using .avs already.

[benwaggoner]

I am using .avs already.
Are you disabling the EE preprocessing as well? =Source is particularly important, as by default it'll trim your frame size to maintain aspect ratio, unless Profile=Source.

I'm honestly not sure what heuristic it uses to determine if an .avs is interlaced or progressive, but setting Deinterlace=Force Off would make sure. That said, I've never seen it get it wrong.

There was one case reported a while ago where AVS + Haali splitter really freaked out and launched a dozen or so instances when running in EE (the same .avs wouldn't even play in other players).

[Dark Eiri]

There was one case reported a while ago where AVS + Haali splitter really freaked out and launched a dozen or so instances when running in EE (the same .avs wouldn't even play in other players).

When I tried TMPGEnc, that happened to me. Every single time.

But yes, I'm forcing preprocessing off. I'll try to tweak some settings to get some extra speed.

[easy2Bcheesy]

Does TMPEG actually use the VC1 SDK? I just can't justify $200 for Expression Encoder 2, even though I love its results, but $99 for a multi-purpose encoder I could live with.

[benwaggoner]

Does TMPEG actually use the VC1 SDK? I just can't justify $200 for Expression Encoder 2, even though I love its results, but $99 for a multi-purpose encoder I could live with.
Not that I know of, but they should. Since the Professional version is just a click-through EULA, we don't have any good way to track what products have included it, other than people telling us.

They definitely should use it! I imagine it'd be useful to make Blu-ray compatible VC-1 for that product, as well as getting the improved WMV support.

[benwaggoner]

When I tried TMPGEnc, that happened to me. Every single time.

But yes, I'm forcing preprocessing off. I'll try to tweak some settings to get some extra speed.
If speed is your biggest concern, you can always start out with the Fastest preset, and then work up from there. I was doing a fastest v. best test recently, and I had a 20 min encode in Fastest and a 6 hour encode in my Insane tweaked version of Best, so there's quite a lot of variability in there :).

[Dark Eiri]

I'm not worried about speed, I'm quality over speed, always. It's just that I can get the same results from x264 in one third of the time, but I'll lose 5.1 support on the 360, so I don't really have much choice here :D. I think Expression Encoder is really worth the price, since it's easily the fastest VC-1 encoder and gives the best results. It's just missing 5.1 audio encoding, but I can do that with any other encoder and just mux it later.

@easy2Bcheesy: Nope, they're using the old WMEncoder libraries. The only way of tweaking any settings is changing the registry. TMPGEnc is really bad compared to EE in most tasks (plus, their CUDA "optimizations" are absolutely useless).