L-SMASH Source

[almosely]

Never had any problems like that with L-Smash or DGDecNV in the last 10 years - so no hit-n-miss for me. Encoding without hw-acc-decoding is much slower.

[FranceBB]

Wow, I'm very surprised to see that it's actually stable even with videos with ref larger than 4.
Anyway, as kedautinh12 rightly pointed out, asd-g is the one keeping the development ongoing for LWLibav.
He generally listens to user, but he prefers to keep track of issues through GitHub rather than through posts on Doom9.
Please report the regression on GitHub by opening an issue here: https://github.com/HomeOfAviSynthPlu...H-Works/issues
I'm sure he'll take a look at it.

[almosely]

Already done before your two posts ;-) And he looked at it.

[Atak_Snajpera]

Originally Posted by almosely

Never had any problems like that with L-Smash or DGDecNV in the last 10 years - so no hit-n-miss for me. Encoding without hw-acc-decoding is much slower.

Have you done some real tests?
Encoding process for example x265 will consume ~95% of CPU time while decoding rest ~5%.

I wouldn't say that extra 5% will give you much faster encoding.

[almosely]

Real tests? I am encoding almost daily for over 20 years. Of course I am talking out of experience. With AVS I started with xvid, then changed to x264 and since a few years to x265. A few days ago I was encoding a 1080p video (usually I encode 720p). With software decoding: 6 fps. With hardware decoding 9 fps. So 50% faster in that usecase ...

[Atak_Snajpera]

So you encoded exactly the same video file twice right?

[almosely]

So, why is that you think everybody else is doing it wrong? ...

[almosely]

So, I did some tests, even reverted to earlier AVS and plugin-versions from that encoding-time, but sadly I cannot reproduce that penomenon anymore. I do not have that particular file anymore.

But I do remember, I started an encode with hw-decoding on (as usual). The estimated encoding time has been about 8 hours and the fps had been around 9. That did not change after having a look at that estimation a few hours later. The next day that encode ended somewhere after approx. 6 hours with an encoder crash. That happens sometimes, rarely, and usually changing a bit of the parameters of requestlinear solves that issue. So I wanted to get that encode finished correctly but did not want to risk another crash. That's why I gave hw-decoding switched off a try. I did not change anything else. Then the estimated encoding time was something about 12 hours and the fps were about 6 and that was true after the job finished - it crashed too, but after approx. 11 hours. So, hw on/off wasn't the reason fot that. Then I had to start a third encode for the last minutes to glue the peaces together in the end. Those crashes started with AVS+ (when prefetch(x,y) was introduced) and I have no idea what's the reason for that. There wasn't running anything in the background that time.

And I remember that I made some tests some years ago with hw-dec on/off and everytime hw-dec on has been faster, a bit, not as much as with that particular file, but faster, without any cost in power (watts). So I use hw-dec on every time.

[Atak_Snajpera]

Originally Posted by almosely

So, why is that you think everybody else is doing it wrong? ...

Placebo effect

source AVC blu-ray 1080p (no avisynth filters!)


source HEVC HDR10 UHD 2160p (no avisynth filters!)


If you use software encoder x265 (with default medium preset) bottleneck is so huge that hardware decoding does not help.
Basically decoder most of time is sitting idle and waiting for requests from encoding process.

In my case decoding process was using less than 5% of CPU time.

I really do not care about that few percent and prefer much stable CPU decoding.

[almosely]

Come on ... You are one person out of billions. The world is bigger than your imagination and experience. What is the meaning of compairing completely different hardware, yours and mine?

I am not using any default presets.

"In my case" and "I really do not care about" is the whole point of this.

I just wanted to report an error within the last two updates, no more.

[Atak_Snajpera]

That's why I asked if you did a REAL tests. Your claim that hardware decoding gives huge speed boost in encoding time is a lie.
If you do some filtering for example denoising/tonemapping in avisynth then you won't even get those 5% because even more time will be used by your cpu.
Hardware decoding is a placebo effect in this case. Deal with it.

[almosely]

You are rude, disrespectful, short-sighted, conceited and unreasonable. I won't exchange another word with you.

[LigH]

Nevertheless, he is right. Compared to the efforts of a video encoder, decoding with or with out GPU support is a negligible difference.

Run your scripts through AVSMeter, compare these "benchmark" durations with the encoding duration: That ratio should tell you where the real bottleneck is.

[real.finder]

even if Atak_Snajpera is right, I think it's better to have both option just in case, also I think hardware decoder can help with weak cpu cases like laptop or low end or old hardware

Originally Posted by almosely

You are rude, disrespectful, short-sighted, conceited and unreasonable. I won't exchange another word with you.

So now we have some other opinions...

Either way, Atak has the knack of coming across like that, so props for you saying so, not many do !!!

[kedautinh12]

Originally Posted by Pauly Dunne

So now we have some other opinions...

Either way, Atak has the knack of coming across like that, so props for you saying so, not many do !!!

Long time to see, old man

Originally Posted by kedautinh12

Long time to see, old man

How did you find me

Yep, keeping an Ultra-low profile these days, kinda lost interest

You're not posting as often as you used to, either.

[tebasuna51]

Originally Posted by almosely

You are rude, disrespectful, short-sighted, conceited and unreasonable. I won't exchange another word with you.

Please guys remember the forum rule:

4) Be nice to each other and respect the moderator. Profanity and insults will not be tolerated. If you have a problem with another member turn to the respective moderator and if the moderator can't help you send a private message to Doom9.

There are 2 questions here:
1) Save time using GPU decode? A little but yes.
2) Last L-SMASH-Works-20230716 have a bug? I think so, then almosely is right and we can say thanks for the report.

1) I have a cheap Nvidia GeForce GT 1030 and running AvsMeter you can see the GPU decode is slow than with my CPU:

AviSynth+ 3.7.3 (r4003, 3.7, x86_64) (3.7.3.0)
Over "Roger Federer v Rafael Nadal Full Match Australian Open 2017 Final.mp4":
Number of frames: 304982
Length (hh:mm:ss.ms): 03:23:19.280
Frame width: 1920
Frame height: 1080
Framerate: 25.000 (25/1)
Colorspace: i420
                                       06/07/2023        NVIDIA GeForce GT1030
                                      FFVideoSource           DgSource 249
                                  --------------------   ---------------------
Frames processed:                 304982 (0 - 304981)    304982 (0 - 304981)
FPS (min | max | average):        281.3 | 1807 | 960.5   149.9 | 489.9 | 474.5
Process memory usage (max):       190 MiB                335 MiB
Thread count:                     25                     17
CPU usage (average):              74.5%                  8.6%

GPU usage (average):              2%                     57%
VPU usage (average):              0%                     74%
GPU memory usage:                 402 MiB                459 MiB
GPU Power Consumption (average):  13.6 W                 15.9 W

Time (elapsed):                   00:05:17.510           00:10:42.809

But when I recode the file with x264 (crf 20) the DgSource avs is fast than FFVideoSource with the same stats (see log files):

DgSource encoded 304982 frames, 59.71 fps, 4513.20 kb/s (5107 sec)
FFVideoSource encoded 304982 frames, 57.89 fps, 4513.20 kb/s (5269 sec)
It is only 2:42 over 1:27:49 encode but maybe with a better GPU can go until the 5:17 used by the CPU decoder.

2) But when I test L-SMASH-Works-20230716 I see than there are something wrong:

                                           LSMASHVideoSource 15/07/2023
                                        ----------------------------------
                                        prefer_hw=0            prefer_hw=1
                                    ---------------------  ---------------------
Frames processed:                   304982 (0 - 304981)    304982 (0 - 304981)
FPS (min | max | average):          45.87 | 2711 | 89.62   26.73 | 580.8 | 80.34
Process memory usage (max):         236 MiB                225 MiB
Thread count:                       26                     16
CPU usage (average):                19.6%                  7.6%

GPU usage (average):                2%                     15%
VPU usage (average):                0%                     23%
GPU memory usage:                   373 MiB                399 MiB
GPU Power Consumption (average):    13.7 W                 14.2 W

Time (elapsed):                     00:56:42.960           01:03:16.199

The full encode (prefer_hw=1) have different stats (see the log) and:
LSMASHVideoSource encoded 304982 frames, 44.84 fps, 5384.67 kb/s (6801 sec, 1:53:21)
And is wrong, the prefer_hw=0 play fine and with same stats than both before(slow than DgSource).

[kedautinh12]

Try test ver
https://github.com/HomeOfAviSynthPlu...ent-1665742509

[tebasuna51]

20230611 version is also wrong
20230507 work fine with prefer_hw=1
20230804 test version work fine now