Is it Possible ?

[jschmoyer]

Hello Claire or Cedric,

Is it possible for to make ConvertXtoVideo run on all processors on a multiprocessor PC?
On my PC currently, it loads Cpu1 (all real cores and all hyper-cores) as it should in the 30-80% usage while working, but the program totally ignores the second cpu (less than 4% usage of any of the real or hyper-cores).

Your program works fine and converts most MKVs to Blurays at just under 3x realtime, I just hate to see that second cpu wasted, because if used as well as the first cpu it would/should cut conversion time to 5x+ realtime.

I'm sure you have more pressing issues, so I'm hoping it's something as simple as turning on a flag or adding to the start command line.

[JJ]

There has been many posts about this before.
In short - it is not possible.
You can use only one cpu when encoding a video as it is continuous stream.
Another cpu can be used while decoding. Usually decoding is much faster than encoding, so it looks like only one cpu is in use.
Video processing uses a lot of memory and read/writes to hd, and if system is not fast enough then cpu simply waits for reading/writing.

You can use more cpus when you convert more than one video at once. Or you can start several different conversions.

[jschmoyer]

Interesting, but I don't think that is correct.
I'm watching Nero Video transcoding an mp4 to BluRay right now and it is using all of the 12 cores and 12 hyper-cores of cpu1 and also the first 4 cores and first 4 hyper-cores of cpu2.
It would appear that 1 (real) core on each cpu is coordinating, as each cpu has single core that stays steady at 40-50% usage.

It seems to be limited to 32 total cores, and it is not using the cores as efficiently as ConvertXtoVideo since it is only loading the cores to 15-25%, BUT it is most definitely using both cpus.
Stopping Nero Video immediately unloads those cores on both cpus.

I don't believe there are any read/write problems as this is an all SSD machine running in RAID 0.
No other software is running on this machine (beyond the usual background OS stuff).

[JJ]

How long does it take to convert with Nero? How long does it take to convert that same project with CxV?

[jschmoyer]

Sorry for the late reply, I've been busy with the holidays.
I think this is the info you asked for.

I started with a MP4 file 01:39:29 length (1.78gb) encoded @ 1920x1080 2433kbps 24fps video, 2ch. audio encoded @ 125kbps 44khz
I asked both Nero Video 2017 and ConvertXtoVideo to produce a BluRay with single item basic menu.
Both programs were set for 100% quality and 2 pass, with BluRay target size set @ 46gb
All monitoring was done with Windows Task Manager , Windows Resource Monitor and Core Temp.
The programs produced the following-

Nero Video 2017 - main video file - 01:39:29 length (21.1gb) encoded @ 1920x1080 40000kbps 24fps video, 2ch. audio encoded @ 192 kbps 48khz
All 48 cores were used - NUMA node 0 (24 cores) were used @ an avg. 20%, NUMA node 1 (24 cores) were used @ an avg. 40% (with 1 core @ a constant 85%)
Windows registered this usage @ a total avg. of 26%.
Nero Video 2017 estimated the size of the BluRay @ 22.5gb and said it would encode @ 40000kbps and the end result was 21.1gb
Transcoding took 01:07:10, 1.48x realtime

ConvertXtoVideo - main video file - 01:39:27 length (2 seconds lost somewhere)(10.3gb) encoded @ 1920x1080 14710kbps 24fps video, 2 ch. audio encoded @ 224kbps 48khz
Only 24 cores were used (ConvertXtoVideo was set for 32 cores) - NUMA node 0 (24 cores) were used @ an avg. 65%, NUMA node 1 (24 cores) not used.
Windows registered this usage @ a total avg. of 33%.
ConvertXtoVideo estimated the size of the BluRay @ 21.2gb and said it would encode @ 30000kbps and the end result was 10.3gb (due to ConvertXtoVideo encoding at half the bitrate promised).
Transcoding took 00:29:55, 3.33x realtime

[JJ]

CPU cores and GPU cores are two different things. Don't confuse them.

Did you check quality of output? Any visual differences?

[jschmoyer]

CPU cores and GPU cores are two different things. Don't confuse them.

Of course they are, why are you even bringing that into the conversation, when it has nothing to do with my question?

Did you check quality of output? Any visual differences?

There is a slight quality difference in detail, best noted in high movement scenes and very dark scenes.
Nero video has the edge in quality, but that is to be expected due to the higher encode rate.
It is not noticeable unless you do a side-by-side comparison and really isn't an issue.

I came here with a simple question - "Is it possible for to make ConvertXtoVideo run on all processors on a multiprocessor PC?"
You answered with some nonsense - "You can use only one cpu when encoding a video as it is continuous stream."
Nero Video uses both cpus to transcode, but not very efficiently. I only picked it as an example because it can generate BluRay structure with the transcoded video, just like ConvertXtoVideo, so a direct comparison of cpu usage was possible.
I did not even mention HandBrake, because all HandBrake does is transcode, but it uses all available cpu cores (in my case 48) and at 80%+.

Why you are seeking to distract from my original question is a mystery to me, but it is becoming obvious you are not the person to answer my original question and also that you don't have any direct experience with multi-processor PCs.
At this point, I'm not even sure you understand how a multicore processor interacts with the OS.
A high post count makes you an authority on nothing but posting, so I'm done responding to your distractions from my question.

If Claire or Cedric would care to respond to my original question - "Is it possible for to make ConvertXtoVideo run on all processors on a multiprocessor PC?" - I would be grateful.
Either way I will remain a VSO user as I enjoy the products, am satisfied with the visual output, and appreciate all the hard work that goes into them.

[felicia]

Hi,
Sorry for the delay.

As you can see

all CPUs are busy with this conversion from Blu-Ray to MKV.

Have you set the number of cores used for encoding & decoding in the conversion settings?
If you remux the video far less cores will be used.
Dependening on which hardware acceleration you use CPU can be nearly ignored.

I hope this answers your question.

BR,

Felicià

[jschmoyer]

Felicia,

In the Conversion settings, encoding and decoding cores are both at set to 32 (the max available).
Can this be increased?
Changing this setting to a value above 32 in the registry has no effect, as it always reverts back to 32.
Also in the registry,this setting indicates threads, while in the program's "Settings" this indicates cores?
Which is it really, threads or cores?

But I believe you've misunderstood the original question.
The problem is this is a multi-processor workstation, it has two (2) 12 core CPUs.

ConvertXtoVideo only uses 1 of the CPUs, not both, when encoding OR transcoding.
I am trying to find out if ConvertXtoVideo is (or can be) SMP aware (perhaps through the command line ?) so that it will use both CPUs( and thus all available cores).
Right now ConvertXtoVideo ignores the second CPU (with it's additional 12 cores), as if it does not see it.
Multiple CPU (SMP) encoding and transcoding is absolutely possible, as the "Handbrake" program does use all CPUs and @ 80% usage (which on this workstation that is 2 CPUs (Intel), 24 cores and 48 threads)

As an update, using v2.0.0.60 -
Transcoding a TS file to BluRay, now 1 CPU (all cores) are used @ 65%, and the 2nd CPU (all cores) are used @ 15%.
Transcoding a MKV file to BluRay, now 1 CPU (all cores) are used @ 70%, and the 2nd CPU (all cores) are used @ 10%.
Transcoding a MP4 file to BluRay, now 1 CPU (all cores) are used @ 65%, and the 2nd CPU (all cores) are used @ 12%.

[cedric]

Hello jschmoyer,

Video decoding/encoding tasks can be distributed to all CPU cores only if CPU decoding/encoding is used.
I mean if you enabled hardware optimizations, CPU cores will only be used for audio chain process (and audio treatment can only be done by 1 core at a time).

Can you give us your hardware optimization settings? Or maybe send us your last log file...

Regards,

[jschmoyer]

Hello jschmoyer,

Video decoding/encoding tasks can be distributed to all CPU cores only if CPU decoding/encoding is used.
I mean if you enabled hardware optimizations, CPU cores will only be used for audio chain process (and audio treatment can only be done by 1 core at a time).

Can you give us your hardware optimization settings? Or maybe send us your last log file...

Regards,

Again I believe you are making an assumption that doesn't ring true.
I have tried both with and without hardware optimization, nothing changes.
In both instances (with or without HO), kernel time, core usage and NUMA node usage remains the same.
This also applies simultaneous conversions, 1-12 it makes no difference in kernel time, core usage and NUMA node usage.
I have not found any setting that affects core usage, 1 cpu is always used much, much more that the other and the % usage is always as I quoted in my prior post.

Attached is the log file for the settings - 32 core decode, 32 core encode, 1 simultaneous conversion, hardware optimization OFF

I see nothing in the file pertinent to the problem, beyond the fact that the Xeon cpus are mis-identified as a single 32/16 (?) unit, when there are actually 2 12core/24thread units.

[cedric]

Hello,

Can you try by disabling all audio and subtitle tracks?
I would like to know if the speed is not lowered by audio/subtitle processins...
Another question: is your source/destination folders are on a quick drive (not a network or USB stick)
Did you tried to change the cor(s) setting to 24?

Regards,

[jschmoyer]

Hello,

Can you try by disabling all audio and subtitle tracks?
I would like to know if the speed is not lowered by audio/subtitle processins...

Encoding takes less time and kernel time, core usage and NUMA node usage go down across all cores.

Another question: is your source/destination folders are on a quick drive (not a network or USB stick)
Did you tried to change the cor(s) setting to 24?

Changing from 32 to 24 cores simply lowers kernel time, core usage and NUMA node usage, but it does not change encode time, since the cores are not being used at anywhere near capacity.

Simple question - Is the program SMP capable or not ??
If it isn't, this is all a waste of time.