Guida a Logitech Media Server, Squeezelite e derivati.

[antonellocaroli]

folks,

as a new user of LMS/Squeezelite, I am still learning and trying to find the right configuration for my setup. My Dac is optimized for native dsd256 playback and hence I would like to up-sample everything to native dsd256 using C-3PO plugin. The server that runs LMS is Intel NUC8I7BEH. I am finding that there are drop outs with certain sample rate and flac file format - the music plays for a second or two and then stop and then plays again then stops and so on...The drop outs happens with 88.2Khz, 96KHz, 192KHz in flac formats only but the same sample rate works fine with other formats, e.g. 192K wav plays fine.

All these sample rate works fine when played through HQplayer with poly-sinc-xtr-mp or poly-sinc-ext2 filter setting which are very compute intensive. I am wondering if sox is more demanding of cpu ? When I play them, I see sox sometimes consuming 30-40% cpu on one core. I tried with dsd64 as well but to no avail and even tried the quality setting in C-3PO to "quick" or "low" but same result. Is there a better configuration to upsample in native dsd256 ?

Appreciate any help and thanks in advance.

Hi,

you have active in advanced > C3PO the multithread ?

[devg]

Hi,

you have active in advanced > C3PO the multithread ?

Yes, multithread is selected.

[devg]

Hi,

you have active in advanced > C3PO the multithread ?

Yes, multithread is selected.

I followed @antonellocaroli configuration which lead me to reconfigure the buffer size. By default its 128 but when multi-threading is enabled, it seems like the suggested value is 32. With 32 it didn't work either and the drop outs was still there. I then reduced it to 24 and viola - it started working. All the sample rates that I couldn't play without drop outs can be played now.

Does the buffer size has anything to do with the amount of free ram available in the system ?

[UnixMan]

Yes, multithread is selected.

Try to disable it. In many cases it does more harm than good.

[devg]

Try to disable it. In many cases it does more harm than good.

Disabling or enabling doesn't seem to have any effect on the drop outs. The buffer size still needs to be 24 (in my case) in order to avoid it.

[marcoc1712]

I followed @antonellocaroli configuration which lead me to reconfigure the buffer size. By default its 128 but when multi-threading is enabled, it seems like the suggested value is 32. With 32 it didn't work either and the drop outs was still there. I then reduced it to 24 and viola - it started working. All the sample rates that I couldn't play without drop outs can be played now.

Does the buffer size has anything to do with the amount of free ram available in the system

sure it has.

If you go low in memory the system start to paginate and this could cause sox to work badly, BUT the value is in megabytes, do you have less than 32 MB of free RAM in your system? That's no good!

If you are low in memory you have several settings in LMS to take care of, but as a general rule, you should have at least 1GB of free RAM for a basic usage, at least double with DSD and same for upsampling. If I were you, I'll go with at least 4GB with your requirements. Never mind about sox quality, always use very hight and nothing less.

Hope squeezelite is running in another system, in this case you could increase (input) buffer sizes, then in LMS -> Settings-> Player -> Audio delay increase a bit the value. If not... You need a very powerfull system with a lot of RAM: with DSD256 the minimum buffer size is 512Mbx 2, ideal is 2GB x 2, that means 8GB total RAM could be enough, but 16 are much better.

[devg]

sure it has.

If you go low in memory the system start to paginate and this could cause sox to work badly, BUT the value is in megabytes, do you have less than 32 MB of free RAM in your system? That's no good!

If you are low in memory you have several settings in LMS to take care of, but as a general rule, you should have at least 1GB of free RAM for a basic usage, at least double with DSD and same for upsampling. If I were you, I'll go with at least 4GB with your requirements. Never mind about sox quality, always use very hight and nothing less.

Hope squeezelite is running in another system, in this case you could increase (input) buffer sizes, then in LMS -> Settings-> Player -> Audio delay increase a bit the value. If not... You need a very powerfull system with a lot of RAM: with DSD256 the minimum buffer size is 512Mbx 2, ideal is 2GB x 2, that means 8GB total RAM could be enough, but 16 are much better.

Hi Marco,

I appreciate your insight and thanks for your reply.

I have 8Gb RAM installed on the server and Squeezelite runs on another low powered NUC. Here is the output from free and as you can see there is enough memory available ~ 3.5Gb.

[root@audiolinux audiolinux]# free -h
total used free shared buff/cache available
Mem: 7.7Gi 4.0Gi 1.5Gi 1.0Mi 2.2Gi 3.5Gi
Swap: 0B 0B 0B


On the squeezelite, I am using -b 2097152:2097152 option as well. However I can't find "audio delay" settings in LMS -> Settings-> Player. Can you be little more specific ?

Also earlier you said "HF before resampling, that will be an overhead of your CPU." Can you please elaborate and how to enable the setting ?

[marcoc1712]

Settings->Player->Audio-> Audio Startup Time. start with 3-5 secs, then increase if necessary.


The last group of settings in C-3PO is DSD input Lowpass filter.



The default is a first order low pass at 15Khz, you could model a different one using up to 4 steps as you like.




I could not remember the optimal values for your case (I'm not using DSD upsampling), Paolo (Unixman) could sure give you some better advice on that matter.


Little buffers mean shorter CPU high load but more interrupts, then higher CPU usage in the long term. It's a trade-off, but filtering out HF should reduce CPU usage.

[devg]

Settings->Player->Audio-> Audio Startup Time. start with 3-5 secs, then increase if necessary.


The last group of settings in C-3PO is DSD input Lowpass filter.



The default is a first order low pass at 15Khz, you could model a different one using up to 4 steps as you like.




I could not remember the optimal values for your case (I'm not using DSD upsampling), Paolo (Unixman) could sure give you some better advice on that matter.


Little buffers mean shorter CPU high load but more interrupts, then higher CPU usage in the long term. It's a trade-off, but filtering out HF should reduce CPU usage.

Got it. Thank you.

I add HF low pass filters and that indeed seems to reduce the cpu load.

However, the "Audio Startup Time" didn't help with the drop outs. I also bought 16Gb ram for the server and that didn't help either. I still have drop outs for buffer size > 24 but as I said earlier, these only happen with 96k, 192k flac files up-sampled to DSD128 or DSD256. I am trying to understand why it is the case. Any hints ?

[marcoc1712]

Got it. Thank you.

I add HF low pass filters and that indeed seems to reduce the cpu load.

However, the "Audio Startup Time" didn't help with the drop outs. I also bought 16Gb ram for the server and that didn't help either. I still have drop outs for buffer size > 24 but as I said earlier, these only happen with 96k, 192k flac files up-sampled to DSD128 or DSD256. I am trying to understand why it is the case. Any hints ?

could you please post the command used when drop outs are produced? (Settings. Player, C-3PO, last executed command).