HQPlayer Linux Desktop and HQplayer embedded

[Louie]

30 minutes ago, Miska said:

 

Yes, it should be fine. You can check the web interface while playing that the offload becomes active.

 

Did you reboot after doing that? Does you GPU appear under "/proc/driver/nvidia" ? If not, try doing "sudo modprobe nvidia" first.

 

I'm assuming here that you are using Ubuntu Server with text mode console. If necessary, add "nomodeset" to kernel command line in /etc/default/grub and run "sudo update-grub".

@MiskaI can see the nvidia entries under the /proc/driver/nvidia. Moreover, if I use "sudo lsmod | grep nvidia" and I can see the below output:

nvidia_uvm              1028096  0
nvidia_drm               57344  0
nvidia_modeset       1183744  1 nvidia_drm
nvidia                       34889728  2 nvidia_uvm,nvidia_modeset
drm_kms_helper        184320  2 nvidia_drm,i915
drm                          487424  4 drm_kms_helper,nvidia_drm,i915

 

Yes, I am using Ubuntu server 20.04 LTS with text mode console, and I have also added the "nomodeset" to the kernel command and reboot.

 

Now, I am using the Filter=poly-sync-ext3 Modulator=ADSM5EC, DSD256 to upscale my 44.1 wav files. However, I still cannot see any status output at the main page offload tab. Thus, I am not sure whether HPQe is using GPU or not.

 

Is it anythings I am missing?

 

Thanks,

 

[michaelvv]

A more optimized install script , thanks to jussi.

It can easily be better and for those who worries about disk usage

this takes up , 3426M

 

#!/bin/bash

##########################
### updated 23/05-2021 ###
##########################

############################################################
### Might still be some packages installed which are not
### necessary for hqplayer , but not that many...
############################################################

### this is for the Intel version
hq_ver=hqplayerd_4.23.1-71_amd64.deb

# Amd    - hqplayerd_4.23.1-72amd_amd64.deb
# Intel  - hqplayerd_4.23.1-71_amd64.deb

libgmpris_ver=libgmpris_2.2.1-8_amd64.deb
hqplayer_user=admin
hqplayer_password=password
hqplayer_folder=https://www.signalyst.eu/bins/hqplayerd/focal/


### setting timezone
timedatectl set-timezone Europe/Copenhagen

sudo apt update
sudo apt -y upgrade

sudo apt -y install net-tools cifs-utils libtool wavpack zip unzip pkg-config gnupg


### AMD Rocm ###

cd /tmp;wget -q -O - https://repo.radeon.com/rocm/rocm.gpg.key  | sudo apt-key add -
echo 'deb [arch=amd64] https://repo.radeon.com/rocm/apt/debian/ xenial main' | sudo tee /etc/apt/sources.list.d/rocm.list


sudo apt update
sudo apt -y install rocm-libs

### mostly hqplayer stuff ###

cd /tmp; wget https://www.sonarnerd.net/src/focal/$libgmpris_ver
cd /tmp; sudo dpkg -i $libgmpris_ver

cd /tmp; wget $hqplayer_folder$hq_ver
cd /tmp; sudo dpkg -i $hq_ver


sudo apt -y --fix-broken install

sudo systemctl enable hqplayerd

sudo systemctl start hqplayerd
sudo hqplayerd -s $hqplayer_user $hqplayer_password

### cleanup.

### jussi's recommended cleanup ###
sudo apt install -f ; sudo apt autoremove --purge ; sudo apt clean

reboot

 

 

[lpost]

3 hours ago, Miska said:

 

OK, I think I'll need to stick with "build 2" method in future.

 

There shouldn't be any audible difference because both produce same output data. But CPU load with "build 2" can be higher since it contains less optimizations.

 

Thanks. I'll have a closer listen today. So far so good with ext3 at PCM768k. Too much for CPU at 1.5M.

[Miska]

8 minutes ago, lpost said:

Thanks. I'll have a closer listen today. So far so good with ext3 at PCM768k. Too much for CPU at 1.5M.

 

I've just updated the 4.23.1 build (3) to be similar to the "AMD optimized" Ubuntu build. Should give good performance, but requires AVX2 instruction set support.

 

[lpost]

15 minutes ago, Miska said:

 

I've just updated the 4.23.1 build (3) to be similar to the "AMD optimized" Ubuntu build. Should give good performance, but requires AVX2 instruction set support.

 

I've give it a go. My i7-6770 from 2016 should have AVX2. So far so good with build (3). Ext3 definitely gets further before a dropout but still occurs every 25 seconds or so at 1.5M but runs perfectly at 768k. No white noise!

 

Is ext3 at 768k technically 'better' than ext2 at 1.5M? I'm not sure I hear an appreciable difference with 15 mins. listening time.

[lpost]

I'm thinking of building a new HQPe OS host. I know the processor choice is a complicated one balancing value, availability, power consumption. I'm looking at either i7 9700K or AMD Ryzen 7 2700X or in these families. ~$300 USD.

 

At this generation of processor specifically for HQPe OS x64/AMD, is Intel preferred or AMD? or is something in the 10th generation a better choice, like 10th Gen Intel® Core™ i5-10400?

 

Any other advice/experience of which processor to pick as of May, 2021? I don't mind spending money but I don't want a high TDP CPU just to achieve DSD512 either. I mostly stick to PCM-PCM at 1.5M.

 

(My old machine won't handle ext3 at 1.5M...and it's time to replace, or start planning anyway)

[sledwards]

1 hour ago, lpost said:

 

(My old machine won't handle ext3 at 1.5M...and it's time to replace, or start planning anyway)

I am observing similar performance with ext3. I get periodic drops using it with PCM1.54, but it works flawlessly with DSD1024. Maybe Jussi has a theory as to why this happens with ext3.

[lpost]

4 hours ago, lpost said:

So far so good with build (3). Ext3 definitely gets further before a dropout but still occurs every 25 seconds or so at 1.5M but runs perfectly at 768k. No white noise!

 

My ears are preferring 1.5M ext2 PCM vs. 768k ext3 PCM. The stage loses some life, some width and density stepping down to 768k.

 

I've had one instance where build (3) didn't play properly. It wasn't the same white noise but clearly not right, unlistenable. Stop, restart play resolved it.

[Louie]

On 5/23/2021 at 10:24 PM, Miska said:

 

Yes, it should be fine. You can check the web interface while playing that the offload becomes active.

 

Did you reboot after doing that? Does you GPU appear under "/proc/driver/nvidia" ? If not, try doing "sudo modprobe nvidia" first.

 

I'm assuming here that you are using Ubuntu Server with text mode console. If necessary, add "nomodeset" to kernel command line in /etc/default/grub and run "sudo update-grub".

Finally, found out the issue, it seem Nvidia drivers v465 cannot be supported and I need to uninstall and install 460. After that, I can see the offload status.

 

Now, I can see hyplayerd  is using GPU by checking with nvidia-smi

 

Thx

[Miska]

8 hours ago, Louie said:

Finally, found out the issue, it seem Nvidia drivers v465 cannot be supported and I need to uninstall and install 460. After that, I can see the offload status.

 

Now, I can see hyplayerd  is using GPU by checking with nvidia-smi

 

Is this with hqplayerd 4.23.1 or something else? Because v465 is what Nvidia ships for the CUDA toolkit used to build 4.23.x.

 

[Louie]

2 minutes ago, Miska said:

 

Is this with hqplayerd 4.23.1 or something else? Because v465 is what Nvidia ships for the CUDA toolkit used to build 4.23.x.

 

I am using 4.23.1.71.

[Miska]

9 minutes ago, Louie said:

I am using 4.23.1.71.

 

OK, strange... Maybe Nvidia messed up something in the driver... Which GPU do you have?

 

Usually it is the other way around, having too old driver means CUDA falls off.

 

[Louie]

2 minutes ago, Miska said:

 

OK, strange... Maybe Nvidia messed up something in the driver... Which GPU do you have?

 

Usually it is the other way around, having too old driver means CUDA falls off.

 

I am using GTX1060

[Miska]

On 5/23/2021 at 8:56 PM, lpost said:

Is ext3 at 768k technically 'better' than ext2 at 1.5M? I'm not sure I hear an appreciable difference with 15 mins. listening time.

 

No, if you use PCM, 1.5M is certainly better. ext3 is not any better than ext2, it is just matter of taste which one you prefer.

 

[sledwards]

I have noticed the size of my log file is becoming excessively large. These 3 lines keep repeating over and over with no user interaction with hqplayerd:

 

  2021/05/25 15:20:56 NAA output discovered 0 Network Audio Adapters
  2021/05/25 15:20:58 NAA output network Audio IPv6 support disabled
  2021/05/25 15:20:58 NAA output discovery from 0.0.0.0
 

Any way to suppress this without actually turning off the log file?

[Outlaw]

On 5/22/2021 at 3:01 AM, perpetualapprentice said:

Followed this advice, triple checked repository list, tried various alternatives, still no success. Web searches for "error while loading shared libraries" is a nice learning experience. Unfortunately, I'm unable to devote more time to this right now.

 

Reverted to hqplayerd_4.22.3-67_amd64 - Everything related to hqplayerd completely purged, "new" install succeeded on first attempt & running without issue.

 

Will survive listening to the fantastic poly-sinc-ext3 via modest PCM upsampling my M1 mini for now.

Me to.Keep having same issues cant seem to get help.Back to 4.22.3_67 running

[Outlaw]

54 minutes ago, Outlaw said:

Me to.Keep having same issues cant seem to get help.Back to 4.22.3_67 running

Got it to work.Had to do fresh install with 4.22.67 then update to latest 4.23.1.71.For some reason was unable to do fresh install to latest version

[Miska]

5 hours ago, sledwards said:

I have noticed the size of my log file is becoming excessively large. These 3 lines keep repeating over and over with no user interaction with hqplayerd:

 

  2021/05/25 15:20:56 NAA output discovered 0 Network Audio Adapters
  2021/05/25 15:20:58 NAA output network Audio IPv6 support disabled
  2021/05/25 15:20:58 NAA output discovery from 0.0.0.0
 

Any way to suppress this without actually turning off the log file?

 

This happens when you have NAA configured as output, but NAA is not available (turned off or similar).

 

Solution is to either turn off logging, turn off HQPlayer machine when not in use (log resides by default in /tmp which should get cleared on reboot), or leave NAA running so that HQPlayer doesn't need to keep trying to reconnect to it all the time.

 

[Miska]

There is now version 4.23.2 available for Ubuntu with AMD GPU support tested on 6700XT. Currently only poly-sinc filters supported on AMD GPUs, there are some problems with other algorithms on the these newer GPUs so those are disabled for now.

 

[audiofool]

@Miska

Love the new ext3 filter - sounds great with PCM output for me.

 

I think something weird is going on with Matrix using hqplayer-embedded-4.23.1-x64

When playing 2 channel FLAC it is almost as if there is a slight reverb added.  If I take out the process files for the extra channels it sounds normal.  Log shows nothing happening with the extra channels, as expected.

 

Also am I correct in assuming the channel mapping are as follows:

0 left front   
1 right front  
2 centre       
3 LFE          
4 left back    
5 right back   
6 left side
7 right side

 

Do my mapping below seem sensible?  I have mapped the side channels since some 5.1 use side instead of back mapping.

 

This is the relevant config for PCM output

<matrix enabled="1" engine="1" expand_hf="0">
            <pipeline channel="0"  gain="-5.5"  mixdown="0" process="/tmp/hqplayer-eq.txt" source="0"/>
            <pipeline channel="1"  gain="-5.5"  mixdown="1" process="/tmp/hqplayer-eq.txt" source="1"/>
            <pipeline channel="2"  gain="-8.5"  mixdown="0" process="/tmp/hqplayer-eq.txt" source="2"/>
            <pipeline channel="3"  gain="-8.5"  mixdown="1" process="/tmp/hqplayer-eq.txt" source="2"/>
            <pipeline channel="4"  gain="-8.5"  mixdown="0" process="/tmp/hqplayer-eq.txt" source="3"/>
            <pipeline channel="5"  gain="-8.5"  mixdown="1" process="/tmp/hqplayer-eq.txt" source="3"/>
            <pipeline channel="6"  gain="-6.8"  mixdown="0" process="/tmp/hqplayer-eq.txt" source="4"/>
            <pipeline channel="7"  gain="-11.6" mixdown="1" process="/tmp/hqplayer-eq.txt" source="4"/>
            <pipeline channel="8"  gain="-11.6" mixdown="0" process="/tmp/hqplayer-eq.txt" source="5"/>
            <pipeline channel="9"  gain="-6.8"  mixdown="1" process="/tmp/hqplayer-eq.txt" source="5"/>
            <pipeline channel="10" gain="-6.8"  mixdown="0" process="/tmp/hqplayer-eq.txt" source="6"/>
            <pipeline channel="11" gain="-11.6" mixdown="1" process="/tmp/hqplayer-eq.txt" source="6"/>
            <pipeline channel="12" gain="-11.6" mixdown="0" process="/tmp/hqplayer-eq.txt" source="7"/>
            <pipeline channel="13" gain="-6.8"  mixdown="1" process="/tmp/hqplayer-eq.txt" source="7"/>
            <pipeline channel="14" gain="0.00"  mixdown="14" process="" source="14"/>

 

 

Also, any measurements on your Denafrips Aries, I see it on your hardware list?

 

 

 

 

 

 

 

 

 

 

[Miska]

3 hours ago, audiofool said:

I think something weird is going on with Matrix using hqplayer-embedded-4.23.1-x64

 

There are no changes in matrix things between 4.22 and 4.23, so it shouldn't be any different.

 

3 hours ago, audiofool said:

            <pipeline channel="6"  gain="-6.8"  mixdown="0" process="/tmp/hqplayer-eq.txt" source="4"/>
            <pipeline channel="7"  gain="-11.6" mixdown="1" process="/tmp/hqplayer-eq.txt" source="4"/>
            <pipeline channel="8"  gain="-11.6" mixdown="0" process="/tmp/hqplayer-eq.txt" source="5"/>
            <pipeline channel="9"  gain="-6.8"  mixdown="1" process="/tmp/hqplayer-eq.txt" source="5"/>
            <pipeline channel="10" gain="-6.8"  mixdown="0" process="/tmp/hqplayer-eq.txt" source="6"/>
            <pipeline channel="11" gain="-11.6" mixdown="1" process="/tmp/hqplayer-eq.txt" source="6"/>
            <pipeline channel="12" gain="-11.6" mixdown="0" process="/tmp/hqplayer-eq.txt" source="7"/>
            <pipeline channel="13" gain="-6.8"  mixdown="1" process="/tmp/hqplayer-eq.txt" source="7"/>

 

This looks strange. Why are all surround channels, both left and right, mixed to both left and right output channels? This would indeed explain reverb kind of effect.

 

[audiofool]

   <pipeline channel="6"  gain="-6.8"  mixdown="0" process="/tmp/hqplayer-eq.txt" source="4"/>
            <pipeline channel="7"  gain="-11.6" mixdown="1" process="/tmp/hqplayer-eq.txt" source="4"/>
            <pipeline channel="8"  gain="-11.6" mixdown="0" process="/tmp/hqplayer-eq.txt" source="5"/>
            <pipeline channel="9"  gain="-6.8"  mixdown="1" process="/tmp/hqplayer-eq.txt" source="5"/>
            <pipeline channel="10" gain="-6.8"  mixdown="0" process="/tmp/hqplayer-eq.txt" source="6"/>
            <pipeline channel="11" gain="-11.6" mixdown="1" process="/tmp/hqplayer-eq.txt" source="6"/>
            <pipeline channel="12" gain="-11.6" mixdown="0" process="/tmp/hqplayer-eq.txt" source="7"/>
            <pipeline channel="13" gain="-6.8"  mixdown="1" process="/tmp/hqplayer-eq.txt" source="7"/>

 

 

15 hours ago, Miska said:

 

There are no changes in matrix things between 4.22 and 4.23, so it shouldn't be any different.

 

 

This looks strange. Why are all surround channels, both left and right, mixed to both left and right output channels? This would indeed explain reverb kind of effect.

 

I was on 4.21 before, but I may not have noticed before.

 

The above matrix is something I was playing with based on one of the examples in RFC 7845 Section 5.1.1.5 and I expected some sort of reverb because of the mixing of the surround channels.  It's with 2 channel that I notice a difference between having an eq filter defined for only 2 channels or having a multichannel config.  I am assuming the hqplayer ignores the additional channels if the source only has 2 channels?  I have simply been deleting channels 2 to 9 when testing - maybe they have to be there?

I also tried the following matrix with the same result with 2 channel audio - it's subtle but sounds slightly wetter for lack of a better description.

            <pipeline channel="0"  gain="-11.5" mixdown="0"  process="/tmp/hqplayer-eq.txt" source="0"/>
            <pipeline channel="1"  gain="-11.5" mixdown="1"  process="/tmp/hqplayer-eq.txt" source="1"/>
            <pipeline channel="2"  gain="-14.5" mixdown="0"  process="/tmp/hqplayer-eq.txt" source="2"/>
            <pipeline channel="3"  gain="-14.5" mixdown="1"  process="/tmp/hqplayer-eq.txt" source="2"/>
            <pipeline channel="4"  gain="-14.5" mixdown="0"  process="/tmp/hqplayer-eq.txt" source="3"/>
            <pipeline channel="5"  gain="-14.5" mixdown="1"  process="/tmp/hqplayer-eq.txt" source="3"/>
            <pipeline channel="6"  gain="-8.8"  mixdown="0"  process="/tmp/hqplayer-eq.txt" source="4"/>
            <pipeline channel="7"  gain="-8.8"  mixdown="1"  process="/tmp/hqplayer-eq.txt" source="5"/>
            <pipeline channel="8"  gain="-8.8"  mixdown="0"  process="/tmp/hqplayer-eq.txt" source="6"/>
            <pipeline channel="9"  gain="-8.8"  mixdown="1"  process="/tmp/hqplayer-eq.txt" source="7"/>

 

 

[Miska]

6 hours ago, audiofool said:

I am assuming the hqplayer ignores the additional channels if the source only has 2 channels?

 

Yes, HQPlayer checks if there's data for the source channel. If source doesn't have data for some channels (like LFE missing), such processing is skipped.

 

[audiofool]

7 minutes ago, Miska said:

 

Yes, HQPlayer checks if there's data for the source channel. If source doesn't have data for some channels (like LFE missing), such processing is skipped.

 

Makes sense.  So nothing I do with the other pipeline definitions should have any influence on 2 channel files?

[audiofool]

One other note, these are 2 channel 44khz 16 bit wav files not flac as I thought, I wonder if the lack of channel layout unlike flac (ie. L R)  is an issue?