HQ Player

[jvvita]

If I understand it correctly, linux is performing better than windows with 4.14. But should there be any difference on performance on running hqplayer desktop on Ubuntu 21 and running hqplayer embedded bootable image?

[chipvn]

The 4.14.0 AMD desktop version works really well on Ubuntu 20.4.3 LTS low latency kernel.

On my passive HDPlex H3 V3 i7-9700 set max CPU speed at 4GHz, the ASDM5ECv2 load is very similar to the load of previous ASDM7EC on 4.13.1.

While I still can upsampling everything to DSD256x44.1 using ASDM5ECv2 with ext3 filter for 1x and poly-sinc-xtr-short-mp for Nx, the CPU load almost reach the 4GHz limit and the 70degC max CPU temperature I want to control also reach. If I want to use ASDM7ECv2, I might need to increase the CPU max speed to about 4.2GHz (I have not tried yet) and might need to accept higher temperature on the CPU.

 

Here is the system load and CPU temp after 1 hour of playing

[bogi]

8 hours ago, Luca72c said:

 

I upsample evetything to DSD256. No petformance issue before 4.14, my server has always been very fast and powerful, using 4.13.1 it copes very well with nearly all filter/modulator combos even with 4x frequencies files. My win10 installation is very basic and optimized, 1% or less total process load in idle, performance has always been even better than using HQP OS + Embedded... Before 4.14...

Then it seems I had bigger reserve on DSD128 than you on DSD256. The fixed 4.14 may be still more demanding on resources than the previous 4.13 version.

[Miska]

2 hours ago, jvvita said:

If I understand it correctly, linux is performing better than windows with 4.14. But should there be any difference on performance on running hqplayer desktop on Ubuntu 21 and running hqplayer embedded bootable image?

 

If you compare equivalent versions; generic <-> generic and amd <-> amd, not much. Reliability and performance consistency is better on HQPlayer OS as it has less things running and is much more streamlined.

 

If you run on "big iron", then the difference is likely smaller than when you have things pushed to the edge.

 

On Ubuntu you trade this reliability and streamlining to get support for GPU offload and flexibility.

 

Note that Ubuntu 21.x is not officially supported, I support only LTS releases (22.04 will be supposedly next). But likely things work also on the short term Ubuntu releases.

 

[jvvita]

Thanks. Ubuntu 20 does not recognize my network card. It works flawlessly on 21

[The Computer Audiophile]

2 minutes ago, jvvita said:

Thanks. Ubuntu 20 does not recognize my network card. It works flawlessly on 21

which NIC are you using?

[bogi]

I found [x] Multicore DSP full (not half) setting has now clear advantages in my i7 6820HQ Win10 case. I don't know how my experience applies to others on Windows, especially if your CPU is not Intel, but in my case:

• Processor load is really LOWER in Multicore DSP full setting (I am still curious how is that possible)
• No dropouts on filter/modulator combinations where my setup is prone to dropouts without Multicore DSP
• ... and what is last but not least ... trebles are less offensive on louder passages, I can easier rise volume, I hear more detail in upper half of frequency range, so I am getting feeling of cleaner sound  (maybe more randomized noise in Multicore DSP scenario ??)

That's when playing from notebook directly - not the recommended setup, but I'm at least playing from RAM disk and using iSilencer to deal with computer noise and using LPS on VBUS of DAC USB input.

[Miska]

1 hour ago, jvvita said:

Thanks. Ubuntu 20 does not recognize my network card. It works flawlessly on 21

 

Note that you can install Canonical's "hwe" (HardWare Enablement) kernel branch which keeps you on their latest short term kernel on an LTS system. Sometimes the installer of later point releases do this automagically.

 

[jvvita]

1 hour ago, The Computer Audiophile said:

which NIC are you using?

Realtek 2.5G. That comes with the Asus Z-590 Prime. 

 

15 minutes ago, Miska said:

 

Note that you can install Canonical's "hwe" (HardWare Enablement) kernel branch which keeps you on their latest short term kernel on an LTS system. Sometimes the installer of later point releases do this automagically.

 

Don't really know how to do it

[AudioDoctor]

8 hours ago, Hammer said:

Hi, any thoughts on why I cannot connect to HQPlayer either with Client or Roon?  Thanks!

 

Desktop or embedded?

[camott]

On 11/4/2021 at 11:11 AM, MikePid said:

Running ASDM7ECv2 with gauss-long on Mac mini M1 (Big Sur). At 1x main cores at 90%.  At 2x main cores are like 95%, but seems to be fine so far. (It’s a dedicated HQP machine)

 

In the exact same setup, I get the dropouts with Nx every 10 to 20 seconds. 1x is fine. Are you actually able to run ASDM7EC2v2 with gauss-long from 96k to 48x256 without dropouts?

[Hammer]

9 minutes ago, AudioDoctor said:

 

Desktop or embedded?

I’m trying to connect to HQPlayer Desktop.  The Windows 10 pc HQP Desktop is installed on has internet and HQP Desktop can see the NAA that’s on my network, but neither Roon Server nor HQP Client, both of which are installed on the same PC, can see HQP Desktop.  Very strange!  I’ve also tried connecting with Windows Firewall disabled without success.  Thank you!

[AudioDoctor]

12 minutes ago, Hammer said:

I’m trying to connect to HQPlayer Desktop.  The Windows 10 pc HQP Desktop is installed on has internet and HQP Desktop can see the NAA that’s on my network, but neither Roon Server nor HQP Client, both of which are installed on the same PC, can see HQP Desktop.  Very strange!  I’ve also tried connecting with Windows Firewall disabled without success.  Thank you!

 

On the top, all the way to the right is a button that has to be enabled to allow netwoprk control of HQP Desktop. enable that and see if it works.

 

edit: If you're trying to use the same PC to control them, don't use the IP address, use "localhost"

 

[Miska]

1 hour ago, jvvita said:

Realtek 2.5G. That comes with the Asus Z-590 Prime. 

 

Don't really know how to do it

 

This gives you their latest lowlatency kernel for Ubuntu 20.04:

sudo apt install linux-lowlatency-hwe-20.04

 

It will then stay up to date with system updates.

 

[Miska]

Here's now third attempt to build fast version for Windows:

https://www.signalyst.eu/bins/HQPlayer4Desktop-setup-4140-3.exe

 

For this one I actually threw out bunch of €€€ stuff.

 

Note! Remember to uninstall previous before installing these!

[jvvita]

17 minutes ago, Miska said:

It will then stay up to date with system updates.

Thank you so very much. But now I have to figure out a way to install the kernel without network access...

[bogi]

1 hour ago, Miska said:

Here's now third attempt to build fast version for Windows:

https://www.signalyst.eu/bins/HQPlayer4Desktop-setup-4140-3.exe

 

For this one I actually threw out bunch of €€€ stuff.

 

Note! Remember to uninstall previous before installing these!

Miska, Thanks for another try!

On my PC, 88.2k -> 5.6M closed-form-fast + ASDM7ECv2, Bauer Crossfeed + Loudness, no other DSP, without other programs running.
Without Multicore DSP the CPU load seem to be similar, about 50% with occasional dropouts.

With Multicore DSP full setting:

The -2 version uses less virtual cores and causes lower overall load about 25% of my CPU - with the same settings. That's what I repeatedly wondered.

The -3 version uses more virtual cores than the -2 version. Overall CPU load is higher, about 40% in my case. But maximum load in these cores is lower than in -2. That makes sense. Since individual virtual cores run on lower load, maybe heavier filters could be used. Just now it is too late in Europe, but tomorrow I will experiment with heavier filters than closed-form-fast. Since I borrowed my Topping E50, currently I cannot try DSD256.

See the difference in Multicore DSP between -2 and -3 version on pictures - the same settings were used:

 

4.14-2:

 

4-14.3:

 

Is the difference between -2 and -3 based on different compiler switches, or could be altered (for example by a checkbox) in HQPlayer settings? Because both -2 and -3 seem to have some advantages. -2 is lower on overall load in full MultiCore DSP mode and -3 is lower in individual core load in full MultiCore DSP mode.

[Miska]

30 minutes ago, bogi said:

Is the difference between -2 and -3 based on different compiler switches, or could be altered (for example by a checkbox) in HQPlayer settings? Because both -2 and -3 seem to have some advantages. -2 is lower on overall load in full MultiCore DSP mode and -3 is lower in individual core load in full MultiCore DSP mode.

 

Please also try the auto-configuration (box grayed) which should give most optimal results always. That way HQPlayer can decide how much of the work to split based on detected hardware properties, without going between all-or-nothing extremes.

 

The difference between -2 and -3 could be made configurable, but I'd rather at least make decision about sensible default. I would bury such setting to the advanced things tuned through environment variables.

 

I did some testing of -3 on my development laptop with Ryzen 9 5900HS and Mobile RTX3060 and have stable DSD256 output with poly-sinc-gauss-long + ASDM7EC modulator. Multicore DSP grayed (auto).

 

[Miska]

1 hour ago, jvvita said:

Thank you so very much. But now I have to figure out a way to install the kernel without network access...

 

For such cases I have a simple cheap gigabit USB-Ethernet dongle with old enough chip. I can then plug it in to have internet access in such cases.

 

Another very viable option for bleeding edge hardware is to use bleeding edge OS instead, that would be Fedora. It stays up very current.

 

[MikePid]

4 hours ago, camott said:

 

In the exact same setup, I get the dropouts with Nx every 10 to 20 seconds. 1x is fine. Are you actually able to run ASDM7EC2v2 with gauss-long from 96k to 48x256 without dropouts?

My DAC doesn’t do 48k well, and I just realized I do get dropouts (rate changing 96k to 44x256). 44k and 88k are fine though. I’m trying sinc-gauss now and it seems to be better for 96k. I’m still playing around with which filters I like better with ECv2, or even if I prefer it. 
 

I love the power and simplicity of Mac mini M1 for HQP. No tweaking or optimizing needed as long as you leave the stuff off that isn’t needed. It can handle more than enough filters for me with the ECs and DSD256. 

[bogi]

6 hours ago, Miska said:

Please also try the auto-configuration (box grayed) which should give most optimal results always. That way HQPlayer can decide how much of the work to split based on detected hardware properties, without going between all-or-nothing extremes.

Using the same setup as in my previous post:
Half ticked (grayed): CPU load between 35 - 40%. I tried to start Firefox to test how much reserve it has against dropping. There were drops during Firefox startup, but then it stabilized and plays without dropouts.

Unticked: Prone to dropouts with no other programs running, about 34% CPU load

On the picture: left grayed, right unticked

6 hours ago, Miska said:

The difference between -2 and -3 could be made configurable, but I'd rather at least make decision about sensible default. I would bury such setting to the advanced things tuned through environment variables.

I would be glad to have possibility to choose using an env. variable.

On my setup with -2 and -3 I am still preferring Multicore DSP fully ticked from the reasons I wrote yesterday: Lower load overall or per core with more reserve to run heavier filters or other program and a bit darker sound with cleaner trebles due to blacker background. Either less noise is generated when full ticked, or noise is more randomized, or iSilencer based on ANC can easier catch it. I tried without iSilencer and the sound difference between ticked and unticked lowered.

[bogi]

I yet tried heavier filters. With multicore full ticked I could run all expect gauss-xl, gauss-xla and ext3, with poly-sinc-xtr-mp be on border.

So I yet compared -2 and -3 versions on that one. The -2 version was less prone to dropouts while starting Firefox (then it stabilized). Teh -3 version needed much more processing power (see pictures below, 40% overall with -2 vs. 69% for -3). I didn't find a benefit of -3 in this scenario.

 

4.14-2:

 

4.14-3:

 

After all the experiments I more prefer the -2 version against -3 when Multicore DSP is full ticked, so I would be glad to be able to set it using env. variable if it would not become the default.

[MasterWarzombie]

i tested the latest version .3. I still have cuts with the same setting with the box grayed out for the DSP core.

[kelvinwsy]

I disabled AO3 on a separate drive running Windows Server 2019 Core Mode and installed 4.14-3; I ran gpedit.msc and enabled hardware driver to ensure OpenGL is enabled.

Installation - No error. Tried to run Hqplayer 4 desktop.exe - Black screen Nada Nothing No error Message.

I repeat that on another driver running Windows Server 2019 GUI mode, 4.14-2 installed and ran smoothly. On this machine a 11600k w/o GPU at 4.7ghz on all core (6 only) I can run EC7 V2 even using Sine Mx filters at DSD256 without any hiccups. 

 

So what am I doing wrong? Is 4.14 incompatible with Windows Server 2019 Core Mode?
4.13.1 ran smoothly without any issues at all!

 

I am really hopeful someone can help with this?

[kelvinwsy]

My 11900k has Intel HD 730 integrated graphics which when enabled under Remote connection option shuld support OpenGl natively..