HQ Player

[Miska]

6 hours ago, bobflood said:

How can certain manufacturers who I won't name claim to be doing DSD 1024 on what appears to be an ordinary FPGA chip?

 

Is there some trick to programing or some unknown capability that I am not aware of or do not understand?

 

I know that HQ Player can do higher than DSD 512 but to date I have not seen anyone be able to actually do it and I don't think there is a DAC readily available that can accept it. It is the same issue as these mega million tap high rate PCM filters being run on a FPGA. The PCM scenario seems at least plausible but the DSD 1024 on a FPGA seems far fetched.

 

Because you can reach the same goal in many ways, if you just want "DSD1024" and are not picky about actual quality.

 

You can do it easily if you lower the quality bar and consider something "good enough". DAC chips do similar thing these days too. So the usual approach, like DAC chips do is to first have some platform (FPGA or on-chip DSP) and then make processing that can run on that what ever limited amount of resources you have. For example DAC chips run proper digital filter up to 8x rate and they they just repeat the same sample value 128 times to make up 1024 "rate" for the modulator. Or some more fancy chip like Chord DAC FPGA runs digital filter to 16x rate and then does linear interpolation (argh) to 2048 "rate".

 

While I prefer to do things vice versa, first make specifications of what I consider "perfect" result and then make processing that reaches the wanted quality level and runs as fast as possible. And then you need to buy hardware that can run it. I don't tailor the algorithm to run on any specific hardware, the hardware needs to be tailored to run the algorithm.

 

For Chord's million taps to 256x which is best they can do now, you can run 16 million taps on cheap GTX 1060 GPU to 512x in comparison... (and it's less than 25% load on that GPU) Like the number of taps alone would matter in first place...

 

Most FPGA's do processing in fixed point and at lower resolution. I do processing in 64/80-bit floating point. Making an FPGA do complex floating point processing already takes immense amount of FPGA space. And I would never resort to something like repeating samples or linear interpolation to reach the target rate.

 

6 hours ago, diecaster said:

FPGA chips can be immensely powerful.

 

Sometimes companies ask what it would take to run my algorithms on FPGA, when I tell them that it would need Xilinx biggest Virtex FPGA's, they quickly forget about it. Such big FPGA's cost way more than CPU/GPU and you are likely not going to find such chips in DACs.

 

[Miska]

Another aspect besides digital filters and oversampling, is in the modulator area. Most hardware implementations use a static modulator design that runs synchronously at the clock rate. While the GHz clock frequencies of CPU's allow making asynchronous adaptive modulators in software. Where the modulator is analyzing signal at the output sampling rate and adjusting parameters if necessary and can even re-do the calculations if the output doesn't match expectations. This analysis part also makes up notable part of the computations.

 

Essentially you can look at how many clock cycles there are per produced output sample. If your clock runs at 4 GHz, you have 177 clock cycles per DSD512 output sample.

 

[vmanuelgm]

Hi Miska.

 

I purchases HQPlayer a while ago and amd trying to install it on Linux Mint 18.3 KDE, but it won't because of some dependencies using the OS own installers.

 

Any tips???

 

Thanks in advance.

[Miska]

24 minutes ago, vmanuelgm said:

I purchases HQPlayer a while ago and amd trying to install it on Linux Mint 18.3 KDE, but it won't because of some dependencies using the OS own installers.

 

Which version are you trying to install? 3.25 is built for Ubuntu 18.04 LTS (Linux Mint 19) while 3.24 was the last one built for Ubuntu 16.04 LTS (Linux Mint 18). Table here is useful for matching the versions:

https://en.wikipedia.org/wiki/Linux_Mint_version_history

 

If you install the package first with "sudo dpkg -i" you can pull the missing (failing) dependencies with "sudo apt install -f" and it'll install those and configure the packages.

 

[vmanuelgm]

36 minutes ago, Miska said:

 

Which version are you trying to install? 3.25 is built for Ubuntu 18.04 LTS (Linux Mint 19) while 3.24 was the last one built for Ubuntu 16.04 LTS (Linux Mint 18). Table here is useful for matching the versions:

https://en.wikipedia.org/wiki/Linux_Mint_version_history

 

If you install the package first with "sudo dpkg -i" you can pull the missing (failing) dependencies with "sudo apt install -f" and it'll install those and configure the packages.

 

Hi Miska, thanks for your support.

 

I am trying to install it on Linux Mint 18.3 based in Ubuntu 16.04.

 

So where can I obtain the 3.24??? Do I need to use the sudo dpkg -i command???

 

Once installed, is it able to play dsd native??? If not, what should I do to make it work under Linux Mint 18.3???

[Miska]

14 minutes ago, vmanuelgm said:

I am trying to install it on Linux Mint 18.3 based in Ubuntu 16.04.

 

So where can I obtain the 3.24??? Do I need to use the sudo dpkg -i command???

 

It is still available on the server, you can modify the download link to point to 3.24.

 

That way I do it, I don't know if some other way works too...

 

15 minutes ago, vmanuelgm said:

Once installed, is it able to play dsd native??? If not, what should I do to make it work under Linux Mint 18.3???

 

It depends on your DAC... Sorry, but I didn't quite spot the DAC from your signature.

 

[eternaloptimist]

20 hours ago, guiltyboxswapper said:

 

Are you able to do 44.1x -> poly-sinc-xtr at 512x 48k base?  My T+A's 44.1k side is broken, hence the question.  Will eventually get this fixed.

Yep. No problem. Tested from different sample rates. 

[vmanuelgm]

30 minutes ago, Miska said:

 

It is still available on the server, you can modify the download link to point to 3.24.

 

That way I do it, I don't know if some other way works too...

 

 

It depends on your DAC... Sorry, but I didn't quite spot the DAC from your signature.

 

Eastern Minimax Supreme. Its able to play dsd native under Windows 10 with your program and others like Jriver.

 

As the genuine drivers haven't been updated since release date, I had to mod the M2Tech 3.34 xmos version to accept my Supreme, and these drivers are able to play dsd native. The genuine ones from Eastern allow only DOP.

 

So, how could I set my Linux Mint up to permit dsd native via HQPlayer???

[eternaloptimist]

21 hours ago, elan120 said:

Very good to know this GPU can elevate performance this much.  Could you try "Closed-form-16M" at 512?  This seems to be the real stress test of all the filters.

From 44.1... CPU 28%, GPU 12%.  Stable, no problems. 

[vmanuelgm]

49 minutes ago, Miska said:

 

It is still available on the server, you can modify the download link to point to 3.24.

 

That way I do it, I don't know if some other way works too...

 

 

It depends on your DAC... Sorry, but I didn't quite spot the DAC from your signature.

 

I am trying to download 3.24 deb but I am not able. Could you please post the exact link to download???

[guiltyboxswapper]

45 minutes ago, eternaloptimist said:

Yep. No problem. Tested from different sample rates. 

Thanks !  Better order a 2080 then!

[shadowlight]

1 hour ago, vmanuelgm said:

I am trying to download 3.24 deb but I am not able. Could you please post the exact link to download???

Try - www.signalyst.com/bins/hqplayer_3.24.0-79_amd64.deb

[Miska]

4 hours ago, vmanuelgm said:

Eastern Minimax Supreme. Its able to play dsd native under Windows 10 with your program and others like Jriver.

 

As the genuine drivers haven't been updated since release date, I had to mod the M2Tech 3.34 xmos version to accept my Supreme, and these drivers are able to play dsd native. The genuine ones from Eastern allow only DOP.

 

So, how could I set my Linux Mint up to permit dsd native via HQPlayer???

 

You'll probably need to try with my custom kernel for Xenial (Ubuntu 16.04), from here:

https://www.sonarnerd.net/src/

 

I'm not updating it anymore, but it should work in this case. Just note that any potential kernel security fixes are not coming to this one. Only to the newer Bionic one.

 

[vmanuelgm]

3 hours ago, shadowlight said:

Try - www.signalyst.com/bins/hqplayer_3.24.0-79_amd64.deb

 

Thanks a lot!!!

 

 

 

50 minutes ago, Miska said:

 

You'll probably need to try with my custom kernel for Xenial (Ubuntu 16.04), from here:

https://www.sonarnerd.net/src/

 

I'm not updating it anymore, but it should work in this case. Just note that any potential kernel security fixes are not coming to this one. Only to the newer Bionic one.

 

Thanks for your answer.

 

Now I am facing a new problem. After installing HQPlayer in Linux Mint, it will refuse to accept my xml key file (generated and working under Windows 10).

 

What can I do???

 

 

EDIT: I answer myself. Didn't know theres a different purchase for Linux, xDDD

[Tannoymike]

Why do I find HQPlayer so sluggish in operation, particularly when hit the stop button or play it takes what seems like several seconds before it stops or starts playing the music, opening the program is also sluggish. Can I use Asset UPNP with HQPlayer. Why also is it so much more difficult to setup then say JRiver which I find so responsive to commands, although I must admit the sound of HQPlayer is very good/

[shadowlight]

4 minutes ago, Tannoymike said:

Can I use Asset UPNP with HQPlayer. Why also is it so much more difficult to setup

 

Can you let us know what part of the setup are you finding difficult?  I am assuming you are using the desktop version?  I have not tried using UPnP server with HQPlayer.  I have always mounted my NAS/Windows share on the system that is running HQP on and just used the library import functionality.

[shadowlight]

52 minutes ago, vmanuelgm said:

EDIT: I answer myself. Didn't know theres a different purchase for Linux, xDDD

Reach out to Miska.  He has previously provided a discount code for 2nd license.

[vmanuelgm]

16 minutes ago, shadowlight said:

Reach out to Miska.  He has previously provided a discount code for 2nd license.

 

 

Thanks, I emailed him!!!

[Account Closed]

9 hours ago, Miska said:

 

Because you can reach the same goal in many ways, if you just want "DSD1024" and are not picky about actual quality.

 

You can do it easily if you lower the quality bar and consider something "good enough". DAC chips do similar thing these days too. So the usual approach, like DAC chips do is to first have some platform (FPGA or on-chip DSP) and then make processing that can run on that what ever limited amount of resources you have. For example DAC chips run proper digital filter up to 8x rate and they they just repeat the same sample value 128 times to make up 1024 "rate" for the modulator. Or some more fancy chip like Chord DAC FPGA runs digital filter to 16x rate and then does linear interpolation (argh) to 2048 "rate".

 

While I prefer to do things vice versa, first make specifications of what I consider "perfect" result and then make processing that reaches the wanted quality level and runs as fast as possible. And then you need to buy hardware that can run it. I don't tailor the algorithm to run on any specific hardware, the hardware needs to be tailored to run the algorithm.

 

For Chord's million taps to 256x which is best they can do now, you can run 16 million taps on cheap GTX 1060 GPU to 512x in comparison... (and it's less than 25% load on that GPU) Like the number of taps alone would matter in first place...

 

Most FPGA's do processing in fixed point and at lower resolution. I do processing in 64/80-bit floating point. Making an FPGA do complex floating point processing already takes immense amount of FPGA space. And I would never resort to something like repeating samples or linear interpolation to reach the target rate.

 

 

Sometimes companies ask what it would take to run my algorithms on FPGA, when I tell them that it would need Xilinx biggest Virtex FPGA's, they quickly forget about it. Such big FPGA's cost way more than CPU/GPU and you are likely not going to find such chips in DACs.

 

 

Thank you Miska for the very detailed answers. I suspected as much.  Bob

[jimdukey]

Has anyone heard the new Ifi Dac?

Or PS Audio?

Or Marantz SA 10?

[Miska]

18 minutes ago, jimdukey said:

Has anyone heard the new Ifi Dac?

Or PS Audio?

Or Marantz SA 10?

 

Which iFi DAC do you mean, the Pro iDSD?

 

[jimdukey]

Yes, that's the new one.

[Sagittarius]

10 hours ago, Miska said:

 

 

Essentially you can look at how many clock cycles there are per produced output sample. If your clock runs at 4 GHz, you have 177 clock cycles per DSD512 output sample.

 

Interesting. I think  this simple calculation sheds some light on the reported performance of various processor designs with the more strenuous HQPlayer algorithms. It makes sense not to waste many of those clock cycles over latency, hence the importance of cache size, or other unrelated processes, which is  probably part of the reason, beside larger total cache size, that some processors with a large number of cores are reported to perform ok with HQPlayer's more difficult loads despite their lower clock speeds as they have more free cores to use. 

 

One question I have is whether the newer non-inclusive victim L3 cache design that Intel have been using with Skylake X processors is still useful for HQPlayer more demanding loads. In other words, would one be (at least theoretically) be better off using a high core Broadwell generation processor with its larger L3 inclusive cache than going with Skylake X?

[eternaloptimist]

8 hours ago, guiltyboxswapper said:

Thanks !  Better order a 2080 then!

Sorry.... I made a mistake in testing....

i cannot get poly-sinc-xtr filters to work from 44.1 (and multiples) to 512k x 48. Just hangs with the progress bar...

No problem with 48 to 512x48 as expected.

Maybe hold off on the order until the DAC is fixed...

[Miska]

Just now, eternaloptimist said:

i cannot get poly-sinc-xtr filters to work from 44.1 (and multiples) to 512k x 48. Just hangs with the progress bar...

 

It will eventually get going if you wait it to get initialized, but it cannot keep up with the audio...