HQ Player

[AnotherSpin]

3 hours ago, DancingSea said:

I realize ultimately we can experiment, but I’m wondering if a guide to HQPlayer filters has been created that focuses on common man language sonic descriptions of the various settings?  The manual contains what is for me, highly technical descriptions of the various filters and dither settings that is mostly meaningless to me. 

 

I would like to have such descriptions as well. 

[Miska]

3 hours ago, tobes said:

@Miska what makes the Closed Form Fast filter different to the other Closed Form filters? - I know it supposedly has less precision/CPU load, though honestly load doesn't seem much different (very low, I only do PCM upsampling - using 2.3Ghz i7 Mac Mini).

What I find odd is that the less precise 'Fast' filter sounds better to me - it has a clarity and naturalness that I don't find in the other Closed Form variants, or the poly-sinc etc filters either.

Predominately using this with a Benchmark DAC3 (Sabre 9028 based).

I recently acquired the Chord Qutest, but so far the DAC3 (with HQP/closed form fast) has the edge in clarity.

 

It should be probably rather called "closed-form-light", but it is simpler version of the algorithm. For higher output rates like DSD512 if the normal closed-form begins to get heavy, then it makes difference in load. If the load is very low to begin with, then the difference pretty much disappears. From that point of view, having fast for PCM-outputs was kind of unnecessary. But the algorithm itself is somewhat different.

 

[tobes]

Thanks for the reply @Miska.

Yes, the CPU advantage with 'Fast' for PCM upsampling is irrelevant - I was curious because I have found myself preferring it to standard closed form, so wondered what it was doing differently.

What filter from the poly sinc series does the closed form fast most resemble?

[tobes]

Thanks for the reply @Miska.

Yes, the CPU advantage with 'Fast' for PCM upsampling is irrelevant - I was curious because I have found myself preferring it to standard closed form, so wondered what it was doing differently.

What filter from the poly sinc series does the closed form fast most resemble?

 

EDIT: Actually as I listen now the standard Closed Form has a more natural sound on (some) acoustic music, the 'Fast' variant has a touch more etch (not the right word because it maintains good dimensionality) - not objectionable - more often that not providing an engaging presence.

[Miska]

4 hours ago, tobes said:

What filter from the poly sinc series does the closed form fast most resemble?

 

Since all closed-form are strictly non-apodizing, for example poly-sinc-hb or the least apodizing poly-sinc-xtr.

 

[baconbrain]

I am considering using HQ Player again. My last install was about 2 years ago with a mR as NAA. The mR is now history, but I have an extra DIY Audio PC that I could use. Perhaps I am having a brain cramp, but I am struggling to understand how I would go about setting HQ Player up within my system chain. Currently I am running Roon Server on my Zenith SE and the Trinnov is a Roon End Point. All via Ethernet.

 

Would appreciate it if someone could point me in the right direction? 

[firedog]

2 hours ago, baconbrain said:

I am considering using HQ Player again. My last install was about 2 years ago with a mR as NAA. The mR is now history, but I have an extra DIY Audio PC that I could use. Perhaps I am having a brain cramp, but I am struggling to understand how I would go about setting HQ Player up within my system chain. Currently I am running Roon Server on my Zenith SE and the Trinnov is a Roon End Point. All via Ethernet.

 

Would appreciate it if someone could point me in the right direction? 

You need either full HQP setup and use it directly from that computer, or you need full HQP setup on a computer and HQP NAA setup on a network connected device that can receive the stream from the HQP computer. 

 

Can the Zenith run HQP? If so, you can use the DIY as the NAA. 

 

Or with Roon, you can install HQP on a computer on the Network, and tell it to send the output to HQP at it’s network address where you installed it. That then is your HQP audio PC

[baconbrain]

1 hour ago, firedog said:

You need either full HQP setup and use it directly from that computer, or you need full HQP setup on a computer and HQP NAA setup on a network connected device that can receive the stream from the HQP computer. 

 

Can the Zenith run HQP? If so, you can use the DIY as the NAA. 

 

Or with Roon, you can install HQP on a computer on the Network, and tell it to send the output to HQP at it’s network address where you installed it. That then is your HQP audio PC

Thx, but with the last proposed setup, that would mean that my Trinnov would have to function as a NAA (which it does not) to be reachable via Ethernet, correct? 

[firedog]

7 hours ago, baconbrain said:

Thx, but with the last proposed setup, that would mean that my Trinnov would have to function as a NAA (which it does not) to be reachable via Ethernet, correct? 

I'm not sure how the Trinnov works. But I'd assume you would use the digital out from whatever device HQP itself is installed on and connect that to an input on the Trinnov. No NAA needed in this setup.

[rickca]

@Miska do you have any experience with SMSL VMV D1 DAC?  I'm curious how this compares to the RME ADI-2 DAC variants you like.

[Miska]

57 minutes ago, rickca said:

@Miska do you have any experience with SMSL VMV D1 DAC?  I'm curious how this compares to the RME ADI-2 DAC variants you like.

 

No, I have not seen anybody selling such in EU. There are quite a bunch of Sabre DACs around. Matrix X-Sabre Pro is available, but a bit too expensive for what I'm ready to pay for such.

 

[fgribas]

@Miska is it possible to make sinc-M output the same rate as input? 

 

When I use a Dragonfly DAC (max 96 kHz) playing 96 ou 88 content, it actually downsamples to 44 or 88.

[Miska]

2 hours ago, fgribas said:

@Miska is it possible to make sinc-M output the same rate as input? 

 

When I use a Dragonfly DAC (max 96 kHz) playing 96 ou 88 content, it actually downsamples to 44 or 88.

 

Not very straightforward, but I think I have better solution for this case coming...

 

[sixman]

Miska, most dacs have their own internal SRC and resample to their own intrinsic rate, as well as apply their own filters to the incoming signal. Does this negate the processing done in HQP to some extent?

 

I always wonder if people with say a PS Audio DS dac get the same benefit of using HQP?

[Account Closed]

39 minutes ago, sixman said:

Miska, most dacs have their own internal SRC and resample to their own intrinsic rate, as well as apply their own filters to the incoming signal. Does this negate the processing done in HQP to some extent?

 

I always wonder if people with say a PS Audio DS dac get the same benefit of using HQP?

There have been posts where people say that for PCM they get the best results using HQP from DACs that are either NOS or can turn off SRC and filtering to become NOS (Holo Spring as an example). For DSD this should not be an issue.

[Miska]

7 hours ago, sixman said:

Miska, most dacs have their own internal SRC and resample to their own intrinsic rate, as well as apply their own filters to the incoming signal. Does this negate the processing done in HQP to some extent?

 

I always wonder if people with say a PS Audio DS dac get the same benefit of using HQP?

 

It doesn't negate the effect, even if you can only upsample PCM 2x, it still has benefits because the first step up from original rate is most critical regarding digital filters. For some DACs you can either bypass digital filters completely, or at least the first stage by sending in highest rate PCM.

 

Since most DACs are SDM type these days, by sending DSD to a DAC that has native direct conversion for DSD you can also bypass it's built-in modulator. Even without that you may still get benefits of running proper digital filters to the final rate, something they are not usually capable of doing.

 

Digital section of modern DACs have these two aspects; digital filters and modulator. Level of effect depends on if you can bypass just the filter (fully or partially) or also the modulator.

 

In case of PCM type DAC (R2R) the modulator doesn't apply anyway, only the digital filter.

 

As bobflood said, for maximum effect one can use a DAC that can operate as a D/A converter only without digital processing.

 

For the direct conversion discrete DAC types, such as Holo Spring or T+A (in DSD mode) things are quite straightforward. For most chip-based DACs as well.

 

For non-direct conversion custom DACs it depends on a particular DAC whether it is better to send there PCM or DSD. In such cases, if not known, sending highest supported PCM rate is usually a safe bet. For Chord, use PCM (unless maybe using Dave with DSD+ mode). For dCS I believe the DACs can natively convert DSD128. For PS Audio maybe PCM, I think it doesn't support direct conversion although the conversion rate itself runs at DSD128. For Devialet, send PCM. For Playback Designs, DSD128 and DSD256 probably result in direct conversion. Same for EMM Labs / Meitner. For Mola-Mola, send PCM.

 

[Yviena]

@Miska Can you take a look at DSD playback with the closed-form filters with ryzen CPUs, it seems there is some weird interraction with it as it seems to make vsync/frametimes in Firefox/3D apps erratic even if cpu usage is only 14% on a 2700x, the XTR filter does not seem to have this problem, and i think i narrowed it down to a CCX issue as setting thread affinity to one CCX largely mitigates it.

 

[Miska]

50 minutes ago, Yviena said:

@Miska Can you take a look at DSD playback with the closed-form filters with ryzen CPUs, it seems there is some weird interraction with it as it seems to make vsync/frametimes in Firefox/3D apps erratic even if cpu usage is only 14% on a 2700x, the XTR filter does not seem to have this problem, and i think i narrowed it down to a CCX issue as setting thread affinity to one CCX largely mitigates it.

 

What OS are you on?

 

Sound strange, or are you using CUDA offload? With CUDA I can very well understand such thing happening.

 

[Yviena]

53 minutes ago, Yviena said:

@Miska Can you take a look at DSD playback with the closed-form filters with ryzen CPUs, it seems there is some weird interraction with it as it seems to make vsync/frametimes in Firefox/3D apps erratic even if cpu usage is only 14% on a 2700x, the XTR filter does not seem to have this problem, and i think i narrowed it down to a CCX issue as setting thread affinity to one CCX largely mitigates it.

 

 

I'm on the windows 1809 build, Cuda offload is unticked, I too find it strange that it's only the closed-form filter that interracts this way, cpu usage seems to have no correlation to the issue.

 

Ram speed is at 3466 Stable so  i see no reason why CCX should affect it but it does...

 

 

 

[Miska]

8 minutes ago, Yviena said:

Ram speed is at 3466 Stable so  i see no reason why CCX should affect it but it does...

 

Sorry my ignorance, but what is CCX?

 

I can also check on my older Ryzen 7...

[Yviena]

9 minutes ago, Miska said:

 

Sorry my ignorance, but what is CCX?

 

I can also check on my older Ryzen 7...

I think this article can explain ryzen CCX much better than i can : https://www.anandtech.com/show/11170/the-amd-zen-and-ryzen-7-review-a-deep-dive-on-1800x-1700x-and-1700/9

 

But like i said limiting HQplayer to one CCX eg core 8-10-12-14 really reduces the micro-stutter, curiously it seems that not all software are affected,  here's a comparison in Firefox between auto affinity, and manually set thread affinity in hqplayer.

First one with manual thread affinity

 

Second one with auto affinity.

 

This is with nothing else but firefox, and hqplayer running as you can see there the spikes are noticeably improved  when affinity is set manually.

[Miska]

29 minutes ago, Yviena said:

I think this article can explain ryzen CCX much better than i can : https://www.anandtech.com/show/11170/the-amd-zen-and-ryzen-7-review-a-deep-dive-on-1800x-1700x-and-1700/9

 

Ahh, lovely abbreviations...  Never been good at remembering such.

 

30 minutes ago, Yviena said:

But like i said limiting HQplayer to one CCX eg core 8-10-12-14 really reduces the micro-stutter, curiously it seems that not all software are affected,  here's a comparison in Firefox between auto affinity, and manually set thread affinity in hqplayer.

First one with manual thread affinity

 

Surprising that closed-form somehow behaves differently from others, because the threading structure is not all that different. But the root cause seems to be that Windows' scheduler is not doing awfully good job. It should understand the CPU topology and thread locality and assign threads reasonably to the cores.

 

What would be really interesting would be to test the same case on Linux as well. I have now both Win10 and Ubuntu on the Ryzen machine. What is the FPS test you are using so I could reproduce the test case?

 

HQPlayer understands some amount about CPU topology and threads have certain hierarchy that should help the OS scheduler too. But trying to overcome job of the OS scheduler usually results in worse performance. Also Microsoft says so "Setting an affinity mask for a process or thread can result in threads receiving less processor time, as the system is restricted from running the threads on certain processors. In most cases, it is better to let the system select an available processor."

 

I may revisit this topic when I work on improvements to the multi-socket/NUMA support, because this is related...

 

[Yviena]

3 minutes ago, Miska said:

 

Ahh, lovely abbreviations...  Never been good at remembering such.

 

 

Surprising that closed-form somehow behaves differently from others, because the threading structure is not all that different. But the root cause seems to be that Windows' scheduler is not doing awfully good job. It should understand the CPU topology and thread locality and assign threads reasonably to the cores.

 

What would be really interesting would be to test the same case on Linux as well. I have now both Win10 and Ubuntu on the Ryzen machine. What is the FPS test you are using so I could reproduce the test case?

 

HQPlayer understands some amount about CPU topology and threads have certain hierarchy that should help the OS scheduler too. But trying to overcome job of the OS scheduler usually results in worse performance. Also Microsoft says so "Setting an affinity mask for a process or thread can result in threads receiving less processor time, as the system is restricted from running the threads on certain processors. In most cases, it is better to let the system select an available processor."

 

I may revisit this topic when I work on improvements to the multi-socket/NUMA support, because this is related...

 

I just used vsynctester.com as you can still see the behavior between auto/manual threads on there, there are probably better software out there to check vsync/frametimes, and scheduler problems, and yeah Microsoft's scheduler is shit compared to linux their adding useless "Features" that a minority of people will use each update, instead of working on performance improvements, and ,long standing bugs.

 

PS: Don't update to the new windows when it releases, they fucked something up between the gpu driver rendering, and the OS leading to more banding when used with NVIDIA.

 

[Conte_Oliver]

Hello I have an installation problem, this is the SO specification: Win 10 1607 (build SO 14393.2848)

What Happen is that the program simply do not open. There is a very brief rotation of the wheel and nothing.

I've tried to disinstall and reinstall many times. If I use RevoUninstaller_Portable to cancel the app for a session I'm able to use HQplayer as soon as I reboot it won't start anymore

The problem surfaced 2 month ago

I do not want to reinstall windows

Thanks fo your kind help

GF

 

 

[bunno77]

On 3/12/2019 at 5:42 PM, Miska said:

 

Really hard to say why, or do something about it, unless I can reproduce it here...

 

So far it's been usually due to missing CPU feature, but if also the 32-bit version doesn't run (just remember to uninstall 64-bit before installing 32-bit), then it's not that either.

 

 

 

So I have kept playing around today. Have been deleting anything hq related, reinstall etc and even disabling antivirus. Nothing has worked. I did a search and found a program called Dependancy Walker that apparently if you drop in the .exe it will tell what is wrong if there is problems. Here is a screenshot of what it showed after I dropped in hq .exe.

Not sure if this is relevant and I don't know if it helps but here it is. Some weird older dates there too