HQ Player

[Flextreme]

   

21 hours ago, Flextreme said:

Recently rediscovered the DSD filter in HQplayer. Love the gigantic (almost overly audiophile) liquid/plastic presentation of Sinc-L/ASDM7/@DSD256 on my AMD 5900x/RTX 3090 pc sourcing a May KTE.

 

Love to try ASDM7EC, but at DSD64 it barely works, higher DSD rates are constantly drop out, even with multicore DSP and Cuda enabled (and working, verified this). My guess my AMD x5900 is the limiting factor (processor or usb limitations?).

 

Did all recent  ZEN 3 platform updates for USB issues, but the fast MAY firmware (30.12) still does not work for me.

 

Anyone here got the May with 30.14 firmware running ASDM7EC at DSD256x on a AMD ZEN3 platform?

 

OK, after some extensive troubleshooting I think if found a solution to run ASDM7EC at DSD256x on a AMD ZEN3 platform without drops. This is probably specific to ZEN3.

 

Solution: Enable multi DSP and Cuda if possible. Fix and limit affinity of HQP to 3 threads on 3 separate logical processors on CCx1.

 

 

 

Create a shortcut with the command below, replace XXX with CMD, cloudfare blocks references to the cmd executable. This will fix HQP to the first 3 CPUs, non SMT.

 

C:\Windows\System32\XXX.exe /c start "C:\Program Files\Signalyst\HQPlayer 4 Desktop" /High /Affinity 15 "C:\Program Files\Signalyst\HQPlayer 4 Desktop\HQPlayer4Desktop.exe"

 

Basically this is the ONLY way I can run ASDM7EC at DSD256x... All other masking/affinity options (more/less threads) will not work for me. At 4 or more logical CPUs, HQP threads seem to start jumping cores and it this seems to leads dropouts. 2 CPU's/Threads via masking is not enough for ASDM7EC. Ultimately,. something looks wrong with the scheduling of threads of HQP on AMD...

 

I'm confident other AMD users run into the same issues as me, hopefully this can help.

 

Edit: forgot to mention that his dramatically reduces CPU temperature, by more than 10c, and therefore my PC runs as quite as it can be. Very nice...

[Flextreme]

For my Susvara/Holo May I am currently using Roon Parametric EQ and then oversample to dsd256x ASDMEC7 with Sinc-L/Mx ... Love it to bits!

 

However, I was wondering if I use EQ/Convolution in HQplayer, instead of Roon EQ, would that yield even better results?

 

If so, which method in HQ player is prefered SQ wise?

 

And how can I translate my parametric EQ settings to HQplayer?

[Flextreme]

19 hours ago, Miska said:

 

Sorry, this was left unanswered, I remembered this when you posted a new one.

 

1) Use if your convolution filters are produced for lower sampling rate than your highest sampling rate content. Note that for PCM+DSD sources, 352.8k is a good compromise to certainly cover enough bandwidth for any audio source (176.4 kHz worth of frequency response). Point is that convolution filter WAV has upper frequency response limit of fs/2 of the filter. HF extend will take on from there and extend the frequency response beyond that point as flat line at the same level where the filter's response ended. Usually, if possible, it is best to ask the filter design software to make filters straight for 352.8k (or 192k is also pretty good already).

 

2) Overlap-save will usually burn about 1.5x more processing power. Mathematically, given the output resolution, both methods should give same result. But both options are offered in case someone finds a case where they would sound different for some reason.

 

 

Let me just state I am beyond impressed with how transparant EQ is within HQP and using oversampling on my May KTE, Bliss KTE, and Susvara. After two years of HQP/Susvara ownership, I am still in awe of the postive impact HQP EQ and OS on have soundstage, detail and musical energy in comparison to regular NOS. 

 

Lately I discovered I do prefer EQ and upsampling with PCM (poly-sinc-gauss-lon/LNS15) instead of DSD (which is typically prefered).

 

I wonder why? Because I use EQ and mayby EQ is applied after upsampling with PCM and not with DSD? I also prefer overlap-save, it seems more detailed and more dynamic. I understand there should be no difference. But I have to admit that this difference is subtle, but I do prefer it.