HQ Player

[brother love]

3 hours ago, GUTB said:

I have an i7-7700K Windows 10 system. This is the first time I've used ASDM7EC since I just upgraded from an old version of HQPlayer 3. It cut out every 10 seconds or so at DSD256. I figured it was because my CPU is passively cooled and I have noted in the past that the passive cooling isn't enough to keep 7700K from throttling under load. I had no idea this was a common problem with 4.7.

 

Not a surprise. HQPlayer 4.6 Desktop DSD256/ ext2/ ASDM7EC (or ASDM5EC for that matter) resulted in drop-outs w/ my dual boot i7-6700k Win10 Pro (Server case/ mobo cooled by 2 low rpm Noctua fans). Haven't tried HQP 4.7.

 

Ubuntu 18.04 HQPlayer 4.6 & latest 4.7 Desktop DSD256/ ext2/ ASDM7EC works great w/ no drop-outs.

[Miska]

56 minutes ago, Yviena said:

Wouldn't  a new DSD protocol that is capable of directly sending 2-6 bits to the dac chip/resistor stage also be another way for better sound quality.

 

It would be nice, but doesn't ease up the computational requirements at all if you still want maximum quality.

 

2 hours ago, Yviena said:

Interestingly  AKM has  separated the dac IC, and modulator in 4498+4191, and increased the upsampling from 8x to 256x

 

Good progress, maybe they have heard of HQPlayer...

 

Well, in the end that is how things used to be in the 90's when digital filters were separate chips from the actual converters. This allows more processing power without disturbing the sensitive analog parts. But it also allows feeding the conversion section straight from something like HQPlayer or other custom DSP.

 

[rickca]

2 hours ago, Yviena said:

Interestingly  AKM has  separated the dac IC, and modulator in 4498+4191, and increased the upsampling from 8x to 256x

Thanks for reminding me of this.  Have you seen any DACs announced that will use this combo?

[Miska]

1 hour ago, fredg_31 said:

Thanks Miska, I’m listening « Trio in Tokyo » with these settings (ext2 / TPDF 44.1->192), relaxing sound, perhaps a bit too much, or I need to forget Daphile sound .

 

I’ll give a try to mp filter later.

 

You can try sinc-S too, that has a bit more bite (in my opinion) on certain recordings that have laid back sound. But you need to set "Adaptive output rate" to grayed for all these filters to work, since sinc-S cannot do 44.1 - > 192 ratio, but instead just 44.1 -> 176.4. Then it'll output 176.4 when necessary and 192 when possible.

 

[Miska]

4 hours ago, StreamFidelity said:

I don't want to argue. The forum also serves as customer feedback and I would like to advertise that I can also participate in future updates with a fixed clock frequency. My arguments:

 

I just can't do much about this... I cannot stick to all old things forever.

 

4 hours ago, StreamFidelity said:

2. Unfortunately, your requirement is no longer valid? In my case, all cores run significantly above this at 4.2GHz.

 

Yes it is, my Xeon W-2245 runs poly-sinc-ext2 + ASDM7EC at DSD256 just fine, two cores running at 3.9 GHz, two running at 1.4 GHz and rest at 1.2 GHz. These things are not so straightforward that clock frequency of X would be enough to do Y, it depends on many different factors. (at the same time I have web browser running and typing this to the forum, etc, so HQPlayer is not the only task)

 

4 hours ago, StreamFidelity said:

But OK. I keep it like @bobflood and stick with 4.6.0. Still, I would be very sorry if I didn't attend future updates.

 

I guess sticking to old version is then the only way. The other option would be to stop moving HQPlayer forward altogether, resulting in essentially getting stuck to 4.6.

 

4 hours ago, StreamFidelity said:

1. I may not be the right yardstick. But many use AudiophilOptimizer and also follow some recommendations from it.

 

And I've told many times that it is asking for trouble and will likely break HQPlayer one or more ways.

 

[rickca]

@Miska the RTX 3070 is supposed to outperform an RTX 2080 Ti.  Do you think the RTX 3070 will be enough to satisfy HQPlayer requirements for the next couple of years, or would you recommend spending another $200 for the RTX 3080?

[Yviena]

1 hour ago, bobflood said:

6 bit DSD sounds like PCM lite to me.

Yeah it would basically be DSM/PCM done in software which would allow for decoupling of the digital, analog side, and the modulation can be done with much higher processing power which in turn means less shortcuts used, and with it not being fixed to a IC chip it can continue to evolve faster when new stuff, or techniques come out.

[Miska]

1 hour ago, rickca said:

@Miska the RTX 3070 is supposed to outperform an RTX 2080 Ti.  Do you think the RTX 3070 will be enough to satisfy HQPlayer requirements for the next couple of years, or would you recommend spending another $200 for the RTX 3080?

 

They never talk about CUDA performance of GeForce cards, so it is hard to estimate how the relative performance will be until it gets tested.

 

But best figures so far are the processing power figures for double precision floating point available in Wikipedia.

 

For RTX 20:

https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_20_series

 

For RTX 30:

https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_30_series

 

Based on those, 3070 is a bit less than 1.5x speed of 2080Ti. If that is the case also for HQPlayer use case, then 3070 is certainly good value!

 

Other interesting part is that 3090 has enough RAM that it should be able to run for example sinc-L filter to DSD512 rates, at price similar to 2080Ti (which cannot do it due to running out of RAM), but much lower price than TITAN RTX which had the same amount of RAM.

 

 

In any case, HQPlayer Desktop 4.7+, Embedded 4.18+ and Pro 4.8+ have already support for this new generation.

 

[NanoSword]

I am getting drop with DSD 256 / ASDM7EC every 20 seconds cpu amd 3950x using euphony .

[jimdukey]

Compared to 16/44, how much More Resolution is 24/96, DSD 64, DSD 128?

Is there a way to express it as a percentage?

Or just Bits-per second?

TIA!

[Account Closed]

8 hours ago, NanoSword said:

I am getting drop with DSD 256 / ASDM7EC every 20 seconds cpu amd 3950x using euphony .

Roll back to 4.6. 4.7 pushes cpu harder. Using 4.6 my cpu (i5 8400) would run at 48-50% total load with ASDM7EC/256, with 4.7 it was being pushed to 53-55% and it could not keep up so I had drop-outs. When I rolled back to 4.6 all worked fine again.

[fred_com]

I just wanted to report that 4.7 is working fine for me on i9-9900K and 2080Ti, no dropouts.

[ASRMichael]

On 9/2/2020 at 5:03 PM, StreamFidelity said:

Jussi, first of all I would like to compliment the HQPlayer. The regular free updates are also exceptional and not a matter of course. Thanks a lot for this.

 

I don't want to argue. The forum also serves as customer feedback and I would like to advertise that I can also participate in future updates with a fixed clock frequency. My arguments:

 

1. I may not be the right yardstick. But many use AudiophilOptimizer and also follow some recommendations from it. For example:

 

 

2. Unfortunately, your requirement is no longer valid? In my case, all cores run significantly above this at 4.2GHz.

 

 

3. If I have counted correctly, there are dropouts with 4 users. These are the ones who wrote here.

 

 

But OK. I keep it like @bobflood and stick with 4.6.0. Still, I would be very sorry if I didn't attend future updates. 

Looking at previous posts between you & miska I would like to understand this further. If I’m understanding correctly? We can no longer set core frequency to the preferred CPU speed? If so, I presume this change will filter down to embedded version also?
 

I know a lot of Euphony / HQP users that set CPU frequency to their desired level. Why? From my experience with Turbo on it very much depends on the quality of the PSU/LPS you use. I’m using SJ LPS DC3 which is very good LPS, & I still set my core frequency to 4.6ghz, not the 5ghz it can do. It simple sound better for me. 
 

So taking this functionality away may not be a good thing. Seems step backwards in functionality. Now I could be missing something as to why @Miska has done this? Maybe Miska can explain what the benefits are? 
 

As I said at the beginning I may not be understanding this correctly, so apologies if I’m wrong in my thinking. 

[Miska]

1 hour ago, ASRMichael said:

So taking this functionality away may not be a good thing. Seems step backwards in functionality. Now I could be missing something as to why @Miska has done this? Maybe Miska can explain what the benefits are? 

 

There are various reasons. For example same reason as required minimum macOS version or minimum GPU model keeps changing. I keep my development tools up to date.

 

Sometimes also code complexity increases to support new features. For example Windows 10 wouldn't run anymore, or would run very slowly, on computers that used to run Windows 3.1. Is Windows 10 step backwards in functionality since it cannot run on the old PC's that used to run Windows 3.1?

 

1 hour ago, ASRMichael said:

I know a lot of Euphony / HQP users that set CPU frequency to their desired level. Why? From my experience with Turbo on it very much depends on the quality of the PSU/LPS you use. I’m using SJ LPS DC3 which is very good LPS, & I still set my core frequency to 4.6ghz, not the 5ghz it can do. It simple sound better for me.

 

If it makes difference to you, you should use a NAA instead to isolate the server. CPU computes same output values regardless of clock frequency, so the data DAC sees is not dependent on CPU clock frequencies.

 

[Account Closed]

1 hour ago, ASRMichael said:

Looking at previous posts between you & miska I would like to understand this further. If I’m understanding correctly? We can no longer set core frequency to the preferred CPU speed? If so, I presume this change will filter down to embedded version also?
 

I know a lot of Euphony / HQP users that set CPU frequency to their desired level. Why? From my experience with Turbo on it very much depends on the quality of the PSU/LPS you use. I’m using SJ LPS DC3 which is very good LPS, & I still set my core frequency to 4.6ghz, not the 5ghz it can do. It simple sound better for me. 
 

So taking this functionality away may not be a good thing. Seems step backwards in functionality. Now I could be missing something as to why @Miska has done this? Maybe Miska can explain what the benefits are? 
 

As I said at the beginning I may not be understanding this correctly, so apologies if I’m wrong in my thinking. 

You may be stuck at 4.6 if these changes affect you.

[Miska]

P.S. There are no code changes on my side regarding this one...

 

[madman73]

Hi @Miska,

my setup is built around a DYI dac based on PCM1704 chip and JLSounds I2s USB vIII interface, so only PCM.

The HQPlayer embedded machine is based on Gigabyte Z68MX-UD2H-B3, Intel i7-3770K not overclocked (for now), 4+4 Gb of ram running @1066 mhz XMP-2133 (11-11-11-31). I also use a NAA with linear power supply. In both case i use gentooplayer as OS.

I like a very detailed sound (ie studio monitor sound), so i prefer poli-sync-xtr-mp, LNS15, @705/768Khz but i'm also appreciating Sinc-L.

Can you help me with:

• the DAC bit setting: now i have 32bit, is correct?
• the best HQPlayer filter/NS setting for my DAC (from a technical point of view) if different from my choices, i will listen to it.  
• CUDA Offload and convolution: i've planned to buy a GTX 1660 Super (i think it's sufficient, the CPU now is around 1%/3%); i have to choose between Gigabyte GeForce® GTX 1660 SUPER™ GAMING OC 6G and the new AORUS GeForce® GTX 1660 SUPER™ 6G with more CUDA Core and higher core frequency but lower memory speed and bandwidth..... Which do you suggest?

Thank you.

 

 

[Miska]

25 minutes ago, madman73 said:

my setup is built around a DYI dac based on PCM1704 chip and JLSounds I2s USB vIII interface, so only PCM.

The HQPlayer embedded machine is based on Gigabyte Z68MX-UD2H-B3, Intel i7-3770K not overclocked (for now), 4+4 Gb of ram running @1066 mhz XMP-2133 (11-11-11-31). I also use a NAA with linear power supply. In both case i use gentooplayer as OS.

I like a very detailed sound (ie studio monitor sound), so i prefer poli-sync-xtr-mp, LNS15, @705/768Khz but i'm also appreciating Sinc-L.

Can you help me with:

• the DAC bit setting: now i have 32bit, is correct?
• the best HQPlayer filter/NS setting for my DAC (from a technical point of view) if different from my choices, i will listen to it.  
• CUDA Offload and convolution: i've planned to buy a GTX 1660 Super (i think it's sufficient, the CPU now is around 1%/3%); i have to choose between Gigabyte GeForce® GTX 1660 SUPER™ GAMING OC 6G and the new AORUS GeForce® GTX 1660 SUPER™ 6G with more CUDA Core and higher core frequency but lower memory speed and bandwidth..... Which do you suggest?

 

PCM1704 is 24-bit component, so DAC Bits should be set to at most 24. Although it is not really linear to such precision, so you can try settings from 18 to 21 and check if you get improvement around these figures. 18 and 19 are pretty safe values, 20 and 21 are more on the border. To some extent depends on the chip grade.

 

Filter setting is up to your personal preferences and also tends to depend on source material. Especially for typical RedBook material you can benefit from apodizing filters. You could also try poly-sinc-long-mp (with Adaptive output rate enabled) and poly-sinc-ext2 (although it is linear-phase instead). sinc-S and sinc-L are other possible filter choices.

 

LNS15 is good choice for noise shaper at these rates.

 

Memory bandwidth is quite essential too, but likely either one is good choice for this case.

 

[Account Closed]

On 9/3/2020 at 10:52 AM, bobflood said:

Roll back to 4.6. 4.7 pushes cpu harder. Using 4.6 my cpu (i5 8400) would run at 48-50% total load with ASDM7EC/256, with 4.7 it was being pushed to 53-55% and it could not keep up so I had drop-outs. When I rolled back to 4.6 all worked fine again.

I decided to go with ASDM7EC/DSD128 and use 4.7. I now have use of some non-2s filters, rate family conversion and channel balance which I did not have before. A side benefit is that the modulator cores (0&3) are running just below 100% and my core temps are in a safer zone. I really do not notice much if any difference in SQ and I have the added flexibility. Maybe the non-2s filters are compensating, who knows. My machine was always on the edge with ASDM7EC/DSD256. I have a passively cooled Streamcom case with ASUS Z370I MB and i5 8400 CPU. When I replace it down the line I will go to some water cooled monster like Chris (@computeraudiophile) has built but for now this will have to do.

[StreamFidelity]

On 9/3/2020 at 10:11 PM, ASRMichael said:

Looking at previous posts between you & miska I would like to understand this further. If I’m understanding correctly? We can no longer set core frequency to the preferred CPU speed?

 

Before I go into this question, a little digression. I work in the IT industry (finance) as a business analyst. I do not program but write technical concepts for the developers. When programs are running well, new requests are always asked. After a certain time this leads to an inflation of the code. That's why there are always times when developers review and reduce code.

 

I don't know where HQPlayer stands. I can only say it from the user's point of view. I was hoping that ASDM7EC would be optimized in performance requirement. At the moment I see a (typical) increase in functions. The developer of the HQPlayer may be pursuing a different strategy. Namely the preparation for ASDM7EC with DSD 512. With new, more powerful processors and especially with CUDA. This is his product and he can use it for what ever.

 

With my philosophy of a fanless audio PC, there are limits to his goals. I have created a time series of the different HQPlayer versions below. The CPU frequency is set to either 4.1GHz or 4.2GHz. Very good sound and improvement performance till 4.6.0. The cores are loaded fairly evenly. Since HQPlayer 4 Desktop 4.7.0 the same setting leads to dropouts. Obviously, a turbo boost of up to 5GHz is expected on two cores. This is not possible in a fanless system. A GPU is also ruled out because it cannot really be implemented without a fan. Whatever. For the time being, I'm sticking to 4.6.0, as I don't want to go back to DSD 128 as an alternative.

 

HQPlayer 4 Desktop 4.1.1

 

HQPlayer 4 Desktop 4.2.3

 

HQPlayer 4 Desktop 4.3.2

 

HQPlayer 4 Desktop 4.3.3 

 

HQPlayer 4 Desktop 4.4.0

 

HQPlayer 4 Desktop 4.5.1

 

HQPlayer 4 Desktop 4.6.0

 

HQPlayer 4 Desktop 4.6.0

 

HQPlayer 4 Desktop 4.6.0

 

HQPlayer 4 Desktop 4.7.0 DROPOUTS!

 

[mfalcon]

@Miska what version nvidia driver is required for 4.7 with Ubuntu 20.4.  My card is the 2080Ti

[Account Closed]

6 hours ago, StreamFidelity said:

 

Before I go into this question, a little digression. I work in the IT industry (finance) as a business analyst. I do not program but write technical concepts for the developers. When programs are running well, new requests are always asked. After a certain time this leads to an inflation of the code. That's why there are always times when developers review and reduce code.

 

I don't know where HQPlayer stands. I can only say it from the user's point of view. I was hoping that ASDM7EC would be optimized in performance requirement. At the moment I see a (typical) increase in functions. The developer of the HQPlayer may be pursuing a different strategy. Namely the preparation for ASDM7EC with DSD 512. With new, more powerful processors and especially with CUDA. This is his product and he can use it for what ever.

 

With my philosophy of a fanless audio PC, there are limits to his goals. I have created a time series of the different HQPlayer versions below. The CPU frequency is set to either 4.1GHz or 4.2GHz. Very good sound and improvement performance till 4.6.0. The cores are loaded fairly evenly. Since HQPlayer 4 Desktop 4.7.0 the same setting leads to dropouts. Obviously, a turbo boost of up to 5GHz is expected on two cores. This is not possible in a fanless system. A GPU is also ruled out because it cannot really be implemented without a fan. Whatever. For the time being, I'm sticking to 4.6.0, as I don't want to go back to DSD 128 as an alternative.

 

HQPlayer 4 Desktop 4.1.1

 

HQPlayer 4 Desktop 4.2.3

 

HQPlayer 4 Desktop 4.3.2

 

HQPlayer 4 Desktop 4.3.3 

 

HQPlayer 4 Desktop 4.4.0

 

HQPlayer 4 Desktop 4.5.1

 

HQPlayer 4 Desktop 4.6.0

 

HQPlayer 4 Desktop 4.6.0

 

HQPlayer 4 Desktop 4.6.0

 

HQPlayer 4 Desktop 4.7.0 DROPOUTS!

 

I agree with the passive cooling approach. It is also clear that going forward it will be a limiting factor. The only way to get sustained high core speed will be with active cooling with liquid being the best option. Heat is the enemy. 

[LoryWiv]

With my modest i7700 and 1060 GTX I am enjoying HQP 4.7, prior settings all run at same CPU / GPU utilization as with 4.6. No dropouts. Admittedly, I do not run fanless. I suspect it would be very challenging for @Miskato develop a truly "one size fits all" product development roadmap given the heterogeneity of end users' host systems. It is totally valid to provide feedback, of course, and to maximize applicability to multiple use cases as feasible is the right goal. That said, I think HQPlayer is a specialty product designed for people who seek a powerful set of tools for music playback, format conversion, upscaling etc...and accept that to utilize that requires some system "horsepower". 

 

Just my 2c, I recognize others may have different priorities. Thanks to @StreamFidelity, @bobflood and others for a worthwhile discussion.

[austinpop]

@Miska A couple more areas with questions as I continue to get more familiar with HQPlayer.

1. Clipping and volume settings: As I understood it, the reason to set volume to -3dB or lower is because during oversampling, the reconstructed values may have higher peaks than the original. Is this correct?
   • Is there an SQ advantage to setting Min = Max?
   • Would it be OK to set Min = -3, Max =0, and then lower volume if and when clipping actually occurs? Reason I ask is that I like the dynamics at vol = 0dB, so would like to use that, except when needed.
   • Or is the damage done simply by setting Max = 0dB, instead of -3dB or lower?
2. HQPlayer Pro: I'm intrigued by people's experiences with offline upsampling. Related to this, I was wondering about dither, as I find my preferred setting can vary by track or album. 
   • Would it make sense to pre-upsample in Pro with no dither, and then apply dither only during playback?
   • Example: a PCM-to-PCM upsampling with (sinc-M, none, 768k), followed by playback of (none, LNS15, auto).
   • Would this be equivalent to upsampling with (sinc-M, LNS15, 768k), followed by playback of (none, none, auto)?

Thanks in advance for your clarifications.

[Miska]

27 minutes ago, austinpop said:

1. Clipping and volume settings: As I understood it, the reason to set volume to -3dB or lower is because during oversampling, the reconstructed values may have higher peaks than the original. Is this correct?

 

Yes, that's the case. -3 dB should be enough for 90+%.

 

36 minutes ago, austinpop said:

Is there an SQ advantage to setting Min = Max?

 

No, it just turns off the adjustment possibility. From DSP perspective no difference.

 

42 minutes ago, austinpop said:

Would it be OK to set Min = -3, Max =0, and then lower volume if and when clipping actually occurs? Reason I ask is that I like the dynamics at vol = 0dB, so would like to use that, except when needed.

 

That "dynamics" is called loudness wars, you get better result by leaving volume at -3 dB and turning up the amplifier by 3 dB.

 

In most modern systems you have no loss of dynamics at -30 dB either.

 

46 minutes ago, austinpop said:

Or is the damage done simply by setting Max = 0dB, instead of -3dB or lower?

 

Damage depends Limiter hits. You'd want limiter count to stay at 0 at all times.

 

1 hour ago, austinpop said:

Would it make sense to pre-upsample in Pro with no dither, and then apply dither only during playback?

 

No, you have precision loss. And in addition whenever you don't use dither you have distortion generator.

 

1 hour ago, austinpop said:

Example: a PCM-to-PCM upsampling with (sinc-M, none, 768k), followed by playback of (none, LNS15, auto).

 

That would add some distortion compared to going straight sinc-M, LNS15 output.

 

More DSP you can bundle into one DSP run, the better.

 

1 hour ago, austinpop said:

Would this be equivalent to upsampling with (sinc-M, LNS15, 768k), followed by playback of (none, none, auto)?

 

No, it is not, because intermediate is limited.