[johnnym]

I have now tested Src-dx and also a rpi4 with Hqplayer. I think, the benefit of Src-dx (or any DDC for that matter) is dependent on the usb implementation of the source. In my case, I hear no difference between direct usb vs Src-dx dual bnc, where the source is a two year old MacBook Pro.

I also found out that for upscaling 16/44.1 kHz files to 24/768 kHz, rpi4 would not work with all the higher end filters such as sinc-M, sinc-Mga, etc. I even confirmed with the Hqplayer developer on audiophilestyle forum, that these filters would not work for 16x upsampling with rpi4. These filters do however work with 24/96 kHz and higher. Also, rpi4 can work as a great network audio adapter for Hqplayer running on a powerful pc nearby. I will therefore stick to the MacBook Pro for now.

This is good to know, thank you.
I have an RPi4 running as NAA in my trial using my MacBook Pro. I thought another RPi4 I have could be used to run the embedded version of HQPlayer for PCM conversion to 768, but I couldn’t get it installed on DietPi. Now I don’t have to slog through install issues only to find it won’t work anyway.

If I can get it working on my MacBook I’ll get a NUC or something more powerful.

 

[87mpi]

If you really want to use an endpoint, throw away that rpi4 stuff made just for playing (including all toys with ARM cpu) and use at least one Nuc with i5 /i7 with x86 cpu (even if mobile class) possibly linear powered if you want really start to really enjoy the music!

If you want even more, use a PC with a very powerful desktop (or even server) class CPU, even if it will only be used as an endpoint.

You will see what a surprise.

 

[MrWalkman]

On an Intel i7-12700H sinc-long (with LNS15) goes to up to 15-18% CPU utilization upsampling from 44.1k to 96k, which is basically unnoticeable. If I enable CUDA offload, then GPU usage goes up to ~65% sometimes, on a 3060 GPU.

Today I tried with the Sony WM1A in USB DAC mode, which supports up to 384k. I had CUDA offload disabled, and I realized that playback is not possible like this with the Intel i7-12700H, it's just choppy, as the CPU can't keep up or something. Enabling CUDA offload gets the GPU at around 80-100%, but at least playback is stable.

Sounds really nice, I like this filter!

 

[kumar402]

If you really want to use an endpoint, throw away that rpi4 stuff made just for playing (including all toys with ARM cpu) and use at least one Nuc with i5 /i7 with x86 cpu (even if mobile class) possibly linear powered if you want really start to really enjoy the music!

If you want even more, use a PC with a very powerful desktop (or even server) class CPU, even if it will only be used as an endpoint.

You will see what a surprise.

What’s wrong in using Pi4 just as NAA as all the processing is done outside and with OS like GentooPlayers or RopieeXL and powered by good LPS it can infact outperform NUC running heavy processing and bloated OS. GentooPlayers can really optimize how Pi works.

 

[Rayon]

What’s wrong in using Pi4 just as NAA as all the processing is done outside and with OS like GentooPlayers or RopieeXL and powered by good LPS it can infact outperform NUC running heavy processing and bloated OS. GentooPlayers can really optimize how Pi works.

I have similar thoughts. Less potential interference due to simpler infrastructure.

 

[shwnwllms]

I use an UP Board with Jussi’s image. This is what he recommends and what he uses in his setup for testing. If it’s good enough for him it’s good enough for me. Sound wise I wouldn’t expect much difference in an NAA but it’s been discussed elsewhere on other forums. Main difference is processor compatibility with rates up to DSD1024 (few NAA’s will accomplish that)

https://up-shop.org/default/up-gate...g-memory-32g-emmc-board-w-vesa-plate-b10.html

Edited: To clarify “processor compatibility” see my subsequent post regarding AMD/Intel vs ARM BIOS architecture differences

 

[Rayon]

Main difference is processing power and compatibility with rates up to DSD1024 (few NAA’s will accomplish that)

But NAA is effectively just passing through data bit-perfect. It should be very light and trivial operation. I agree that compatibility is a different matter. Also I'm not sure about their internal clocks as such. For example Holo May reclocks everything anyway, so the accuracy of the clock shouldn't matter, but I'm not sure about other dacs.

 

[GoldenSound]

But NAA is effectively just passing through data bit-perfect. It should be very light and trivial operation. I agree that compatibility is a different matter. Also I'm not sure about their internal clocks as such. For example Holo May reclocks everything anyway, so the accuracy of the clock shouldn't matter, but I'm not sure about other dacs.

The only time the clock of the streamer/source device will matter is if you're using a synchronous protocol like AES/SPDIF/I2S.

If you're using USB the DAC's clock is in control of all timing

 

[GoldenSound]

If you really want to use an endpoint, throw away that rpi4 stuff made just for playing (including all toys with ARM cpu) and use at least one Nuc with i5 /i7 with x86 cpu (even if mobile class) possibly linear powered if you want really start to really enjoy the music!

If you want even more, use a PC with a very powerful desktop (or even server) class CPU, even if it will only be used as an endpoint.

You will see what a surprise.

Don't use a beefy PC just as an endpoint. You're just needlessly introducing a noisier source to your chain.

 

[87mpi]

Let's start from the assumption that listening with headphones is absolutely not comparable with listening in a room with loudspeakers. Two totally different worlds. But let's move on. Concepts are of little use. It's the proof that counts. And the ears don't lie. I assure you that in 5 years I have personally tried every sort of endpoint device from rpi3, rpi4, to usbridge, usbridge signature, nuc i3, i5, i7 and laptops, desktops and servers, in practically any possible configuration with different linux operating systems like audiolinux, Gentooplayer (purchased several licenses), ropieexl etc,.. Euphony v3, v4, v4+stylus, v4+HQPe etc... I tested almost all players out there, Roon, roon rock, roon +HQPe (purchasing the license) configurations with 1 pc, 2 pc, 3 pc separating everything possible and on a network with router powered in line with excellent ethernet cables. I tested any windows 10 version (1507,1511,1607,1709,1903,1809,ltsc 2019,ltsc 2021,21h2,22h2 Windows 11 and jriver,jplay,foobar,HQPlayer desktop 4 Hqplayer desktop 5 (tried few days ago) and got to this conclusion. I have a holo spring 3 KTE, a really amazing dac. It plays pcm and dsd almost at the same level. In my opinion pcm is to be preferred even if not by much. Believe me. Leave measurements aside and listen. The ears don't lie. Too often we think that a choice on paper should be the best because actually on paper it doesn't make sense, but when tested it turns out to be the wrong choice. We should just try and listen without preconceptions and prejudices. I repeat. Believe me. Try, try and try again and again. Don't take what they tell you or what you read as truth, just because you think it is. In this world, people are always wrong. Question everything. I don't take money indeed I have spent so much that if I think about it I start to cry. I am not a fan boy of any brand and I do not sponsor anyone. But it really helped me. This is why I feel like recommending these things, the result of my direct personal experience in the field. Free to believe it or not.

Greetings to all fans and music lovers.
Antonio

 

[GoldenSound]

Let's start from the assumption that listening with headphones is absolutely not comparable with listening in a room with loudspeakers. Two totally different worlds. But let's move on. Concepts are of little use. It's the proof that counts. And the ears don't lie. I assure you that in 5 years I have personally tried every sort of endpoint device from rpi3, rpi4, to usbridge, usbridge signature, nuc i3, i5, i7 and laptops, desktops and servers, in practically any possible configuration with different linux operating systems like audiolinux, Gentooplayer (purchased several licenses), ropieexl etc,.. Euphony v3, v4, v4+stylus, v4+HQPe etc... I tested almost all players out there, Roon, roon rock, roon +HQPe (purchasing the license) configurations with 1 pc, 2 pc, 3 pc separating everything possible and on a network with router powered in line with excellent ethernet cables. I tested any windows 10 version (1507,1511,1607,1709,1903,1809,ltsc 2019,ltsc 2021,21h2,22h2 Windows 11 and jriver,jplay,foobar,HQPlayer desktop 4 Hqplayer desktop 5 (tried few days ago) and got to this conclusion. I have a holo spring 3 KTE, a really amazing dac. It plays pcm and dsd almost at the same level. In my opinion pcm is to be preferred even if not by much. Believe me. Leave measurements aside and listen. The ears don't lie. Too often we think that a choice on paper should be the best because actually on paper it doesn't make sense, but when tested it turns out to be the wrong choice. We should just try and listen without preconceptions and prejudices. I repeat. Believe me. Try, try and try again and again. Don't take what they tell you or what you read as truth, just because you think it is. In this world, people are always wrong. Question everything. I don't take money indeed I have spent so much that if I think about it I start to cry. I am not a fan boy of any brand and I do not sponsor anyone. But it really helped me. This is why I feel like recommending these things, the result of my direct personal experience in the field. Free to believe it or not.

Greetings to all fans and music lovers.
Antonio

I have a two channel set-up as well as my headphone set-up. And I've tested a large number of devices in the digital source category as well, always doing my subjective evaluation before any measurements.

I've yet to find a situation where when comparing two devices, the one that sounded better didn't turn out to be better in terms of jitter performance or output noise when I measured them after. That after all is the reason we spend (sometimes a LOT of) money on these devices, to get ones that will bring improvements even if many hardcore objectivists would claim they don't matter or would be inaudible.

IMO no one should rely entirely on measurements, to do so would assume we can build a complete picture of how something sounds from the available measurements which we cannot yet do.
But you also should never rely only on sighted listening tests because to do so would be to assume we are immune to expectation bias, placebo, and also just to assume that we should not bother to understand WHY something sounds better. Which would be silly.

Using a beefy PC for an extremely simple task just means you're producing tons more noise. If that effect is preferable to you that's fine, but my experience has been quite the opposite and IMO getting PCs generally out of your audio chain is one of the biggest upgrades you can make.

 

[tdx]

I have a two channel set-up as well as my headphone set-up. And I've tested a large number of devices in the digital source category as well, always doing my subjective evaluation before any measurements.

I've yet to find a situation where when comparing two devices, the one that sounded better didn't turn out to be better in terms of jitter performance or output noise when I measured them after. That after all is the reason we spend (sometimes a LOT of) money on these devices, to get ones that will bring improvements even if many hardcore objectivists would claim they don't matter or would be inaudible.

IMO no one should rely entirely on measurements, to do so would assume we can build a complete picture of how something sounds from the available measurements which we cannot yet do.
But you also should never rely only on sighted listening tests because to do so would be to assume we are immune to expectation bias, placebo, and also just to assume that we should not bother to understand WHY something sounds better. Which would be silly.

Using a beefy PC for an extremely simple task just means you're producing tons more noise. If that effect is preferable to you that's fine, but my experience has been quite the opposite and IMO getting PCs generally out of your audio chain is one of the biggest upgrades you can make.

Yes but isn't a beefy PC a necessity to use HQPlayer's most advanced filters? So if that's the case, are you recommending a PC with HQplayer but then something like a Sonore Rendu or some other network addon in the chain so the PC isnt directly connected to the DAC via USB?

Just want to make sure I follow your thought process....

 

[GoldenSound]

Yes but isn't a beefy PC a necessity to use HQPlayer's most advanced filters? So if that's the case, are you recommending a PC with HQplayer but then something like a Sonore Rendu or some other network addon in the chain so the PC isnt directly connected to the DAC via USB?

Just want to make sure I follow your thought process....

It is a requirement yeah, the PC doing the HQP Processing needs to be powerful enough to handle it, but the NAA task is exceptionally easy. It's just passing the data to the DAC and not doing any actual real computing.

The optimal setup would be to have a beefy PC that can handle whatever HQP config you want, and then stream to a high quality network endpoint.
If using USB then you just need the USB output to be clean and there's a wide range of devices that'll suit that.
If using I2S/AES/SPDIF then the streamer's clock is important and so you can end up spending a lot to get the best result.
Which of the two is better will depend on your particular DAC and your particular streamer.

 

[ra990]

Correct, the Hqplayer upscaling runs on the server and the audio packets are sent via TCP into the client NAA endpoint, which simply passes them from the network over to the USB in FIFO fashion. Theoretically, the lightweight NAA endpoint isn't using much CPU or power and is going to be less noisy into the connected DAC.

 

[87mpi]

Theoretically... But practically is another thing.

Even if on paper the endpoint performs a very simple task, delegating the entire processing to the server, the device connected to the dac is fundamental, and its cpu is even more so.
It uses as endpoint a device with ARM cpu (like a raspberry, a rendu etc.. ), then a mobile x86 (like a notebook or a nuc), then another one of desktop class (like a common pc).

With the same expenditure (practically almost zero) and configuration, the device with a more powerful CPU will sound better, even if it should theoretically consume, heat up and produce more noise on it.

Using the network introduces much more noise than that produced by a CPU.

The sound has a completely different weight.

Seeing is believing.