Building a DIY Music Server

[Gurkel]

Found this forum because I am also building a server for audio files/streaming at home, but:

 

why are you building super computers or at least machines which are expensive enough to buy one and have enough cpu capacity to idle around most of the time? I won't use Roon but had a quick look at the hardware specs to run it: i3, 4GB RAM, that's it. And for Tidal win7, Intel p4/2.3GHz and 512 MB...

 

My plan is to host the files on a SBC like a pi4 or reuse my rock64 I already own and run something like Moode on another SBC (this time a pi4 because of the software support), push the data stream to a DAC and then to the amp.

[Gurkel]

6 hours ago, bit01 said:

That is the question! In part it seems that most DAC designers are not able yet (have tot figured out how) to mitigate all noise on the input digital signal, which seems to affect the analog converted signal in various forms.

Whatever the case it seems that when you have low noise power supplied to a 'super computer' as you called it, idling most of the time and achieving low latency when computing from both the h/w and a streamlined OS - you have a good result for SQ. Punch (macro dynamics) , 'silent' background revealing more detail, longer (fuller) note decays, timbre etc  all evidently improved on my DACs moving from an NUCi7 to i9-10900K gaming PC configured for best SQ as per guidelines found in many of the pertinent  threads on this forum!

Happy experimenting!

 

ATB

b.

 

Dumb question: do you have some skills in maths/phyiscs and maybe a bit about computer hardware?

 

Because what you are telling me is impossible to achieve the way to plan to do it,.

 

The idea with the multi processors and "optimzed OS" is completely wrong, which is something anyone knows who is overclocking his computer, but let me explain it. Modern CPUs contain multiple cores for the fastest possible way of doing calculations, plus run at very high frequencies. The higher the frequency they run, the faster they calculate. Problem is: they consume a lot of power which must be removed by a dedicated cooler, and the higher the frequency a cpu runs, the more heat must be removed. So heat is a limiting factor of the performance of a CPU, and Intel and AMD do anything to optimize the way how heat is generated.

On trick every CPU with more than one core does, is the following. It analysises which core is busy with doing calculations (and generates heat) and moves the process to different core after some milliseconds. By doing this, the CPU can spread the generated heat over a bigger area. So not a single spot gets really hot but the whole CPU warms up. And this gives the cooler the chance to move more heat away, resulting in a better performance of the CPU due to higher possible frequenciey it can run. This function is hard coded into the CPU and cannot be modified by any software, leave alone an "optimized OS". It also cannot be modified by changing the clock speed, so no matter if you run your computer at 1GHz or 4GHz, it will do the same process swap all the time. Last but not least I am sure that the tidal or Noon server running on a windows machine is not multithreaded but a single thread. If you run such server software on a machine with 40 threads, you have 39 idle threads and the one doing something is bored by the amount of information it has to push around.

 

The "low latency" for audio streams is not required. The digital signal going out on an USB port to the DAC is in the range of 350kb/s, which is an amount of data far far far away from a range where you need some more "horse power" to do the job. My rock64 (a simple SBC) can handle 80MB/sec without issues. And low latency also is not required because the data exchange is asynchronus, which means: the data doesn't have to arrive "just in time" and if the computer is busy doing something else at the moment, everything breaks. A short delay is captured by the buffer on the receiving side.

 

 

I stumbled across the description of one of these prebuild HQ servers (Takio? Taiko?), had a look at the product details and was a bit amused to see that someone really gets away with this and actually sells these computers.

 

 

So, why spend thousands of dollars if it's not required?

[Gurkel]

56 minutes ago, The Computer Audiophile said:

I can tell you the CPU in my PC acts very different from what you suggest. 

 

I run upsampling to 1,536 kHz or DSD256 and convolution. That isn’t going to happen at this level on a Rock64. 
 

 

 

 

 

Can you send a screenshot of the process explorer (windows) or top (Linux)? Wanna see how much CPU power this encoding requires. All cores please, don't think resampling is multithreaded.

 

This still doesn't make the claim with the "modified OS" more true as it doesn't change the hard coded routines in the CPU, and these routines  already do what you try to achieve on OS level - just in a much more efficient way. This OS thing sounds to me like a red hering to me or: you try to solve a problem which doesn't exist. 

 

But now I'll look around a bit more to see what others use.

[Gurkel]

9 hours ago, MarcelNL said:

I think the background is similar to having a power amp with excess power....when do you need the last few Watts if you're listening to the first few mW most of the time.

 

Well, it's more like you are driving around at 30 km/h and tell others you need 500 horsepower and bi turbo so you can accelerate to 32 km/h when you use a 20 core computer to run Tidal which (according to the webpage of tidal) requires 512MB RAM, win7 and a CPU sold 10 years ago.

[Gurkel]

3 minutes ago, ted_b said:

Gurkel,

Welcome to AS.  I think the general tenor of this thread is that a server with rigorous power yet delicate balance puts out the best sound. 

 

Yes, and I also plan to build one, but before wasting money on stuff I don't need on the one side vs buying cheap parts I have to replace immediately I do some research - in boards like this one here. And after stumbling across these high power machines and the claims with the pure sound I started digging around.

 

Because I like computers and my job is to write programs/generate workflows for the data exchange between machines, I had a closer look at the data sent to the USB port by a network streamer and did some calculations.

 

So far I am sure there is no difference in the digital data of such a port. That's why I asked around if there is someone here with the required background knowledge to explain what's going on.

 

 

[Gurkel]

26 minutes ago, The Computer Audiophile said:

There isn’t a CPU currently made that can process DSD512 with EC modulators for playback without dropouts.

 

HQPlayer is the app. It can be coded to be multi threaded. You’re suggestion that resampling isn’t multi threaded is a bit off. It’s up to the developer of the app. 
 

HQP also offloads to the GPU Cuda cores. 
 

 

 

DSD512 is not relevant here, but I found someone here in the board who does upsampling to DSD256, and he is doing this on "They built me a custom fanless, 4 processor i5 computer that runs HQP with output at DSD256. Andrew is the owner and very helpful. Just tell him what are looking to do and your budget and he can advise you.", and another one with "so I finally landed to a passively cooled I5-8400 that is capable to do DSD256/ASDM7"

 

Above rock64 level but below the current cpus - no need for a super computer. And if you push the calculations to the GPU via Cuda, then a fast CPU is not required at all.

[Gurkel]

4 minutes ago, MarcelNL said:

 Just build it, listen to it and be happy...or open your mind and listen to more options and THEN make up your mind which route to go for best Sound Quality. It's not that this bunch of folks likes to invest in hardware just for fun. I don't need a high power CPU for my system at all, yet a fact is that my music sounds much better when it's processed that way.

 

to use a car analogy, a good 4.2L V8 with 350 or so HP accelerates a lot different than a biturbo car with 350 HP...there is no substitute for cubic inches except more cubic inches

 

6 minutes ago, ASRMichael said:

This thread is not about theory, it’s about building a diy server. Clearly you have no plans to follow what other users here have tried.

 

That's why I am asking, and at the same time I come up with my opinion/knowledge and would like to discuss this with someone who knows a bit about the technology used in a computer. 

 

We are talking about good  sound quality here - that's the goal - and how to achieve it. Beside asking around I am also doing my own research and learn some tjings. And I came to the conclusion that PCM (used for data transfer over USB) and DSD are both completly indendenat of any analog influences of the network streamer. As long as you don't run into a ground loop, the data stream cannot be changed by the PSU, by the USB cable or similar things.

 

And then I find this board here where people spend thousands of dollars/Euros on hardware, and I wondered if they did a research or just blindly follow ideas they don't understand.

 

Which brings me back to topic: Anyone here with the required knowledge who can explain the whole stuff? I am talking about PCM encoded analog signals and DSD signals.

 

[Gurkel]

2 minutes ago, MarcelNL said:

@Gurkel I somehow believe we are miscommunicating here, the purpose of this thread is not to lowest grade CPU that can run something but finding the system that offers the best sound quality.

Please see my last response. Anyone here who understands PCM/DSD encoding?

 

If the signal a cheap DIY server pushes to the USB port is identical to what an expensive computer does, why should I buy the expensive one. I haven't heard of any case where someone found out he got different data when running a program on computer A than he got when running the same program on Computer B. Which is great, because otherwise computers would be useless.

[Gurkel]

7 minutes ago, The Computer Audiophile said:

You personally should not purchase such a computer. All good?

 

doesn't answer the nagging question about the relation between the digital audio signal and the influence of analog signals on it (and the resulted sound quality)

 

But otherwise. If I don't do upsampling and try to generate data out of thin air, does your answer mean that the output the computer generates (and therfor the SQ) is independant of the chosen hardware?

 

So if I use an i5 to transform the raw audio data (FLAC, WAV,..) into the data format which goes to the USB port (PCM/DSD) and from there to the DAC/AMP, it's the same data an I7 or a pi4 generates? And same data means same Sound Quality generated by the DAC.

 

So as long as I don't do any heavy calculations in the middle I am fine with a pi?

[Gurkel]

On 2/10/2020 at 11:53 PM, Nenon said:

 

 

I have 8 CPU cores / 16 threads on my server. You can rarely see any of them exceeding 1% CPU utilization and the frequency quite often goes over 4 Ghz and rarely under 3.5 Ghz (without explicitly configuring any CPU frequency settings). This is when the magic happens.

 

BTW, since this a "Building a DIY Music Server" feel free to post some pictures of your build. 

 

Exactly this is the reason I am asking around. Another one was the TO post (page 1) where someone posted the hardware details for hie Tidal streamer.

 

 

And I found this Upscaling Mika machine which is powered by an 6950X. Did a quick comparison with mine, and even at stock speed (3.6 Ghz) the Intel is only 3% faster - which is ok but nothing special. But As I don't plan to generate data out of thin air, I don't require this.

[Gurkel]

15 minutes ago, Mops911 said:

As I understand and learned here, all(?) DAC upsample internally. I read Hugo Chord, a 10k DAC, uses less power if fed upsampled data. So as I understand, there is consnsus here for the reason that a powerful CPU can upsample much better than a feeble DAC. Better filers, noise shapers, more Taps, what have you

 

Only a side topic here. I don't do it, I am just interested in DIY server for streaming, so back to topic.

 

Page 1, first comment and TO of this thread:

 

 

 

Quote

 

Hi everyone! I am building a computer for someone else and decided to share what I am doing with everyone. 

 

Let me start with some of the high level requirements:

• One box solution to eliminate some of the clutter.
• Optimized for Streaming (Tidal / Qobuz).
• The best quality USB output

 

•  Intel Core i9-9900K
•  
•  
•  
• USB Output: PinkFaun USB Bridge with ultraOCXO clock

 

•  

 

There we go again. My knwoledge tells me the PCM/DSD signal of an USB port is not influenced by an analog signal, and you don't need an i9-9900K to stream Tidal/Qobuz.

 

Who knows how PCM/DSD works?

[Gurkel]

6 minutes ago, Mops911 said:

If your server doesnt, your DAC will.

 

Dude, you only scratched the surface. Read on, people get obsessed with clocks and phase noise (frequency domain) or Jitter (time domain). You see Nenon soldering better clocks on this motherboard....

 

 

It is SO much more than brute power and nobody knows. But in the end, its your system, you can grind it and eat it :-D

 

 

 

 

Leva that aside. If you send data over the internet, each data package has it's own CRC code attached, and if the packet fails the check it is transitted again (if you use TCP). No switch alters data, there is no data clenaisng of whatsoever analog side effects here. Pretty obvious but people buy it.

[Gurkel]

On 5/2/2021 at 5:43 AM, auricgoldfinger said:

It seems to me this thread went completely off the rails about 3 pages ago.  I suggest deleting them would keep the thread properly focused.  This discussion rapidly devolved into the usual objective/subjective arguments.

Maybe the reason was because my data was transferred over WiFi which altered how it sounds and added a bit of a harsh sub tone.