Design a PC/Server for ROON and HQ Player

[Jud]

10 minutes ago, Bullwinkle said:

Now that document interests me for the following reason. Although I haven't been here long I've tried to read deeply into a few threads where folks discuss hardware. I have not noticed CA denizens stressing a need for ECC memory (though my short time here means I may have missed it). Right out of the gate this FreeNAS document is making ECC sound like a necessity; NONE of the processor/chipset combos we've lately discussed here support ECC memory.

 

I wonder whether any CA denizens have strong beliefs about ECC.

 

This is for server-level reliability without regard for ultimate gaming speed.  ECC is what you want for that use case.

[Bullwinkle]

@tboooe It's the Supermicro c7z270-CG (which may also be listed with an -L or -B suffix which just refers to packaging, L being the retail package and B being the bulk package).

[Miska]

5 hours ago, Bullwinkle said:

Now that document interests me for the following reason. Although I haven't been here long I've tried to read deeply into a few threads where folks discuss hardware. I have not noticed CA denizens stressing a need for ECC memory (though my short time here means I may have missed it). Right out of the gate this FreeNAS document is making ECC sound like a necessity; NONE of the processor/chipset combos we've lately discussed here support ECC memory.

 

I wonder whether any CA denizens have strong beliefs about ECC.

 

I have not felt that I would have encountered so many memory errors in my audio computers that I would need ECC. Instead I just try to get high speed DDR4 with as low as possible CL.

 

I do have ECC on my development workstation, but that's just a Xeon E5 workstation purchased as-is with on-site warranty. Not something I built myself. But I would not necessarily buy it as machine just for running HQPlayer.

 

[Miska]

5 hours ago, Bullwinkle said:

 

I can believe this is true compared to ASUS and MSI. One of the reasons I went with a Supermicro z270 board was that they use Nichicon's highest endurance class of capacitors.

 

Supermicro is not really available here, so that's why I have not paid attention to it.

[Miska]

6 hours ago, volpone said:

 

Thank you Miska. I'm also interested in GB Mobo for a future powerful HQP / Roon server setup.

But, when using NAA ethernet connected client on a distinct and distant nanodevice like SBC (RPI, Cubox, MicroRendu ...), do you care much about isolation and "noisy" server HW ?

 

Yes, it doesn't matter in case NAA is used. But the machines in question are directly connected to DACs, so that's why I paid a bit more attention to that particular detail.

 

[leftside]

I'm thinking of setting up a dedicated box for Roon and HQP. I am now 2 days into installing and configuring Roon + HQP with my existing two computers:

PC server: Windows 10 Pro, I5-4590 3.3 GHz 4 cores 4 logical processors, 8 GB RAM, GTX 970 (running Roon and HQP)

Laptop NAA: Windows 10 Pro, I7-6600 2.6 GHz 2 cores 4 logical processors, 16 GB RAM (running Roon)

 

I have it all up and running and upsampling to DSD256 for my Lampizator Atlantic DAC connected to my laptop. I'm liking the results so far, but the sound occasionally cuts out due to my wireless network (I presume). I'm going to try running Roon and HQP on the laptop, but I think I need a "dedicated" box(es?) connected directed to the DAC for optimal results. 

 

Are there links somewhere that I can investigate further in regard to this dedicated music transport? I could build one up myself with various components, though to save time I'd prefer to purchase something prebuilt.

[tboooe]

3 hours ago, leftside said:

I'm liking the results so far, but the sound occasionally cuts out due to my wireless network (I presume).

I am able to listen to DSD128 and Redbook streaming from my NAS via a wireless AC network with zero issues.  

 

What kind of network do you have?

How is the server communicating with the NAA?  

What are our HQP settings?

 

Just for a datapoint, using an old i7-3770s CPU, I am able to convert to DSD512 using the -2 filters.  

[leftside]

I have the following network setup: 

Router: NETGEAR Nighthawk X6 Wireless AC3200 Tri-Band R8000

The PC is connected to the router via ethernet

The laptop is connected wirelessly via the 802.11ac protocol on the 5 GHz band

I'm running the HQP networkaudio.exe on the laptop to enable NAA

 

HQP filter settings:

poly-sinc-shrt, ASDM7, 44.1k * 256

 

Maybe I should look at the -2 filters? I'll also do some more reading tonight. Thanks.

 

 

[tboooe]

A couple of things to look into:

 

First check out the CPU usage on your server pc.  

Try connecting the NAA laptop directly to the router and see what happens.

Definitely try the -2 filters esp if your CPU usage is high

Do you have any other devices on the 5ghz band?  

[leftside]

All good tonight. Most of the filters cause the CPU(s) to run between 40% - 60%. The poly-sinc-hb seems to be a bit higher. I can't run the regular xtr filters without the PC stuttering. 

 

A lot of the filters sound really good. I seem to be leaning towards the poly-sinc-short filter the most, and most definitely prefer DSD over PCM. I'll experiment more with the various filters and various types of music, but anyway I digress. After 1 afternoon (Sunday) and two evenings (Monday and Tuesday) I think I'm all set up.

 

I'd still like to compare my laptop to a custom/dedicated music transport and also with a higher spec PC server to be able to run the xtr filters.

[Miska]

AMD has also launched some interesting new products...

 

http://hothardware.com/news/amd-ryzen-9-threadripper-cpu-lineup-41ghz-16-core-32-thread-beast

 

http://hothardware.com/news/amd-naples-zen-architecture-makes-epyc-debut-for-datacenter-market

 

[Solstice380]

6 hours ago, leftside said:

 

All good tonight. Most of the filters cause the CPU(s) to run between 40% - 60%. The poly-sinc-hb seems to be a bit higher. I can't run the regular xtr filters without the PC stuttering. 

 

 

That level of CPU utilization will give some stuttering.  I have an i7 6700K with GTX970 GPU and can't run the non-2s xtr filters.  Not much OS optimization, though.  

 

I find the non-mp filters to sound more like delta-sigma DACs, even through the chipless DSD512 module in my Golden Atlantic.  I tend to prefer the -mp filters when upsampling to the higher DSD rates.  Miska has said that the -mp filters are the best (most accurate?) for transients.  Maybe I'm more sensitive to the pre-ringing of linear filters that I feel make the sound harsh on even well recorded jazz and rock.

[christoph]

7 hours ago, leftside said:

and most definitely prefer DSD over PCM

Do you have the separate chipless DSD engine in your Atlantic or are you using the DSD part of the R2R PCM/DSD256 board?

[tboooe]

2 hours ago, Solstice380 said:

That level of CPU utilization will give some stuttering

My CPU is at 70-80% when converting to PCM to DSD512 and I dont get any stuttering.  I do get occasional stuttering when I converting DSD t64 or 128 to DSD512 though.  I also do 2 ch convolution.

[Bullwinkle]

Is it possible that the stuttering is due to processor differences of some sort? E.g., because tboooe's  i7-3770s has Intel hyperthreading (so that the four cores can "be" eight threads) while leftside's i5-4590 does not (so that the four cores are four threads)? At any rate it would be useful for everyone on this thread to know the source of the difference observed here, I think.

[tboooe]

Just now, Bullwinkle said:

 

Is it possible that the stuttering is due to processor differences of some sort? E.g., because tboooe's  i7-3770s has Intel hyperthreading (so that the four cores can "be" eight threads) while leftside's i5-4590 does not (so that the four cores are four threads)?

 

Good observation.  However, I have disabled hyperthreading.  

 

I do wonder if the stuttering is due to @leftside connecting his NAA via wireless versus having it hardwired to this network.

[Solstice380]

13 minutes ago, tboooe said:

My CPU is at 70-80% when converting to PCM to DSD512 and I dont get any stuttering.  I do get occasional stuttering when I converting DSD t64 or 128 to DSD512 though.  I also do 2 ch convolution.

 

You are lucky!  What are your individual cores at?  What CPU, etc.?  50% on a core can be max for some processors.  I can do DSD 64/128 to 512 without issues with non-2s filters, except xtr so I use the 2s variant.  So many variables in our servers / networks it's often difficult to determine where the issue is. 

[Solstice380]

7 minutes ago, tboooe said:

I do wonder if the stuttering is due to @leftside connecting his NAA via wireless versus having it hardwired to this network.

 

Good point!

[tboooe]

8 minutes ago, Solstice380 said:

What are your individual cores at?  What CPU, etc.?

My individual cores are at 70-80%.  I have multicore DSP enabled as well to further ensure the CPU load is evenly spread between the Cores.  My CPU is an old i7-3770s with hyperthreading turned off.

[ted_b]

4 minutes ago, Solstice380 said:

I can do DSD 64/128 to 512 without issues with non-2s filters, except xtr so I use the 2s variant.  So many variables in our servers / networks it's often difficult to determine where the issue is. 

I thought this too, but Jussi reminded me (and I posted about it) that DSD-to-DSD does NOT use filters, but instead only modulators and the seldom-used integrators from the DSF/DIFF settings.  So I continue to ask, as should you, why does setting filters affect performance when doing DSD-to-DSD if they are not used?

[leftside]

5 hours ago, Solstice380 said:

 

That level of CPU utilization will give some stuttering.  I have an i7 6700K with GTX970 GPU and can't run the non-2s xtr filters.  Not much OS optimization, though.  

 

I find the non-mp filters to sound more like delta-sigma DACs, even through the chipless DSD512 module in my Golden Atlantic.  I tend to prefer the -mp filters when upsampling to the higher DSD rates.  Miska has said that the -mp filters are the best (most accurate?) for transients.  Maybe I'm more sensitive to the pre-ringing of linear filters that I feel make the sound harsh on even well recorded jazz and rock.

My processor seems to be ok with that range, but anything above 60% brings it to it's knees. Mostly it is around 40% but sometimes spikes to 60%. Same as you - can't run the non-2s filters.

[leftside]

4 hours ago, christoph said:

Do you have the separate chipless DSD engine in your Atlantic or are you using the DSD part of the R2R PCM/DSD256 board?

I don't have the separate chipless DSD engine (maybe next year....) and am using the DSD part of the R2R PCM/DSD256 board.

[leftside]

2 hours ago, tboooe said:

Good observation.  However, I have disabled hyperthreading.  

 

I do wonder if the stuttering is due to @leftside connecting his NAA via wireless versus having it hardwired to this network.

I'm pretty sure it's the processor on the server. As mentioned, the xtr filters cripple it (I RDP into the PC from the laptop to view the CPU with Process Explorer). It's ok with spikes up to 60%, but is far happier in lower ranges.

 

The laptop will also sometimes cause the sound to cut out momentarily if I browse the Internet to a "demanding" site. This for me is reason enough to consider a dedicated music transport connected to the DAC (which I could control from the laptop).
 

[leftside]

Looks like this should do the job for a one-box solution for Roon+HQP doing PCM to DSD256 and PCM to DSD512 conversion with all filters (stereo). I would install Windows 2012 R2 server and Audio Optimizer myself.

https://www.endpcnoise.com/quiet-music-servers/streacom-music-server

 

Asus Z-170 Deluxe mobo 1151/Skylake

i7-7700 Kaby Lake 1151

16GB Corsair DDR4 2666 (or would 32GB be preferable?)

Samsung 850 Evo 250GB SSD

SOtM Audiophile PCIe to USB

1 * SOtM In-Line Audiophile SATA Power Noise Filters (would I need more than 1 filter?)

 

 

 

[christoph]

5 hours ago, leftside said:

I don't have the separate chipless DSD engine (maybe next year....) and am using the DSD part of the R2R PCM/DSD256 board.

the DSD part of the R2R board is already quite good, but the chipless DSD engine is IMHO quite a lot better. So be prepared to even better sound as soon as you order the chipless DSD engine