A novel way to massively improve the SQ of computer audio streaming

[bodiebill]

30 minutes ago, Superdad said:

Amazing to see all this discussion of power supplies without a single mention of one of nearly the most consequential factor, output impedance. Computers are a very “bursty” load, and given adequate current and even just average noise performance, the supply with the lower broadband output impedance will always win.

 

Can this be measured if it is not mentioned in the manual/specs?

[bodiebill]

On 1/16/2020 at 9:39 PM, bodiebill said:

Thanks Nenon, that makes sense. I guess experimenting with several setups will tell.

But it looks like this would a good starting point:

- SBooster 19V/1.75A => HDPlex 800W DC-ATX => motherboard

- HDPlex 200W 12v/10A => CPU

- SR4 12V/2A => USB (iFi micro iUSB3.0)

 

Or for the latter, even better (for additional power cleansing) : 

- SR4 12V/2A => MPAudio DLS-HPULN 10.5V/5A(2A) => USB (iFi micro iUSB3.0)

 

Alternatively, looking at your last remark (multiple rails of same PS):

- HDPlex 200W 19v/10A => HDPlex 800W DC-ATX => motherboard

- HDPlex 200W 12v/10A => CPU

- SR4 12V/2A => MPAudio DLS-HPULN 10.5V/5A(2A) => USB (iFi micro iUSB3.0)

 

After fabricating another short DC cable with Supra CAT8 to replace the stock HDPlex cable for the motherboard, I am now powering as follows:

 

- HDPlex 200W 19v/10A => HDPlex 800W DC-ATX => motherboard

- HDPlex 200W 12v/10A => CPU

- HDPlex 200W 12v/2A (variable output) => MPAudio DLS-HPULN 10.5V/5A(2A) => USB (iFi micro iUSB3.0) => Amanero/Lampizator

 

This is very nice sounding with great slam and dynamics, and not strident. I begin to wonder whether whether my dissatisfaction was with the HDPlex LPSU or its DC cables. With serious cables there seems to be a clear improvement!

 

For the USB I did try with (12V => 10.5V) and without (12V) the MPAudio (which takes 9-12V), but the former is clearly better. More rest and ease! So the MPAudio stays in for now.

 

So for now I am listening to this setup, which has the beauty that I am using one AC power cable less.

 

In a few weeks I expect a Denafrips Terminator, which some say makes the likes of iFi micro iUSB3.0 or ISO Regen redundant...

[bodiebill]

On 1/17/2020 at 5:10 PM, austinpop said:

Yes, please publish your findings here when you get them, along with build details and the exact workload (OS, music player, if HQP - what settings) you used. One useful piece of info that will be useful to others will be the relative consumption between ATX and EPS (CPU). 

 

Yes, again - this is not meant to be an exact science. Typically, your PSU should have a capacity several multiples of your observed consumption. But it will tell you, for example the relative weight of ATX and EPS, so you can decide how/where to allocate your resources into the capacity of each rail.

 

Also be sure to observe other infrequent, but intensive workloads, like: library scanning, OS and software updates, for example.

 

And finally, of course, a reminder to look at both peak and steady state.

 

A Brennenstuhl power meter arrived today and I did some measurements.

I may need some help here as I am somewhat baffled by the first results. Please note that I only measured my minimal audio endpoint PC with AudioLinux headless (meaning GUI-less) + NAA. (I am not in a hurry to measure my heavy-lifting server as I am happy with its current power supply.)

 

First the audio PC specs:

mobo:    Supermicro X11SAE-M
cpu:    Intel Core i3-6100T (35W TDP)
memory:    2x Apacer D31.23185S.001 4GB ECC DDR4 2666MHz (wide temperature)
PSU internal:    HDPlex 800W DC-ATX converter

NIC:    Intel X520-D1

HD/SDD:    none

 

CPU / EPS

I used a simple 12V SMPS:

- steady state (nothing running): 0.04A

- playing any kind of audio file (redbook, hi-res PCM or DSD) streamed from the HQPlayer server): 0.14A

 

ATX

I used a 19v Chinese linear PS, wattage unknown as I did not have an SMPS that fitted:

- steady state: 0.17A / 30W

- playing music: does not change, always at 0.17A / 30W (does not change)

 

So maximum total current drawn approx. 0.3A. Such low amperage? It does not make much sense to me. But it is consistent with the fact that I used to power the whole PC (ATX+EPS) with a 36W / 1.75A / 19V SBooster without problem.

 

Probably someone here can tell me what I am doing wrong?

[bodiebill]

10 hours ago, Nenon said:

Your equipment is not capable to catch microbursts. If it was able to update let's say every nanosecond, you would see bursts that might be even hundreds of times higher. 

 

OK, it was worth a try.

 

The 100 multiple that you mention for bursts may be correct for gaming servers or servers running HQPlayer with EC modulators, but will be much lower for my audio endpoint which ONLY receives an audio stream and passes it true. Actually the audio endpoint's burst multiple should be lower than approx. 6 as a 1.75V power supply could handle everything (ATX + EPS together) without a problem for more than a year.

[bodiebill]

2 hours ago, austinpop said:

While your meter cannot give you microbursts as @Nenon says, you can still get useful info.

 

Since I don’t know what AC voltage your location has, these current values are hard to interpret. Try setting the meter to display watts, rather than current. Look at watts during

- power on

- system idle

- music playback

Assuming you’re in a 220v country, your data already shows something interesting:

- EPS and ATX draw about equal amounts of power: 220x0.14 = 31W approx, and 220x0.17 = 37W approx. Validate this by remeasuring watts on your meter.

 

Assuming these power measures are correct, I bet you would benefit from a much heftier PSU - of equal or better quality - to power both ATX and EPS: 6a, 10a or even higher.

 

Remember - amps on your meter are AC amps at 220v, while the PSU current we’re talking about are DC at 19v or 12v.

 

Thanks @austinpop !

 

I measured the ATX and EPS/CPU wattage drawn and the results are as follows. The AC here delivers 230V, so based on the wattage I calculated the current: amperage = wattage divided by 230.
 

	ATX		EPS/CPU
max values:	W	A	W	A
start cycle	21.8	0.09	15	0,07
system idle	19.1	0.08	4.5	0.02
playing music	19.4	0.08	7.4	0.03
stop cycle	18.7	0.08	7.8	0.03

 

Pretty low maximum values overall, as measured per cycle/use. The HDPlex I am now using delivers 10A which seems overkill, so I would not hesitate to use my soon-to-come 38W/2A SR4 for either ATX or EPS, or even both. Just to find out which sounds best: the 'best' or the most powerful PS...

 

Apparently, for this system, it is true that any PS that can successfully post and boot the PC will be more than enough to handle all running processes.

 

Interestingly enough the EPS draws less than the ATX, so Nenon's rule "put the most powerful power supply on the CPU" may apply to a powerful server with high-TDP CPU, but not to such a minimal endpoint PC.

 

 

[bodiebill]

1 hour ago, austinpop said:

Exactly. Listen for yourself, and see which you prefer. 

 

I already compared:

 

(1) HDPlex 200W 10A 12V rail => EPS/CPU

with

(2) HDPlex 200W 2A LT3045 15V (variable) rail => MPAudio DLS-HPULN 5A 12V out => EPS/CPU

 

In theory the MPAudio can supply 5A but this must be limited to the 2A of the HDPlex. 

15V to 12V is a big drop. This has the advantage of lower ripple but could generate a lot of heat. Not in this case: while playing music the MPAudio hardly becomes lukewarm as the current is so low.

 

And (2) is the clear winner: better bass, more presence and less strident highs. So in this case the 'higher quality' PS that is the result of several LT3045's wins from the powerful PS.

[bodiebill]

On 1/19/2020 at 9:10 PM, bodiebill said:

Thanks @austinpop !

 

I measured the ATX and EPS/CPU wattage drawn and the results are as follows. The AC here delivers 230V, so based on the wattage I calculated the current: amperage = wattage divided by 230.
 

	ATX		EPS/CPU
max values:	W	A	W	A
start cycle	21.8	0.09	15	0,07
system idle	19.1	0.08	4.5	0.02
playing music	19.4	0.08	7.4	0.03
stop cycle	18.7	0.08	7.8	0.03

 

Pretty low maximum values overall, as measured per cycle/use. The HDPlex I am now using delivers 10A which seems overkill, so I would not hesitate to use my soon-to-come 38W/2A SR4 for either ATX or EPS, or even both. Just to find out which sounds best: the 'best' or the most powerful PS...

 

Apparently, for this system, it is true that any PS that can successfully post and boot the PC will be more than enough to handle all running processes.

 

Interestingly enough the EPS draws less than the ATX, so Nenon's rule "put the most powerful power supply on the CPU" may apply to a powerful server with high-TDP CPU, but not to such a minimal endpoint PC.

 

For my audio endpoint I made new measurements on the ATX part as I found a 19V SMPS for it. Now the comparison is more fair as I used SMPS'es for both ATX and EPS/CPU. The result is:
 

	ATX		EPS/CPU
maximum	W	A	W	A
startup cycle	19.1	0.08	15	0.07
idle	16.7	0.07	4.5	0.02
playing music	16.9	0.07	7.4	0.03
stop cycle	17	0.07	7.8	0.03

 

Power draw at the ATX side (via HDPLEx 800W DC-ATX rail 1) is even lower than before when measured with the LPSU, but still higher than at the EPS side.

 

Note that the W & A values when playing music are independent of the audio format (redbook, hi-res PCM or DSD256). On the server there is a huge difference (as I can see from the temperatures) but for the endpoint, which just streams and does not process, there seems to be no difference.

 

These figures will save me some money 🙂

[bodiebill]

Heads up...

Apparently, ATX12VO ("12 volt only") is going to replace the current ATX standard with 24 pins:

 

https://www.guru3d.com/news-story/atx12vo-future-power-supplies-will-not-have-24-pins-atx-connector-anymorebut-10-pins.html

 

Could this be bad news for audio as the motherboard will have to perform more (conversion) tasks to power components, potentially adding noise? 

[bodiebill]

8 hours ago, Chopin75 said:

If you use NUC with intel, it only works for  non-ECC Apacer RAM.

 

Actually, some i3 processors do work with (and utilize) ECC RAM.

For instance, I am using Apacer wide-temperature ECC RAM with i3-6100T (35W TDP) and Supermicro X11SAE-M on a Spartan endpoint with just AL and NAA.

[bodiebill]

5 minutes ago, Chopin75 said:

Oh yes, I heard about that but apparently they run in non-ECC mode 

 

no, they run genuine ECC:

https://ark.intel.com/content/www/us/en/ark/products/90734/intel-core-i3-6100t-processor-3m-cache-3-20-ghz.html

[bodiebill]

handy list for ECC-compatible Intel processors:

https://ark.intel.com/content/www/us/en/ark/search/featurefilter.html?productType=873&0_ECCMemory=True

[bodiebill]

Following some posts on an ISO Regen topic here:

I just made this adapter that allows me to split usb power and signal. The thing on the left is a Chinese 5V/2A LT3045 PSU powered by batteries.

 

I made this in preparation for the Denafrips Terminator that I expect to receive in a few days. However, I already tried it on my Lampizator 'Lite/Big 7'. As the Lampi does not draw any current I just removed the little PSU, to bypass the power, and already I heard a slight step up in SQ.

 

[bodiebill]

On 3/1/2020 at 2:58 PM, bodiebill said:

Following some posts on an ISO Regen topic here:

I just made this adapter that allows me to split usb power and signal. The thing on the left is a Chinese 5V/2A LT3045 PSU powered by batteries.

 

I made this in preparation for the Denafrips Terminator that I expect to receive in a few days. However, I already tried it on my Lampizator 'Lite/Big 7'. As the Lampi does not draw any current I just removed the little PSU, to bypass the power, and already I heard a slight step up in SQ.

 

 

The Terminator arrived last week and I am now listening with:

    audio endpoint => Lush => iFi micro iUSB3.0 => Lush^2 (data) => Terminator

where the 5VBUS power for the Terminator USB is supplied separately by a battery powered LT3045 LPSU. See pic.

I need more time before I can comment on SQ.

[bodiebill]

29 minutes ago, StreamFidelity said:

The 5V is only used as a trigger by the Terminator. The real USB power comes from the linear power supply built into the terminator. I think it`s more important to supply the USB controller from endpoint with clean power.

 

Ah, so the the Terminator 5V trigger would not require a top notch LPSU?

Actually, my ISO Regen gets very clean power via 9V Studer900 LPSU => ldovr LT3045 7.1V.

[bodiebill]

15 hours ago, austinpop said:

I have had the same off and on experience with fiber. I currently happen to like it between the sNH-10G and eR, but only after I switched to the Startech - and now Planet Tech - transceivers. Ultimately, you just have to experiment.

 

For some time my setup has been:

modem/router => MikroTik SFP switch => fiber => Intel fiber NIC => server PC with HQPe

same MikroTik => fiber => Intel fiber NIC => audio PC with NAA

 

The last few weeks, after changing to a Denafrips Terminator DAC, many things improved, but I was not totally happy with the sound for classical music. High strings were a little thin and strident.

Inspired by this thread I changed the setup to:

modem/router => MikroTik SFP switch => fiber => Intel fiber NIC => server PC with HQPe

modem/router => copper ethernet => audio PC with NAA

and I like that better. Strings are fuller. The bloom is back.

I do not think it is the fiber as such but maybe the Intel fiber NIC in the audio PC is adding noise/jitter?

 

I could go all-copper, but maybe keeping the fiber upstream to isolate the audio system from the rest of the house is better? Like in:

modem/router => MikroTik SFP switch => fiber => Intel fiber NIC => server PC with HQPe

MikroTik => fiber => 2nd MikroTik => copper ethernet => audio PC with NAA

 

Any ideas?

 

(PS I just ordered a Cisco 2960G-24TC-L to experiment with.)

 

[bodiebill]

32 minutes ago, austinpop said:

I don't have experience with those switches and NIC, so in a general sense, the only advice I can offer is to continue experimenting.

 

In my experience, I find that the fiber vs. copper difference is a subtler, second order effect. This is just a suggestion for you to consider - your efforts might yield more fruit by first cleaning up the PSUs in the chain, starting with your audio PC and working backwards.

 

Are you willing to consider an etherREGEN? If so, I'd suggest it instead of the 2nd MikroTik, exactly as you've shown: fiber in, copper out to audio PC.

 

Finally, transceivers do make a big difference, and I've benefitted from the findings of others here who've recommended the Planet Tech MGB-TLX transceivers. It specifically adds density and weight to SQ, so may be useful for your thinness/stridency observations.

 

Thanks! Interesting that the transceivers make a difference also, never thought of that. So many variables to experiment with!

 

I have to control myself and for now will experiment with what I have or what is on its way to me (i.e. the Cisco switch I mentioned and a JCAT net card Femto).

 

I am not sure which functionalities the Femto card and the ER have in common, and what they do different. Is it beneficial to have both?

[bodiebill]

1 hour ago, John769 said:

FWIW: My chain is control pc ->femto net card ->Cisco 2962CG (incl 2x SFP port) optical -> sonore OM ->audio PC. In other words, part copper and part fibre. Will be interested to read how you get on.

 

Unexpected setup. I was planning to put the Femto downstream, in the audio endpoint. Any reason you chose to put in in the control PC?

[bodiebill]

6 minutes ago, John769 said:

I only have one m2 (or PCIE) slot in my NUC i7 endpoint and have opted to fill that with an optane card (for OS). Would you say the femto card is wasted in the control pc? S/w is windows.

 

I would not say wasted: some (f.i. the vendors) recommend to put the Femto card in both the control and the audio PC. I would predict it to have more impact downstream than upstream, but no certainty until tried...

[bodiebill]

3 hours ago, austinpop said:

your efforts might yield more fruit by first cleaning up the PSUs in the chain, starting with your audio PC and working backwards.

 

The audio endpoint PC is powered by an SR4, the ISO regen going into the DAC by a Studer900 via ldovr LT3045, and the MikroTik's by decent LPSU's. For my (optically isolated) control PC I went back from a HDPlex 200W LPSU (got incredibly hot!) to a nonlinear PSU, a Seasonic Titanium 400W fanless. The latter sounds more at ease.

[bodiebill]

51 minutes ago, John769 said:

Interesting.. I suppose I could try reversing things, so the OM was at after the control and the net card in the audio, and see how it sounded.  Plenty of time right now;)

 

Yes, I would be interested to know your verdict! 🙂

[bodiebill]

1 hour ago, dgilz said:

I've been using a very similar setup to your last model :

Mac Mini => Mikrotik RB260GSP => Mikrotik fiber => Mikrotik RB260GSP => Cat6A copper => SOtM SMS-200 => SOtM tX-USBHub => DAC

Each of these devices are plugged to LPS.

The Cat6A copper is a deliberate choice : the one sounding best.

On the 2nd Mikrotik switch the Sotm is the only device plugged in ethernet.

I wouldn't comment on the sound because the DAC and amp and speakers are doing a lot after this chain of devices.

All I can say is this is the best SQ I've managed to get, and I like it very very very much : it is natural, full, and brings a lot of joy in my home.

 

My 2cts advice : try, try and try again, change your cables, change the order of your installation, etc. not only it is fun to try new ideas, it provides a better understanding or control of your audio installation.  

 

Interesting, and thanks for sharing. I am also more happy now with the upstream fiber / downstream CAT6 setup. Indeed a more natural and fuller sound than with fiber alone. Although I tend to think that this is not due to the media itself (glass vs copper), but to the noise that the Intel fiber-NIC creates in the endpoint.

 

So already a clear improvement, and that is still without the Cisco 2960G-24TC-L switch and the JCAT NET card FEMTO that are both on their way to me...

 

And I agree with the try&try&try strategy! It is the only way. 🙂

[bodiebill]

2 hours ago, Sebastian T. said:

@bodiebill Did your SR4 arrive? If so, how does it compare to the Sbooster LPS?

 

Yes I have used the SR4 now for over a month to replace the SBooster to power my low-footprint audio endpoint PC with 19V. It seemed a step up in SQ and have kept it in since.

 

However, I still want to experiment to hear whether it would better be applied to power my ISO Regen (9V into a LT3945 unit that outputs 7V), or a JCAT net card FEMTO (9V into two LT3045 units to output 5V).

 

 

[bodiebill]

I did not want to present is as apodictic as I have been fooled before (i.e. hearing what I wanted to hear), but I did notice a quieter and more at-ease presentation.

Maybe the delta was not as big as I expected because of the ISO Regen and Denafrips Terminator DAC. The latter is less sensitive to noise as it has its own reclocker.

I am sure I would never swap it back for the SBooster, except of course if it has a greater impact elsewhere in the chain... Which I will try out one of these days...

[bodiebill]

7 hours ago, Nenon said:

That's part of the challenge. There is no good quality power supply on the market that I know of that can adequately power up this beast. There is no passive cooling chassis that is compatible with the FCLGA3647 socket, and you have two of them. There is no passive cooling chassis where this motherboard would fit in without drilling. It is very difficult to find good quality industrial RDIMM RAM, so that's a custom order. I am doing custom CPU heatsinks and custom cooling pipes, custom LPS, etc. etc. The list goes on and on. 

 

Somewhat confused. How do you and austinpop get such good results in what I understand to be a single-PC setup without a good powerful LPS?

 

My current (satisfying) setup is dual PC with

 

- HQPe server: ASRock Z390M-ITX/AC, Intel Core i7-7900K (95W TDP), 2x Apacer DDR4 RAM 2666MHz 8GB, JCAT net card FEMTO, Seasonic 400W fanless platinum nonlinar PSU

and

- NAA endpoint: Supermicro X11SAE-M, Intel Core i3-6100T (35W TDP), 2x Apacer 4GB ECC DDR4 2666MHz (wide temperature), HDPlex 800W DC-ATX converter, Paul Hynes SR4 19V/2A LPSU

 

When I go single PC by connecting the server directly to the ISO Regen and Denafrips Terminator DAC, the sound becomes harsh and digital again.

 

Maybe this could only be done succesfully with the likes of SR7?

 

 

[bodiebill]

2 hours ago, Nenon said:

Without a good powerful LPS? I think you got this part wrong... I am using an excellent multi-rail LPS, which is a combination of Sean Jacobs DC3, DC4, and his latest high current design. It's as good as it gets and costs a small fortune. Haven't had a chance to compare to a custom DR SR7 but will do one day. The power supply is the most important single component in a digital source. 

 

However, I think there is a fundamental difference in the way you tweak an upsampling system than what austinpop and I do with Euphony/Stylus. Please note the screenshot I have posted earlier - 2 CPUs / 20 cores / 40 threads. Two of the 40 threads are utilizing 2% CPU, two are utilizing 1% CPU, and 36 are at 0%. That's what makes Euphony sounds so good in my case.  Assigning processes to specific cores gives you a high level of isolation. Combine this with the extremely underutilized resources that are literally waiting for what to do and a world class LPS, and all the caching, playing from RAM, etc. in Euphony/Stylus, and the result is incredible. 

 

Add HQPlayer to the picture and that changes everything. HQPlayer needs a lot more processing power and keeps your CPUs busy. That generates a lot of noise that would eventually make its way to your DAC. Using a two-box solution in this case helps isolating some of that CPU noise. But that comes with its own complexities and downsides. What is important is the net result at the end. 

 

And for the net result you have to look at the entire chain when you design a digital source - from the file you are playing all the way to the analog output. That is a complex chain with even more complex interactions between the various components that includes hardware, software, power supplies, cables, the interface with the DAC, etc. Everything needs to be taken under account.

 

I typically start with the DAC and go upstream from there. Some DACs benefit a lot from upsampling. For those DACs you have to decide if upsampling with all the challenges you need to address is worth. My DAC does not benefit from upsampling, which makes this decision much easier. I owed a Denafrips Terminator DAC in 2018 but my source was not at that level at the time, so it's hard to comment on this particular DAC. I was happy with this DAC the first week only and then I realised it's not to my taste. But obviously there are a lot of considerations to make when you are deciding if you want to upsample in software or not. That's usually my first and biggest decision. Then you need to take other factors into account. Are you streaming online music? Or are you playing local files? If streaming online, your network needs special attention - cable modems, routers, VLANs, switches, power supplies for all your network devices, power cables, vibration isolation, ethernet cables, etc. This is all part of your digital source. If you are playing local files, the size of your music collection plus room for growth matters. You have to determine what type of storage you would be using. You need to decide if you are okay with a more simple player like Stylus or you need the sophisticated interface of Roon.  

 

And here is an example of interaction. If:

a. You are streaming from the Internet and get your network sorted out; and

b. You are okay with the Stylus player, that's running in ramroot and caching all the music before playing; then

You don't need to install internal storage for your local music. You don't need USB drives, SSDs, or HDDs that would generate additional noise in your computer. You can install a NAS on your network, and that would most likely give you the best results. Obviously you need to boot the OS to RAM from somewhere but that is another topic. 

But if you change just one of these variables, things can go downhill. For example, let's say you do all that and then you decide to switch from Stylus to Roon (a variable changed). In that case, playing local music from a NAS may not be the best solution for you. Roon does not cache your entire track/album to RAM before it starts playing. It would use a lot more network and other I/O activity, which would generate more noise that may be audible. I mean network activity by itself generates audible noise. Also, Roon uses more hardware resources. So, in that case, if you switch to Roon, a local PCIe NVME storage might be a better solution. 

 

This is just a very simple illustration of interaction between software and hardware. Everything matters and everything needs to be considered. I can go into a lot more complex interaction examples but it would take me hours to write it down.

 

@bodiebill I can't tell exactly why one box solution with Euphony did not work for you. But looking at your two boxes - I don't see any of them as suitable for one-box Euphony solution. Your HQPe server does not have good and clean power. Your NAA endpoint is not very powerful. It seems like you are taking the USB output from your motherboard, which is not something I've had good results with (even if you use isoRegen). Something like the JCAT USB XE card powered by a really really good LPS would help a lot. But you also need a powerful server with very clean power. And different motherboards / different CPUs sound different, so you have to make the right component choices too.

I don't know the exact details of what you have done, but it almost seems like you have designed a good HQPe-based two box solution and just tried to put Euphony one of the boxes and eliminated the other. It does not work that way I am afraid. You would have much better results considering all the variables and designing your digital source to tailor your specific requirements. That's what DIY is all about. I am not surprised your two-box solution sounds better. 

 

Thanks Nenon for your extensive explanation. Food for thought!

 

Sorry I missed your PS solution. I reacted to your "There is no good quality power supply on the market that I know of that can adequately power up this beast" and missed the context.

 

Indeed, in my case the 2-box solution sounds considerably better, and that may be because I gave up on powering the HQPe server with a LPS. I did so before, with a HDPlex 200W, but that ran way too hot using the EC modulators. I then chose for the Seasonic for the isolated server, which even seemed to sound better, though that could be my phantasy as the heat made me uncomfortable.

 

Actually, I am not using Euphony at all. Just HQPe / Audiolinux on the server and NAA / Gentooplayer or Audiolinux on the endpoint. But I guess the principles you describe would apply to AL and GP also.

 

I recognize what you say about the amount of work done by the single box, and how it influences SQ. Today I tried the single-box setup with EC modulators vs no upsampling by HQPe. The latter (using the Terminator's upsampling) sounded clearly better, whereas in the 2-box setup the reverse is the case.

 

I experimented with the 2 setups inspired by this thread and your reports. However I am really happy with the 2-box setup and the Terminator, and for now reluctant to spend a fortune on a powerful LPSU for the server to find out whether it could be improved further.

 

Regarding stock usb: an Elfidelity usb card is on its way to be tried soon on the endpoint. Cheap enough, and if I notice an improvement I could consider upgrading to a JCAT usb card in the future. Currently I am playing with NAA => Lush => iFi micro iusb3.0 => Lush^2 => ISO Regen => Terminator. I did notice though that the Terminator benefits less from such usb cleaners than my previous DAC (Lampizator Lite 7).

 

So many variables!