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

[elan120]

12 hours ago, tims said:

Did you make this silver cable yourself (by twisting single silver conductors into a star-quad configuration) or buy a silver star quad cable off the shelf?

The base of this cable is a double shielded Neotech 16AWG solid core silver cable purchased from Audio Sensibility in a twisted pair configuration.  I first setup the cable in JSSG360 configuration with the double shield, and then added the third layer to make it JSSG360^3.

[elan120]

1 hour ago, Bricki said:

Did you purchase the cable as bulk or pre terminated? I assume you altered the existing sheilds to JSSG360 configuration before adding your 3rd layer?

Cable was purchased in bulk without termination, and the existing shields were used to setup JSSG360 configurations.

[elan120]

On 9/18/2019 at 7:02 AM, auricgoldfinger said:

I have been using industrial RAM in my 7i7DNHE endpoint and 7i7DNBE server for the past several months.  Specifically, this memory:

 

https://www.soselectronic.com/products/apacer/75-b93gj-g010b-307859

 

When I was considering this tweak, my first thought was "That seems like a lot of money to spend on memory for a NUC!"  My second thought was "How am I ever going to conduct any A/B testing?" 

 

I decided to start with the endpoint.  Fortunately, the improvement was evident from the moment I re-booted, negating the need for A/B testing.  The best way I can describe the change is to say that the listening experience is more immersive.  The music is more immediate and present. 

 

I subsequently purchased the same RAM for the server.  While the improvement was not as dramatic as with the endpoint, there was definitely an additive effect.

 

I was already very far down the transparency rabbit hole when I undertook this experiment.  Spending $150 (or $300) may not make sense for everyone.  For example, I would allocate budget dollars to a 19v Farad3 or SR4-19 with high quality silver leads before upgrading the RAM.  Getting the power supply right should take precedence.

 

 

I concur the findings.  Installed the RAMs two days ago, and like these Apacer memories quite a bit.

 

Here is a picture with Apacer DDR4 SDRAM in action:

[elan120]

1 hour ago, RickyV said:

 

How would you or @auricgoldfinger rank the apacer tweak, on core isolation level, on ramroot level or clock upgrade level? 

Have you tried one simm, I think I like it. 

The difference with Apacer memory change is quite apparent.  Based my subjective listening, the level of improvement is on par or a bit better compared to Ramroot.  Music sounds more three-dimensional and rounded, the “you are there” feeling is further enhanced.  Overall, I feel this is another worthwhile upgrade that should be added to all the other changes/tweaks mentioned in this long thread.

 

Once I have more burn-in time on these RAMs, I do plan on trying single SIMM.

[elan120]

18 hours ago, Gavin1977 said:

What’s the capacitor near the back plate of the NUC MB?

That is the Evox cap.

 

 

16 hours ago, Dev said:

Did you buy the industrial type ? Any links to the place you bought the  memory from ?

Yes, these are the Industrial RAMs.  Aside from SOS, they can also be purchased from Mouser with a higher price.

[elan120]

1 hour ago, Bricki said:

just confirming that you had all 3 layers of shield connected at both ends and each shield separated by a layer of insulation... with nothing connected to ground?

Yes, confirmed.  Please see pictures below as reference.  Looking forward to your impression when you are done.

 

First two layer shields connected together without connected to ground:

 

I left about 6mm of section exposed (Red Circles) without heat shrink tubing to connect to the third layer shield:

 

Third layer added:

 

 

[elan120]

33 minutes ago, Bricki said:

My braided tinned copper sleeve arrived earlier than expected and I got a chance to put a 3rd layer on my curious USB cable.

For the first 2 layers I used the 9.5mm sleeve... That was probably over a year ago...for this 3rd layer I used 12.7mm. I tried to replicate the TANF configuration that @TheAttorney came up with in the lush^2 thread...as far as I can tell this basically involves the 3rd layer being connected to the underlying jssg360 layers at only one end, that end being the output end for the direction of the music.

 

This 3rd layer has given me even blacker backgrounds with more clarity and detail and timbre... There is more depth and individual images are less convoluted and more separated. Everything once again sounds richer and more natural... vocals sound even more dimensional and have more subtle gradations. Small background sounds in recordings are more easily discernable and every individual image is more focused and more accurately placed in the sound stage in terms of depth and width.

 

Next step is to add a 3rd layer on my Supra cat8 Ethernet cable between my NUC endpoint and server. I will use this same TANF configuration and report back again.

Very good to know JSSG360^3 is working out on your USB cable as well.

 

I also changed out my Gotham output cable to the same 16AWG solid silver wire in JSSG360^3 configuration two days ago on my MPAudio power supply that is powering my tX-USBUltra, and it is showing the same improvement as my other JSSG360^3 silver DC cables, so I am quite happy with the result.

 

Below is a picture showing the JSSG360^3 silver output DC cable (on the left).

[elan120]

12 minutes ago, Johnseye said:

 

Elan did you modify the capacitor in that PSU yourself or was it stock? 

John, if you were asking the capacitors on the bottom of the picture, then no, they are stock, but if you refer to the capacitor at the output side of DC cable, then sort of, I added this capacitor.

[elan120]

On 10/4/2019 at 7:15 PM, Bricki said:

I have now completed a 3rd layer on my Supra cat8 Ethernet cable used on the bridged connection from my server to NUC endpoint.

I did the same TANF configuration and connected just one end to the underlying jssg 360 layers.

The end with the longer section of heat shrink denotes the end that connects to the jssg 360 shielding.

 

After listening for a couple of hours this seems to have even more impact than the USB cable... maybe it's the fact that this cable has to pass over my furutech TCS31 ac cable that goes to my sps500 or maybe it's the fact that it's 1meter long as opposed to the USB cable that is only 280mm... Whatever the reason I think it has more impact in my system.

 

The impact is very similar to the USB cable. The background is blacker again and the music even richer... The depth has increased further still as well as focus. Vocals again seem to be more dimensional and everything pops more.

 

The music has a lot more emotion and feeling to it. Small nuances are more evident and bring more character... just listened to son of neck bone by the beastie boys and the impact of the drums and the timbre of the cow bell is just making it so real 😁👍... I wonder what I'll sheild next 🙄

 

On 10/5/2019 at 6:14 PM, ray-dude said:

That’s awesome, thank you for the report! Now to figure out an easy way to add a forth layer to the Lush^2

This JSSG360x2 is quite interesting and I decided to give it a try with the materials just arrived to one of my Lush^2 cable.  So far, with about an hour of playing time, it is clear that I have my new favorite.  Now I am going to modify my other Lush^2 cable to this configuration (A: Y-W,  B: R-B).

 

Here is a picture of my modified Lush^2:

[elan120]

3 hours ago, Bricki said:

Awesome! My next step is to try this jssg360x2 on my ethernet cable 😁👍

Thank you for the experiment result.  I completed the second Lush^2 cable modification earlier, been listening the music for the past two hours and the result is even more evidence.  I also plan to experiment more configurations especially JSSG360x2link and JSSG360x2bullseye mentioned by @ray-dude, but as of now, this is the best configuration I have experienced.

[elan120]

17 minutes ago, Bricki said:

These are the chips on the Apacer RAM... I think they are Samsung but I don't know for sure

Yes, you are correct, they are Samsung memory chips - SEC = Samsung Electronics Corporation.

[elan120]

1 hour ago, RickyV said:

Staying with the same chip brand, is the inside of the chip different, is pcb different so what makes it wide temp range?

Comparing with previous Apacer RAM, although they both are using Samsung memory chips (see below), the latest release sound better.

 

Apacer 75.B93GJ.G010B:

 

Apacer D21.23180S.001:

[elan120]

2 hours ago, RickyV said:

What is the technical  difference between your two types? And sq wise?

Apacer 75.B93GJ.G010B is 2400 DDR4 RAM, where Apacer D21.23180S.001 is 2666 DDR4 RAM.  Using D21.23180S.001 show a bit more detail, smoother, richer, and more musical.

[elan120]

7 minutes ago, RickyV said:

 

ah I just found it post 15089. Thanks

Here is the chip number difference:

[elan120]

5 minutes ago, Miska said:

 

Has someone tested a lower latency memory? At least HyperX (Kingston) Impact DDR4 offers faster CAS latency 15 for DDR4-2666 speed.

 

I originally use Hyper X Impact DDR4 2400 with CL14, and Apacer DDR4 2400 clearly surpass it in SQ, and now Apacer DDR4 2666 is another step up.

[elan120]

11 hours ago, Nenon said:

I just tried the SR orange fuses on my DAC and computer power supply, and I quite liked them. Thought I would share. 

I wonder if their UEF Blue Power Cable having similar effect, since the DAC I have doesn't have fuse so no way to try out SR fuses.

[elan120]

7 minutes ago, soares said:

Any idea on the price and when it will be available? Thanks

Check this post for more details.

[elan120]

13 minutes ago, Jeremy Anderson said:

I'm scratching my head a little about the difference between the -0.5V and the -2.0V. That's a big swing.

These LT3045 regulators have quite low of drop out voltage (the minimum input-to-output differential voltage needed to maintain regulation at a specified output current), however, both load current and temperature will have an effect on actual drop out voltage, where higher the load current and/or higher the temperature will drive the drop out voltage to be higher.  When single regulator is used, the below two graph can be used to determine the drop out voltage, where typical drop out voltage at full load (500mA) is 260mV.

 

 

One thing to keep in mind that too high of drop out voltage will create higher heat, where good heat dissipation will be important to avoid potential early regulator failure.

 

 

[elan120]

2 hours ago, Jeremy Anderson said:

Thanks for the clarification, @elan120!  Looks like the graphs you found stop short of the actual amperage requirements (although one could extrapolate). Assuming each device is consuming ~1A, and there are three of these in parallel on each board, then it stands to reason that we'd be looking at ~330-350mA per IC, or 250mV typical drop-out voltage at 125C which is crazy hot. I'd much rather see temps peaking at 40-50C on my "warm" electronics (but 25C is way better if ambient temperatures and passive dissipation will allow for it). That's closer to 220mV dropout... which maps to the blue line on the left graph and the green line on the right graph.

 

Am I reading that right?

Each LT3045 is capable of outputting 500mA of current.  DXP-1A5DSC from www.ldovr.com use three in parallel for their 1.5A output.

 

As for power dissipation and temperature, the power dissipation for LT3045 is calculated using:  IOUT(MAX) • (VIN(MAX) – VOUT) + IGND • VIN(MAX)

 

where:

• IOUT(MAX) = 500mA
• VIN(MAX) = Maximum input voltage, typically it will be +/-5% of the nominal input voltage.  i.e. 5Vin = 5.25Vin(max)
• IGND = Ground pin current - refer to graph below:
• 

 

So using an example of 5V input and 3.3V output, the power dissipation will be: PDISS = 0.5A • (5.25V – 3.3V) + 12.5mA • 5.25V = 1W

 

For a MOSP package LT3045, the thermal resistance is in the range of 33°C/W to 35°C/W based on the copper area (below), so to calculate the junction temperature rise above ambient temperature, using thermal resistance at 34°C/W, it will be approximately 1W • 34°C/W = 34°C.

 

The junction temperature will be ambient temperature plus the junction temperature, assuming ambient temperature is 30°C:

TJ = 30°C + 34°C = 64°C

 

 

As you can see from the above sample calculation, higher drop out voltage will result in higher power dissipation, which will drive the junction temperature higher.  Since LT3045 having a maximum junction temperature between 125°C and 150°C depending on the device grade, to keep the regulator long in service life, it is essential not to use too high of drop out voltage.

 

Hope this helps.

 

 

[elan120]

11 minutes ago, ray-dude said:

 

I'm still early in the process of really understanding the various filters in HQP, but if you have a Chord DAC and don't have a Hugo mScaler or Blu2 and want a sneak peek at what it would sound like, I suggest the following: PCM only, auto family, sinc-M filter, no dither, 768k sampling limit as a pretty damn good starting point. poly-sinc-long-lp was also quite good (and I prefered it on initial listen), but I preferred sinc-M after extended listening.

 

In general, if you want to mimic the HMS sound, stick with linear phase filters (timing accuracy), disable volume adjustments in HQP (let the DAC handle that), do as much in PCM land as you can (Rob's DACs play nicer with PCM than DSD, also a LOT less CPU load to upsample), target 768k as max sample rate (assuming you have a Chord DAC that can accept that rate on USB...basically everything newer than the TT), and use auto-family to restrict to powers of 2 upsampling.

 

I find HQP excellent, but not quite where HMS is.  It is an open question whether more taps in the filter (any chance for a sinc-2M PCM filter @Miska? I'd be delighted to beta test!), more precision in the filter coefficients (no idea what precision HQP uses...Rob has claimed 16 bit precision in his sinc coefficients), or some magic in the noise shaping (this is the most opaque part of the pipeline for me, with the most secret sauce) is the remaining gap.  

 

For now, I would characterize both of them as excellent, with the edge to HMS.  With updated filters (or better understanding of how to optimize the filters in my chain), things are danger close.  Jussi is to be congratulated for a remarkable product, and an extraordinarily generous full feature free 30 minute trial mode.

 

 

Very nice description of HQP @ray-dude, have you had a chance to test the EC modulators with DSD256?  In my system, I prefer them slightly more, but certainly your suggested PCM only, auto family, sinc-M filter, no dither, 768k sampling is very likable as well.  On a side note, when upsampled to PCM1536, SQ elevate to another notch, however, I don't have the luxury to compare with Chord DAC as you do.

[elan120]

2 minutes ago, bodiebill said:

I installed the HDPlex 800W DC-ATX converter that arrive today in my low-footprint audio PC. It only uses the first DC-in for 24 pin ATX and 8 pin EPS. (SATA is de-activated in this PC.)

The sound difference is immediate. A much broader spectrum with more dynamics. Alas also more strident (irritants), so I hope that breaking it in for a week or so will make a difference. 

Good to know it is making such a positive impact.  What do you use to feed the DC-ATX converter?

[elan120]

23 minutes ago, bodiebill said:

 

Currently SBooster BOTW ECO 19V, but I ordered an SR4. (The 800W replaced a HDPlex 250W DC-ATX.)

Sounds like your system is "relatively" low power as SR4 has a maximum output current at 2A.  I am also considering this same converter, but my newly built server will need a bit more current to run upsampling from HQPlayer, so I am still on the hunt for a suitable LPS.

[elan120]

I also have very positive experience with 16AWG solid silver in JSSG360^3 configuration for several of the power supplies to various devices.  After reading all these DC cables, I am inspired to make some new cables for my server using Neotech 18AWG 7N wires in JSSG360 configuration instead of silver.  Once I have the 24ATX cable made, I will swap both out and see how much difference they make.

 

Here is a picture of the finished 8pin EPS cable I completed this afternoon:

[elan120]

1 hour ago, ted_b said:

Sorry if I didn't explain my point.  I realize that an endpoint can be a renderer  But it can also be a rather simplified FIFO buffer, as in Jussi's NAA architecture.  I am asking whether you (or Romaz, in this example) believe these different functions would have different cpu, power, noise profiles....or is low-power-low-noise no longer the perfect endpoint box in ANY circumstance?

 

1 hour ago, cat6man said:

 

great question.

running hqplayer and having an NAA architecture at the endpoint, does increased processing power still sound better?

 

I recently done some comparisons between two NUC's with results might be helpful in this question.

 

I ran an Intel NUC7PJYH in an AKASA enclosure as endpoint when I first switched from single PC to a two PC setup late summer last year, it was using AL as OS and NAA streaming ROON/HQPd from my server.  It was working nicely for a while until I changed endpoint to an Intel NUC7i7DNHE couple month later, also in an AKASA fanless case running AL/NAA.  I stayed with this setup since that time, but continue to tweak this NUC such as clocks, internal DC cables, RAMs, external power supplies...etc., it continue to improve the SQ and recently, I built a new server running AL/HQPe that has a sizable step up in SQ, I got curious how much impact endpoint can still contribute with this new server, so I put NUC7PJYH back in using the same AL/NAA setup as well as the same power supply and RAM, the result still clearly show NUC7i7DNHE is a step up in SQ.

 

Although there are other things I did on NUC7i7DNHE that can't be transferred to NUC7PJYH, but one obvious difference is the CPU power.  There are clearly more things would be nice to find out such as will even higher power CPU be helpful in raising the SQ?  What is the best balance?  I am not sure the answers to these, but in my system, there is no one "perfect" endpoint setup.

[elan120]

3 hours ago, Miska said:

 

Have you tried going with something low power? Like the UpBoard or RasPi4 running from 5 V LPS?

 

Preferably with the bootable NAA image so that you get same OS consistently, even on different hardware.

 

I ran RPi3 with LPS-1 for a while before moving to single PC setup and now back to two PC.  Never tried UpBoard or RPi4, but I might revisit and try using the RPi3 I already have at some point down the road to see what happen now with the new PC.

 

When I did the comparison earlier, it was using the same OS with same configuration, so the consistency should be present.