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

[Johnseye]

14 minutes ago, ElviaCaprice said:

 

It will work in serial, not parallel.  I originally powered my sCLK EX NUC with two LPS 1's in serial.  Replaced by SR7 which is far better.  Problem with serial is increased impedance.

 

The Y cable would be parallel right? 

 

[Johnseye]

23 minutes ago, mormogi said:

Very very good powered ssd do the same or similar effect . 3 hpuln series +perfect AC. (not sps500 or lps1.2!!!)

its too expensive I know!!!

 

I was powering my SSD with an LPS-1.  Loading in RAM means no need for an SSD.  This is totally different.

[Johnseye]

34 minutes ago, Superdad said:

 

As mentioned on the phone, this is trivial to do and I have voltage set resistors arriving to do max. current tests at both 15V and 19V.  

The JS-2 won't be able to offer 7+ amps at 19V (as it does for 12V), but I'm hoping it will make it to 4.5A easily. Could simply substitute the set resistor on one of the four settings (presently 5/7/9/12V) on one of the output jacks.  19V/4.5A is still 86W.  Plenty for this class of NUCs.

Still, I am not yet convinced of any benefit to 19V.  We have loads of happy NUC and Roon Nucleus users at 12V. 

 

I'd be interested. I think you'd have a market. Not only could 19v NUC's be powered but also mITX boards. 

[Johnseye]

51 minutes ago, RickyV said:

 

There are other gen 8 NUCs i7 with 15w tpd.

 

https://ark.intel.com/compare/130392,130393,126140,130394

 

Are these any good or is the NUC8I7BEH better?

 

Rick ?

 

3 of those are the same processor.  Lower TDP is better.  The difference between 4.2 and 4.5 GHz is not worth the extra power.  I don't know if there are any impact-full instruction sets the 4.5 has over the 4.2 but I would guess for our purposes, none.

[Johnseye]

1 minute ago, rickca said:

OK I understand.  But even then you don't need a discrete graphics card.  I think the Bean Canyon NUC8 series would be a better choice.

https://www.anandtech.com/show/13113/intel-officially-launches-bean-canyon-nucs-with-coffee-lake-u-processors

 

Why?

[Johnseye]

23 minutes ago, rickca said:

Because the Bean Canyon series don't have a discrete graphics card to drive up the price and TDP.  Hades Canyon NUC8i7HVK is 100W TDP.  Bean Canyon NUC8i7BEH is 28W TDP.  Hope this is clearer.

 

Hades Canyon is Intel's attempt at a gaming NUC. 

 

Ok, you're comparing NUC 8 only, not 7.  

[Johnseye]

48 minutes ago, auricgoldfinger said:

This post is intended primarily for the benefit of those who own a streamer made by one of the popular manufacturers and who may be wondering how the AL NUC streamer compares to those products.  

 

 

 

Recently, I sent my sMS-200ultra to a dealer for the Neo upgrade.  The dealer also replaced the two SOtM-upgraded stranded silver internal DC cables with a pair of 20 AWG solid core Neotech silver cables that received the JSSG 360 treatment.  The first photo shows the 2-layer shielding using the JSSG360 approach. The second photo shows the upgraded SOtM and custom dealer DC cables side-by-side.

 

 

 

 

 

The Neo upgrade is a subtle, yet welcome, change that eliminates the slightly bright SOtM “house sound.”  However, the real star is the new DC cables.  The sound is much more open, and the background seems completely silent.  Highlights include richer and more textured vocals and a sharper image, with the placement of instruments being much more precise.

 

 

While the sMS was away, my worst obsessive-compulsive tendencies overcame me amidst all the discussion of using an AL NUC as a Roon endpoint.  On a lark, I contacted @lmitche to ask if he would build one for me.  I expected him to politely decline, but to my surprise, he was quite enthusiastic about the idea.  I laid out my requirements, and he asked questions to familiarize himself with my setup.  With that information at hand, he suggested options, and we reached an agreement on price.  The NUC was in my hands a few days after I sent the funds.  Larry is very professional, and a great guy, to boot.

 

 

Given the elevated performance of the modified sMS, I was a little apprehensive about the NUC.  Keep in mind that the sMS has been given every conceivable upgrade, including the internal JSSG cables, internal EMI absorber, and re-clocking by a Reference 1 double regulated PSU powered OCX10 via Habst Digital BNC cables.  Despite all of these upgrades, the NUC easily beats the sMS in terms of detail and dynamics.  In fact, it is really not much of a contest.

 

For perspective, my chain in its simplest form can be described as follows:

 

 

FMC > switch > streamer (sMS/NUC) > tX-USBultra > DAVE > HE1000 v2

 

 

The OCX10 re-clocks the sMS and tX, which in turn re-clock the switch and both FMCs.  (The FMCs and switch were modified by SOtM.)  The OCX10 and tX are each powered by a Reference 1, the streamer and switch are each powered by an LPS-1.2, and the downstream FMC is powered by a JS-2 rail.  I utilize a Synergistic Research ground block and HD cables to ground various components and cables, as well as a Gutwire HD cable to ground the DAVE. 

 

 

For my listening comparison, I simply substituted the NUC for the sMS, keeping the LPS-1.2 as the power supply for both.

 

Great to hear of another validating what those of us experiencing with Audiolinux and the NUC.

Would love to hear your impressions of with and without the tX-USBultra in line.

[Johnseye]

1 hour ago, ElviaCaprice said:

I think you guys are getting too hung up on NUC's as if they have some magical qualities.  If it's just audio as an endpoint using the mobo USB, LAN, AL headless with 4GB memory, then I would go even smaller mobo.

 

Like a Pico-ITX,   http://www.jetwaycomputer.com/JBC400P591.html

 

In fact, I've got one sitting around, have to give it a go as an endpoint at the summer home.

 

5 minutes ago, Cortes said:

+1

 

Indeed, any modern laptop with SSD,  low consumption Intel CPU, and wireless disabled in bios ,is much more clean than   a crappy NUC.  I can't see how a modern laptop running any linux distro in the ram is not better than a NUC. 

 

 

 

 

 

 

Maybe 

 

 

 

Seriously though you need to read through Roy's hypothesis again.  In his opinion processor cache is a factor.  That Jetway you linked has 2M cache.  I think we're all guessing and trying to figure it out at this point, but one thing's for certain is that these NUC 7s sound great.

[Johnseye]

1 minute ago, Cortes said:

 

Cache?. My desktop is an AMD Threadripper with 32MB of cache, and running linux. Maybe is an audophile jewell and I had not realized. ;-).  This looks a serious snake oil.  

 

If you had a 32MB proc cache on a board the size of a NUC using 15w TDP maybe you could compare. 

 

Snake oil that anyone who's actually heard it agrees from a device sold by Intel not specifically for audio? It's a discovery. 

[Johnseye]

16 minutes ago, Bricki said:

I think Roys conclusion was that it had more to do with core count than cache size. When he compared the i7 nuc with one core active (the smart cache allowing all 8MB to be used by one active core) it didn't sound as good as all cores active. I think he concluded that roon bridge was multi threaded and benefited from all cores active... ?☺️

 

It has more to do with how much cache an active proc has access to.

 

 

 

 

[Johnseye]

2 minutes ago, Bricki said:

John, you're right that certainly was his original theory. Later he did this comparison above that changes things a bit ☺️

 

Good catch, I missed that.  Cores over cache.

[Johnseye]

I asked Jussi @Miska to add ramroot to his bootable HQPlayer NAA and he quickly put it in his build.

 

https://www.signalyst.eu/bins/naa/images/

naa-3552-x64.7z 

I like what I'm hearing.  I've only listened to a couple songs at first listen it's similar if not the same as what I hear with Audiolinux headless running HQPlayer NAA.

[Johnseye]

14 minutes ago, Superdad said:

this NUC makes much obvious things like HQ Player SRC filters and dither options.  And now I absolutely can not stand HQP PCM>DSD conversion.  For reference, folks should know I have a heavily modified Singxer SU-1 feeding I2S to a Holo Spring Level 3.

 

I 100% agree about the filters and conversion.

I also have a modified SU-1 but am not sold on it with the NUC yet.  I just started doing some comparison with both the tX-USBultra before it and out all together.  I think I should spend more time with it.

[Johnseye]

6 hours ago, austinpop said:

SQ Differences between Roon player and Squeezelite on Audiolinux

 

Now that I had squeezelite working at last, I was finally able to address the question of SQ differences between the 2 ecosystems - Roon and LMS. Recall that on the shipped OS, InnuOS, the SQ of the LMS/squeezelite combo is consistently better than Roon.

 

Having brought up Audiolinux on this box, running in ramroot, and having found that Roon in this configuration exceeded the SQ of both Roon and LMS/squeezelite on the original OS, the question then became: what about LMS/squeezelite on Audiolinux? Would it in fact again leapfrog Roon?

 

Short answer: Yes. But only if properly tuned.

 

More Detail

 

Let's first talk about memory. This box has 8GB of physical RAM. Despite my efforts to trim the size of the root partition, here is the output of the free command after inital boot:

 

              total        used        free      shared  buff/cache   available

Mem:        8026548     1846292     5461128         556      719128     6074440

Swap:             0           0           0

 

 

Notice that there are approx 1.8GB in use, with 5+GB free.

 

After Roon is started and running for a few minutes, used memory has grown by ~600MB, which confirms Roon is allocating memory as it goes and as needed.
 

              total        used        free      shared  buff/cache   available

Mem:        8026548     2242756     4945320         560      838472     5678540

Swap:             0           0           0

 

I first ran squeezelite without setting any value for the -b option. From the help, here's what this option does:

  -b <stream>:<output> Specify internal Stream and Output buffer sizes in Kbytes

 

Next, I tried several values for this parameter on the command line:

1. default, no setting

2. -b 1048576:1048576

3. -b 1572864:1572864

4. -b 2097152:2097152

Here is the output of free in the -b 2097152:2097152 case:

              total        used        free      shared  buff/cache   available

Mem:        8026548     6067792      993068         564      965688     1853428

Swap:             0           0           0

 

Notice that the memory used has grown by about 4GB, which confirms that the 2 numbers in the -b option are additive. Also note that free memory has dropped to under 1GB (ulp!), but what the heck - let's live dangerously.

 

Comparing squeezelite case 4 (4GB buffers) with Roon, the SQ difference is quite obvious! squeezelite sounds so much better!

1. Yet another jump in dynamics, but also
2. it feels like the noise floor dropped even more
   • bigger and deeper image
   • better instrument separation

I was stunned. But what about stepping back from the precipice (very little free memory), and reducing the buffer size? With each step from 4GB (option 4) to 3GB (option 3) to 2GB (option 2), there were drops in SQ, although they were small. The final step - going from 2GB buffer (option 2) to default buffer - was the biggest drop in SQ.

 

Here's the kicker - running with the default buffer size (i.e. no -b parameter set), the SQ of Roon and squeezelite were indistinguishable. In other words, the SQ benefit from squeezelite comes from setting (and presumably pinning) large buffers.

 

Key Takeaways

1. Those of you who are running squeezelite with LMS on Audiolinux, please give this a try and report what you hear.
2. But... beware of overcommitting memory! With squeezelite running for a while, run "free" and see what your situation is. Then looking at how much free memory you have, set a safe buffer size budget, and try the -b option. If you're running the squeezelite-R2 that Piero recommends, then this will need to be set in the /etc/conf.d/squeezelite file in the params value.
3. Even those of you who are running squeezelite on other Linux variants might want to give this a try.
4. Finally where does this leave us Roon fans? Frustrated! We need to get the Roon architect(s) to see these results and prototype a suitable form of this in their player. Of course, Roon is a more complex player, and they have to consider all the target OSes it can run on. But I think there is some gain to be had with clever coding.
5. We (Roon users) must advocate for this.

 

On a similar note I'd like to share some of my experiences now which I'll summarize another day in another post.

 

Audiolinux GUI based version, LXQT

There are system and package updates that can be run in Audiolinux; pacman and yaourt.  I have found that running these updates at the stock image size causes issues.  There's simply not enough space.  I have to run gpartd to extend the partition on my USB stick, then update.  I haven't verified yet whether the end result is under 8GB and the packages just need a temporary place to be stored until install.

 

ROON CORE on LXQT

My Roon database pushes the size of the OS beyond 8GB.  I have a lot of music.  Fortunately I have 16GB of RAM in my server.  Again, extending the partition on the USB stick before loading into RAM eliminates a lot of headache.

 

These NUCs should be able to handle 16GB.  I know mine can go up to 32GB but I believe (needs validating)  these Celeron versions are stated as 8GB.  I have read that the Celeron versions will work with 16GB without issue.

[Johnseye]

2 minutes ago, ronfint said:

I had exactly the same problem. Following  Piero's guide, I moved my RoonServer database to a second usb flash drive, and everything then worked smoothly. I am a total linux novice, so Piero's help was invaluable.

 

Right now I am running AudioLinux on a Retina MacBook Pro. I'm waiting for the dust to settle before deciding on a NUC for my Roon Server.

 

Piero informed me you can extend the partition of the USB stick to its max capacity and the image will still only load the data.  So if your stick is 128GB you can extend to that size.  When you load into RAM it will only create a partition the size of the RAM.  I did not test this.  I kept my partition size under 16GB.

 

I'm assuming you don't have more than 8GB of RAM in your Retina?  Does storing the Roon db on another USB drive slow anything in Roon?  At a minimum you need to keep a UBS stick installed in the MAC/PC/NUC which generates more noise.

[Johnseye]

3 minutes ago, rickca said:

Happy Thanksgiving.  I want to thank the incredibly generous people here who share their observations and ideas and help me understand how to get the most out of my system.

 

It's like being in a fairy tale. ? 

 

 

Happy Thanksgiving!

 

[Johnseye]

4 minutes ago, austinpop said:

 

It's not one thing or another. All these things seem additive. 

 

 

Yes they are. Well put.

[Johnseye]

1 hour ago, beerandmusic said:

i was asking davide and he is using ssd for music storage...he might as well have a bootable partition.

 

Good grief dude. You can't be that stupid. Who cares who you asked. The answer is the same. If you've read any of this thread the answers are already there. It's not my thread but I'd consider you an instigator and ban you. 

[Johnseye]

2 hours ago, seeteeyou said:

 

Intel just happened to offer Denverton with 16 cores / 16 threads @ 24W / 32W TDP

 

https://ark.intel.com/products/codename/63508/Denverton

https://www.intel.com/content/www/us/en/design/products-and-solutions/processors-and-chipsets/denverton/ns/atom-processor-c3000-series.html

 

Intel Atom C3000 “Denverton” Processor Targets Low Power Servers

https://www.cnx-software.com/2016/07/16/intel-atom-c3000-denverton-processor-targets-low-power-servers/

 

Flex ATX (with 2 PCIe slots) for over $900

 

https://www.provantage.com/supermicro-mbda2sdv16ctln5fo~7SUPM5TH.htm

https://www.wiredzone.com/supermicro-components-motherboards-embedded-processor-a2sdv-16c-tln5f-10027787

https://www.dihuni.com/shop/do-it-yourself-embedded-iot-and-server-chassis-and-boards/supermicro-a2sdv-16c-tln5f-intel-atom-c3958-soc-16core-256gb-m-2-1xpcie-4x10gbe-1xgbe-embedded-iot-gateway-security-appliance-server-motherboard/

 

Mini ITX (with 1 PCIe slot) for $750 or so

 

https://www.atacom.com/program/atacom.cgi?ADD=MBMI_SUPE_A2_H3

https://www.cdw.com/product/supermicro-xeon-16c-motherboard/4933542

https://www.interpromicro.com/commerce/catalog/product.jsp?product_id=30667&amp;czuid=1543025568775

https://coreprocomputers.com/products/supermicro-a2sdi-16c-hln4f-o-intel-atom-c3955-ddr4-sata3-usb3-0-v-4gbe-mini-itx-motherboard-cpu-combo

https://www.dihuni.com/shop/do-it-yourself-embedded-iot-and-server-chassis-and-boards/supermicro-a2sdi-16c-hln4f-intel-atom-c3955-soc-16core-256gb-12sata-4gbe-12v-embedded-iot-gateway-display-kiosk-security-appliance-server-motherboard/

 

C3955 on its own would already cost $434 so that's why we've gotta pay quite a bit more for such fanless SoC with the highest number of cores. It's "still" an Atom but its performance ain't too shabby when compared to stuff like Xeon Silver

 

https://www.servethehome.com/intel-atom-c3955-16-core-top-end-linux-benchmarks-review/

 

I have a mini ITX from April of this year and the NUC sounds better with a similar proc. Both running Audiolinux. Although the ITX is running gui and NUC is running headless. 

 

Not sure about the Atom but "suspect" it's not as good. 

 

 

[Johnseye]

8 hours ago, Dev said:

 

 

Not sure if you are aware but Intel has been following this thread for a while before they released the Celeron NUC this year that many of us here like it for audio. Get one while you can ?

 

You're joking.

[Johnseye]

5 hours ago, misterspense said:

So that means two things are different, NUC vs mini itx, and headless vs GUI. So you're saying a NUC with headless AL sounds better to you, compared to mini itx with GUI AL.?!?!

 

Yes. There's another difference. My ITX is clocked by the sCLK and it was used as a stand alone whereas the NUC was endpoint. However both were the last component before the dac. Still not quite apples to apples.

 

I'll be trying the ITX headless soon.

[Johnseye]

4 hours ago, rickca said:

John, can you please remind me what software you're running on these machines?

 

Headless AL/NUC is running RoonBridge and LXQt AL/ITX sCLK is running Roon Server?  And previously LXQt AL/ITX sCLK was running Roon with your DAC directly attached?  Or are you using HQPlayer in this mix?

 

mITX is runing Audiolinux lxqt with Roon Core and HQPlayer Desktop.  I will be trying it with Audiolinux headless, Roon Core and HQPlayer Embedded.

 

NUC is running Headless Audiolinux with HQPlayer NAA.  I have also run it with Roon Bridge and Jussi's HQPlayer NAA.  All in RAM.  Jussi's version is very close to Audiolinux in SQ.  I need to put more time in comparing the two.  Roon Bridge sounds a little thin by comparison.

 

All PCM is upsampled to 352 or 384 and DSD to 512.

[Johnseye]

4 minutes ago, davide256 said:

So whats the fallback if the NUC7CJH is being discontinued? The out of stock/lack of product  market situation doesn't bode well.

 

I use the gen 7, i7 version. There's no magic to the NUC7CJH. I don't think anyone's tested a gen 6 version. You can also get the gen 8.  By gen I mean NUC gen, not Intel proc gen. There are gen 7 NUCs with gen 8 procs.  It just depends how much you want to spend. Remember, the more cores and bigger cache the better. 

[Johnseye]

10 minutes ago, rickca said:

Do you have NUC7i7DNHE or NUC7i7BNH?

NUC7i7DNK1E

[Johnseye]

1 hour ago, lmitche said:

Will your NUC enter the bios? I had what I thought was a dead NUC, but happened to notice the Ethernet light flickering in a random fashion and realized it was working. I was using a TV as monitor. This was a problem with CEC in some bios releases that results in a blank screen. Once attached to a proper computer monitor all was good. Later I set the "play sound at boot" options on. Big mistake, with no boot. The 4.6 bios fixed that.

 

Nevertheless, I am sure Intel will honor a warranty claim and send a new device.

 

It could be your ram causing the issue. Do you have an extra stick? 

 

When you get the Akasa case see if you can power your SSD separately. That way you can unplug it.