Audiolinux Server configurations, Software, Hardware, and Listening Impressions

[lmitche]

While it's a long way from conclusive, I am enjoying an experiment with an 8 core Ryzen 7 2700 machine as Roonserver on Audiolinux. This replaces the 4 core 6700k used previously. The prior machine used a Asus ROG 270g motherboard. This one is an Asus ROG B450-i motherboard. The optane boot drive is in the direct to cpu m.2 slot. Running without a GPU increases SQ.

 

SQ is at a new level, with more transparency and punch across the spectrum. Various AL tuning options have a big impact, especially isolated cores. The machine is idle 99.5% of the time during playback.

 

With AL 5 the 4 port Startech Asmedia card has stopped working. Because of this, the motherboards native USB3.1 and NUC ports are being used instead.

 

Stay tuned for more. It is early days and there is much to be learned.

[lmitche]

3 hours ago, AnotherSpin said:

 

Which exact isolation cores option you use?

The menu isn't used.

[lmitche]

3 hours ago, rickca said:

Didn't you have the processor graphics disabled in BIOS on the i7-6700K?

What do you mean by NUC port?  Are you using the motherboard ethernet or built-in wifi?

Sorry, NIC. WiFi card removed.

[lmitche]

7 minutes ago, luisma said:

Hey Larry, why aren't you running headless 1.4.0? I'm sure you have a valid reason running the full version that's why I'm asking. I was under the impression headless with minimal footprint would "enhance" some things even more.

 

Hey Luismaz who said I was not running headless 1.4.0?

[lmitche]

18 hours ago, Gavin1977 said:

The CPU power circuitry of the ROG Strix Z270G Gaming is a powerful 10-phase (8+2) design, with the phases distributed to the four different input voltages that the modern 1151 CPUs require (Core, I/O, Graphics, and System).

 

The Asus ROG B450-i VRM is in a 6+1 phase configuration with each phase utilizing an IR3553 40A PowIRstage for the CPU VCore, and then a single 60A IR3555 for the SoC.

 

Convetional wisdom would be that the 8+2 phase design would offer an advantage in terms of power quality & ripple.  A 6 Phase system will reduce ripple by 50%.

 

I'm interested in your findings - so have there been any other changes.  Are you comparing like for like?

 

 

 

Hi Gavin,

 

My motherboard criterion were Ryzen 8 core compatibility, an mATX or mini-ITX board, a well implemented and documented BIOS, the highest number of VRM phases possible, and lastly no cooling fan on the PCH.

 

You can refer to the excellent Tomshardware AMD motherboard spreadsheet, https://docs.google.com/spreadsheets/d/1wmsTYK9Z3-jUX5LGRoFnsZYZiW1pfiDZnKCjaXyzd1o/edit#gid=2112472504 .

 

To narrow things done from there, please use google to learn how the user and professional reviewers rated each board. Given the criterion and additional information above, please let us know which board you recommend.

 

Many thanks,

 

Larry

 

P.S. I am using the same power supply,  as well as the same memory sticks as the 6700k build.

[lmitche]

9 hours ago, luisma said:

 

Since you mentioned this

Thought you were referring to 3.5.0

wrong assumption

Luisma, sorry not to more clear. I haven't seen a Linux X-windows GUI here since AL headless was introduced.

[lmitche]

6 hours ago, AnotherSpin said:

menu, kernel, packages

It is best to update the menu, other packages and then the kernel in that order.

 

Nevertheless, there are times when this practice hasn't worked and a fresh install is required.

[lmitche]

6 hours ago, AnotherSpin said:

 

I see. If I want do it from AL menu, what should be selected from:

 

1) All, but the first

2) Half

3) Only one (random, excluding first)

4) Disable

 

What is the reason for going with isolated cores in general? Thank you.

AnotherSpin,

 

Isolated cores dedicates one or more cores to an application process, like Roonserver, preventing OS background tasks and interrupt processing from happening on the isolated cores. This means the application is "pinned" to that set of processor(s) and there is reduced context switching overhead overtime. This results in improved sound quality.

[lmitche]

6 hours ago, Gavin1977 said:

Thanks - @lmitche are you able to provide any further insight into how to identify a motherboard with the  best VRM for audio?  I have read quite a lot including this really good VRM guide.  I am running MSI Z97I Gaming, it's a 6 phase design with Super Ferrite Chokes and it sounds way better than a NUC.  My power supply is a Corsair SF450, so ripple is about 25mV, not great but something I'm going to improve on - either way it's still a step up in audio quality - transients are amazing.

 

As an experiment I've ordered a MSI Z87 Xpower second hand from eBay to see what a 32 Phase design sounds like and also a Super Flower PSU - as these have ripple arround 5mv (it's a LLC resonant PSU design, very best you can get before moving to a linear power supply).  Quite a cost effective experiment/build as these parts can be had off eBay for cheap.

 

@luisma What power supply are you using in both cases?

Gavin,

 

I am using a Hdplex 200 watt supply to power the Ryzen chip and motherboard. The Hdplex replaced a 1600 watt titanium Super Flower ATX power supply with very low ripple in the 5 to 10 mv range.  SQ took a nice jump up in quality with the Hdplex.

 

Do the same at your own risk. It is easy to create more demand for power from the Ryzen 7 then can be supplied by the Hdplex. If this happens the supply will overheat and shutdown, or create a smoke condition as has happened with others.

 

Which NUC, powered with which power supply?

[lmitche]

1 minute ago, AnotherSpin said:

 

As use AL headless NAA only on NUC endpoint, should 3) Only one (random, excluding first) be selected? Thank you.

It is easy enough to try each option and to have a listen. The difference between options is pretty apparent here, but there is no single rule as to which sounds best.

 

 

[lmitche]

1 hour ago, RDOK said:

Hi all,

 

I installed the latest AL version on a PC to be used as RoonServer. All is well but it will only play standard resolution files, no high-res. Not even 48 kHz.

No difference if I play from Qobuz or the Nas. Previously Roon Rock was running on the same PC and that played all resolutions without problems. The only change is from Rock to AL.

I tried changing CPU freq (although cpu load is <10%), realtime prio, isolated cores on/off but no change.

Any ideas would be very welcome

Is your DAC directly connected or are you using an endpoint?

 

You need to look in Roon: Settings: Audio and ensure the DAC is setup properly.

[lmitche]

30 minutes ago, latenz said:

Hi

I am trying to build my server as a onebox-solution. maybe if it’s better for soundquality I will need an endpoint too.

But first things first. What would you recommend for highest soundquality in Audiolinux Roon Hqplayer--Setup. At the moment I try with a intel i7 gamingboard to boot from usb-stick and put the roondatabase on the optanedrive. Maybe this is stupid but I thought this is what romaz and co recommend but they are now on euphony....

next step is ram-mode when everything is up and running...

 

my mainquestion- is it worth to split roondatabase in Optane and system in ram or should I put everything on the optanedrive...

 

i would be glad to have some more input to this...I will post my findings...

 

thanks 

 

Just load the os from the optane drive into ram.

[lmitche]

1 hour ago, rickca said:

So the big question is why does the new system sound better?  8 cores vs 4 cores?  AMD vs Intel?  Motherboard differences?  Please let us know if you manage to determine the key contributing factors.

Truthfully, we will never know. With the system running just above idle, power utilization is minimal and temperatures low. Double the cache probably has an impact as well. Of course turning various bios knobs still has a great impact. Lastly the lack of a gpu on the chip has to have an impact.

[lmitche]

6 minutes ago, AnotherSpin said:

 

Would you share your preferred BIOS settings for processor and power?

I am sorry but they aren't mine to share.

[lmitche]

1 hour ago, latenz said:

Yes...but the roondatabase is to big for the ram 

The roondatabase can live on the Optane simultaneously in another partition. Poke around, the instructions for doing this were posted last year by Piero and/or Austinpop.

[lmitche]

5 hours ago, Gavin1977 said:

Thanks Larry - this is a very useful chart and one I was not aware of.

 

I would highlight the following theory as the basis for my experiment, sourced from http://mail.personal-view.com/talks/discussion/21417/motherboard-vrm-power-guide/p1

 

 

The interleaving effect is important - so phase doubling is ok.

 

It was just out of interest that I tested this theory.  I had my old Gaming PC lying around, so I thought I would see how it sounded compared to my Intel NUC7i7DNHE.  The DNHE model is apparently the 'best' NUC for audio reproduction at present according to reports on this website. So I put the Euphony USB stick into my gaming PC and...

 

My gaming PC sounded signficantly better on all fronts!  

 

There was significantly more detail, it was tonally lush and had huge dynamics in the presentation.

 

This was not what I was expecting, as I had not optimised the PC for audio playback other than using Euphony.  Now, truth be told I was only using the stock power brick for the DNHE NUC, this should be expected to be relatively noisy with ripple likely in the 60mV or so range I would expect.  The switching mode power supply in my gaming PC is a Corsair SF450, ripple is measured at 25mV which is good, but not outstanding for a SMPS.  So the SF450 should probably sound a bit better than the NUC's stock power brick.  However, 25mV is nowhere close to a good linear power supply where ripple would be 1mV or less (and much, much lower if you go for a Paul Hynes Design).  So what's going on, and why do I hear such an improvement!?

 

After some research, I believe that Voltage Regulation on high-end gaming motherboards is part of the magic sauce.  It is probably the reason why some gaming PC's can sound far better than 'lower powered' computers/streamers/renderers etc...  This is potentially one reason why very expensive servers such as Antipodes CX/EX, Sound Galleries etc... sound so good, as the cheaper, smaller, lower powered servers just don't have a lot of money spent on voltage regulation on their motherboards. The Sound Galleries Server uses a gaming motherboard from Asus... which has a high phase count.

 

Another good article on Voltage Regulation is available here: https://www.maketecheasier.com/what-is-vrm/

 

The motherboard I experienced these improvements in sound quality on is a mini-ITX MSI Z97I Gaming AC which is a 6 phase digital power design - I believe it's a 'True' 6 Phase design, 'True' because it doesn't use a doubler and it also has exactly 6 high quality Super Ferrite Chokes (SFC) on the motherboard.

 

Of course space on an ITX motherboard is limited, but there are some older gaming motherboards out there that do have more phases - for example the ASUS Z87I-Deluxe has a power riser to achieve 12 phases.  

 

ATX sized gaming motherboards can have 8, 12, 16, even 24 phases of power control.  What I have read is that a VRM with 24 phases of power is best.  Gigabyte produced one of these in a motherboard called the GA-P55-UD6 - but this is very rare and only supports older processors.  Motherboard manufacturers seem to have moved away from high phase counts for their latest designs, and have settled around 12-16 phases as perhaps they've found the sweet spot commercially.  But if you can find one of these older motherboards on the market then perhaps you have a very cost effective solution for producing a very high quality music server.

 

Why is 24 phase voltage regulation on a motherboard a good thing?  See Table 3 on this datasheet - 

https://www.renesas.com/eu/en/www/doc/datasheet/isl6617.pdf

 

With 24 interleaving phases the ripple is significantly reduced from that provided by the SMPS.

 

So, my present motherboard only has 6 phases of power regulation, I have now purchased a very rare MSI Z87 XPower XL-ATX motherboard which has a staggering 32 phase voltage regulation and also a stupidly high bulk output capacitance (which I believe will further improve the dynamics of the sound).  A ueful historical list of gaming motherboard VRMs can be found here: https://web.archive.org/web/20160914210453/http://sinhardware.com/images/vrmlist.png

 

So I think there might be merit is going for some of the older motherboards due to there very high phase count (we will see!)

 

The very best ATX SMPS for ripple are made by Super Flower, specifically the 550W version of their Leadex Gold PSU's - no other manufacturer comes close at present.  Leadex Gold SMPS measure under 5mV of ripple (they use LLC resonance conversion, which also pops us on the DIY audio forums as well).  My theory is that 5mV of ripple will then further smoothed out by 32 phases of voltage regulation and I believe it will provide exceptional performance.

 

This is a VERY cost effective solution as well - the largest expenditure is a gaming motherboard with a high phase count (some of these are £300-400 new), but an older model with a second hand processor could be had for £150 or so on eBay.

 

The 550w Leadex Gold PSU is only £60-70 which is amazing for only 5mV of ripple.  Forget about Pico PSU's... I used them in prior media centre builds and they don't offer the same level of audio performance as they're still influenced by the power brick used upstream.

 

This is just a different way of achieving the common goal of better sound reproduction.  I was not expecting my intel NUC to be relegated to the sidelines so quickly - I could buy a linear power supply for it, but if I'm achieving these levels of performance then I cannot see the value in spending hundreds on a bulky linear power supply.  Afterall, Chord Electronics, Benchmark and Hypex all use SMPS, so it can work given the right design approach. 

 

I hope this helps others.

 

p.s. 
The processor I am using is a 4790k - just because it's what I already had in my gaming PC.  It would be interesting for me to switch out this CPU for a lower powered variant - however (other than the 8MB L1 cache) I'm not sure it would make much difference(?  I need to do some more reading about this though) as I think the magic sauce is high phase count voltage regulation.  Underclocking the processor from stock is an option, but I have read elsewhere that some people find that some instruments are not rendered as well.
 

Hi Gavin,

 

Your theory that more VRM phases may be contribute to better sound quality may be true, but if you are using SMPS power supplies, like the stock Intel NUC supply or any ATX SMPS supply, you are working from a poor foundation.  A good low ripple high transient linear power supply will have a much greater impact on SQ.

 

The NUC you referenced, a NUC7i7DNHE, is used here daily as a Roon endpoint. It is powered by a Paul Hynes SR4. It sounds wonderful and enables enhancements to the server to pass straight through to the rest of the chain. I keep a stock Intel SMPS on hand, and the difference is night and day. Likewise a NUC I7 running from an Uptone LPS1.2 at 12 volts makes beautiful music at a reasonable price and is easily obtainable. Just lower the NUCs P1 value to 10 amps and you should be good to go.

 

Likewise the big Super Flower low ripple ATX SMPS cannot hold a candle to the Hdplex linear power supply with it's 12 and 19 volts rails powering the CPU and Hdplex DC to DC ATX supply in the server.  The difference in noise floor is instantly noticeable.

 

The new Taiko Audio extreme machine uses a two processor Xeon motherboard with 7+1 VRM phases. It also has a killer low ripple, high transient linear power supply.

 

Your write-up is excellent, however an image comes to mind of a proud owner extolling the virtue of the shock absorbers in his new car while he has tires filled with 10 lbs of air pressure.

 

Many thanks for the VRM facts, as they have inspired an experiment to be conducted later today.

 

Stay tuned,

 

Larry

 

P.S. BTW, the NUC referenced above runs at 400 mhz. Instruments are rendered in their full glory.

[lmitche]

VRM update - There are extensive AMD bios options used to tune VRM behavior on the ASUS B450-I motherboard. It would take months to try all the settings. Nevertheless a nice SQ gain was found by disabling VRM spread spectrum and changing the VRM frequency to 500khz. This change was from "auto" settings which means that the prior values are not clear.

 

While the SQ change was a nice bump, it wasn't as large a lift as upgrading to an LPS supply from the ATX supply.

 

 

[lmitche]

22 minutes ago, Ricardo007 said:

@imitche

Can u do this with euphony ? 

Yes, there are controls for CPU isolation in Euphony. However, as Euphony is a closed system so it is a bit tough to know the flexibility of those controls.

 

AL is wide open, so anything in Archlinux is easily in hand. This allows me to build configurations beyond the scope of the Euphony (and AL) menus.

[lmitche]

1 hour ago, latenz said:

I also did a fresh install with the new headless download...cause update from inside did not work for me ... even the suggested way crashed the system...with the new image everything was smooth...

The same problem happened here, so I downloaded the most recent image and spent hours tuning it.

 

All is good! SQ is terrific!

[lmitche]

27 minutes ago, renan said:

well i checked and it take at list 15 min to load the 3-4 Go of the system in RAM. Yes I think this is quite long but never had something different with AL in Ramroot on my pc player.

My pc player is :

motherboard supermicro x10sba-o celeron intel j1900

8 Go RAM

Belkin tweaked PCIe USB card connected to the DAC

usb flash drive

12V linear power supply 

Are you using a usb 2 or usb 3 drive?

[lmitche]

3 hours ago, renan said:

Didn't think about it.

 

Just checked and:

USB stick is connected to the USB3 port but my drive is a usb2 ..... So I have to change for a usb3 drive?

Thought so. No need to change if you are happy with the 20 min load.

[lmitche]

Hi @Gavin1977,

 

Your VRM writeup was interesting.  Many thanks for the effort.

 

In interpreting the results you should consider transient response as well as ripple. My observation is that transient response trumps ripple as a predictor of sq for a power supply.  This property is part of the magic of the Paul Hynes supplies.

 

Transient response is seldom measured in reviews of atx power supplies. They seem to focus on ripple instead. 

 

Here is an old but good explanation of the problem.

 

https://www.anandtech.com/show/2404/5

 

Understanding this, it should be no surprise that it is easier to find a high transient power supply to power a low current device like a nuc than a high transient power supply for a high current motherboard and processor.

 

You are lucky to have the one of the few high current high transient power supplies in the SR7. If I owned the same I'd be thinking about moving back to a single box solution. In the meantime an hdplex 200 watt powering a high current server and low current SR4 or LPS1.2 powering a NUC is the best of both worlds.

[lmitche]

Having read recent posts in the "Novel" thread it seems we are heading in the "gear head" direction again, where the fastest most bad-ass machine is said to produce the best SQ.

 

Hmmm. . .

 

SQ here is crazy good with a recent series of tweaks that move things in the opposite direction.  As evidence here are processor loads and temps for my server (AMD Ryzen 7 2700) and endpoint (NUC I7) while playing roonserver and roonbridge respectively on the most recent version of Audiolinux.

Server:

 

CPU temperature

Tdie:        +26.2°C  (high = +70.0°C)
Tctl:         +26.2°C

 

CPU load

 

Linux 5.2.19-rt11-3-rt-bfq (archaudio)        10/21/2019      _x86_64_        (8 CPU)

 

08:37:43 PM  CPU    %usr   %nice    %sys %iowait    %irq   %soft  %steal  %guest  %gnice   %idle x
08:37:43 PM  all    0.31    0.00    0.36    0.04    0.01    0.01    0.00    0.00    0.00   99.27 x
08:37:43 PM    0    0.01    0.00    0.02    0.00    0.00    0.00    0.00    0.00    0.00   99.97 x
08:37:43 PM    1    0.01    0.00    0.03    0.05    0.01    0.00    0.00    0.00    0.00   99.90 x
08:37:43 PM    2    2.46    0.00    2.67    0.07    0.04    0.01    0.00    0.00    0.00   94.75 x
08:37:43 PM    3    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00 x
08:37:43 PM    4    0.01    0.00    0.03    0.05    0.01    0.00    0.00    0.00    0.00   99.90 x
08:37:43 PM    5    0.01    0.00    0.02    0.05    0.01    0.00    0.00    0.00    0.00   99.91 x
08:37:43 PM    6    0.01    0.00    0.03    0.03    0.00    0.00    0.00    0.00    0.00   99.92 x
08:37:43 PM    7    0.01    0.00    0.07    0.03    0.02    0.03    0.00    0.00    0.00   99.84 x

Endpoint:

 

CPU temperature

 

Package id 0:  +31.0°C  (high = +100.0°C, crit = +100.0°C)

 

Core 0:        +31.0°C  (high = +100.0°C, crit = +100.0°C) 
Core 1:        +31.0°C  (high = +100.0°C, crit = +100.0°C)
Core 2:        +29.0°C  (high = +100.0°C, crit = +100.0°C)

Core 3:        +29.0°C  (high = +100.0°C, crit = +100.0°C)

temp1:       -273.2°C                                                      x
temp2:        +27.8°C  (crit = +119.0°C)
temp3:        +29.8°C  (crit = +119.0°C) 

temp1:        +33.5°C

 

CPU load

 

Linux 5.2.19-rt11-3-rt-bfq (audiolinux)       10/21/2019      _x86_64_        (4 CPU)

 

08:51:48 PM  CPU    %usr   %nice    %sys %iowait    %irq   %soft  %steal  %guest  %gnice   %idle x
08:51:48 PM  all    0.06    0.00    1.11    0.02    0.12    0.01    0.00    0.00    0.00   98.69 x
08:51:48 PM    0    0.02    0.00    0.05    0.04    0.01    0.00    0.00    0.00    0.00   99.88 x
08:51:48 PM    1    0.02    0.00    0.04    0.01    0.01    0.00    0.00    0.00    0.00   99.92 x
08:51:48 PM    2    0.02    0.00    0.08    0.01    0.01    0.03    0.00    0.00    0.00   99.86 x
08:51:48 PM    3    0.17    0.00    4.29    0.00    0.47    0.00    0.00    0.00    0.00   95.07 x

 

The ambient temperature of the room is 22°C. 

 

SQ is holographic.

[lmitche]

1 hour ago, Dev said:

@lmitche & @Forehaven

 

what motherboards are you using for the Ryzen 7 and how are you powering it ? I have been meaning to build a high performance server for sometime. Can you folks please list the parts used ? AL fully supported on Ryzen ?

Hi Dev,

 

We are using various MicroAtx and Mini-Itx motherboards mostly Asus. It is best to shop around for 65 watt CPUs. Yes AMD is fully supported in Audiolinux, with a menu switch to change from AMD to Intel and vice versa.

[lmitche]

1 minute ago, lmitche said:

Hi Dev,

 

We are using various MicroAtx and Mini-Itx motherboards mostly Asus B450s. It is best to shop around for 65 watt CPUs. Yes AMD is fully supported in Audiolinux, with a menu switch to change from AMD to Intel and vice versa.