Building the ideal(ish) Music Server

[Guest rmpfyf]

Changing the clock is overrated. Far and away the best thing for  sound quality improvements is improving the quality of power  that gets to the clock, CPU and so on.  I dont intend  wasting too much more time fiddling with software tweaks.

 

It does sound markedly different, though that's not always better. though right now I'm on TSC (what Linux selects). Was amused that I could force the PC to sound better on HPET under some build conditions; it's effectively a legacy timer. 

 

The clock doesn't really get significantly affected power. It's a crystal on most motherboards and has a decent amount of phase noise irrespective of what powers it. It's not a precise thing, and even with that in place software's still important - fundamentally, timing in a PC might have a hardware reference but it's all interrupt-driven. Linux is pretty flexible here - tick or tickless kernels, you name it. I changed the actual crystal on the motherboard for a TCXO on it's own power and soon for an OCXO. That's been a serious bump in SQ (it's also been amusing to see the PC just boot faster under a better clock source). It was also a great excuse to buy a solder vacuum

 

Wasting or otherwise is your call. Not a value exercise for some, not really accessible for others (Snakeoil is quite good as an introduction to what's possible and then some). Any modern OS is a long way from firmware on a good CD transport that often just sounds better - there's a serous amount of work to make it simpler and workable.

[agisthos]

Changing the clock is overrated. Far and away the best thing for  sound quality improvements is improving the quality of power  that gets to the clock, CPU and so on.  I dont intend  wasting too much more time fiddling with software tweaks.

 

You do both. Upgrade the clock, and feed it with its own regulated and isolated PSU.

[moonbeam]

You do both. Upgrade the clock, and feed it with its own regulated and isolated PSU.

 

Could you please provide photos/details of the conversion you did?  Did you have any issues during the installation?

[agisthos]

Could you please provide photos/details of the conversion you did?  Did you have any issues during the installation?

 

I did it to my Cambridge Audio 752, well to be more precise, Joe Rasmussen did it.

Ease of such a mod all depends on the hardware. At least with the Oppo board design, many people are doing clock mods so there is a common guide and pool of experience to work from.

But with every PC motherboard being unique it may not be so simple.

[TP1]

I did it to my Cambridge Audio 752, well to be more precise, Joe Rasmussen did it.

Ease of such a mod all depends on the hardware. At least with the Oppo board design, many people are doing clock mods so there is a common guide and pool of experience to work from.

But with every PC motherboard being unique it may not be so simple.

 

 

We have been looking at this - but based on the experience of others who have tried it, results can be hit & miss,  or failure.    PPA have a clock Mod for a specific motherboard  but i would think the most benefit would be gained from modding  the clock/power supply to the  SOTM or equivalent USB card . SOTM have a clock upgrade path and  it does seem to improve SQ - although how much of this is due to the clock or the upgraded power supply feeding it is not certain.  @@Chanh has taken to powering the clock on his SOTM card directly from a clean fast reacting power source and bypassing the on board step down regulators. The result gives an audible improvement.  

 

There are some motherboards that do sound a lot better than others,  and this is  a good way to go for real gains in SQ  if  the system has plateaued .  I have mainly be using lower powered mobos but have gone through quite a number,  and for this class of motherboard, those using PCI USB card - the ASRock D2700  is the best I've tried and  for PCIe USB cards, its the  AS Rock N3700. The N3700  has excellent dynamics and precision and black backgrounds , as does the D2700 which also creates a massive, airy  soundstage. Both use polymer capacitors throughout  and are well built.  Polymer caps have a lower ESR and faster response than aluminium electrolytic and this would  certainly help the  overall sound quality.   

Edited May 20, 2016 by Tasso

[AccuTidal]

....

Edited July 31, 2016 by Chanh

[Guest rmpfyf]

@@Tasso - PPA's mods fit a range of boards with a bunch of generic chipsets, I'm just not sure they cover Atom chipsets. Actually I'm pretty sure they don't - the NM10 Express chipset on the motherboards you mention has some pretty specific clock requirements (32.768kHz crystal, pages 56, 74 and others of http://www.intel.com/Assets/PDF/datasheet/322896.pdf). This crystal drives the Real Time Clock (RTC) on a relevant Atom board, and from that derives a number of other system timings, but bear in mind it's a kHz reference driving a few MHz sources - not ideal. There are other inputs required; PCIe and LAN may have their own oscillators (100MHz and 50MHz or some multiples thereof - usually 25MHz, e.g. http://www.intel.com/content/www/us/en/embedded/products/networking/ethernet-connection-i218-gbe-datasheet.html). 

 

32.768kHz is common for an RTC as dividing it by a binary counter (2^15) gives one second increments.

 

Intel Core chipsets are different and have a same crystal for the RTC but a completely different crystal for system needs. A range of frequencies are usually derived from this Integrated Clock Controller, which you can't change easily (it's part of the chipset), but you can change the input clock for something better (pages 51 and 102 of http://www.intel.com/content/www/us/en/chipsets/9-series-chipset-pch-datasheet.html). That clock on a H87/Z87/H97/Z97 is 25MHz, for which many good low-phase-noise oscillators exist. The clock architectures on these platforms are considerably more robust, if only because they're engineered to support more devices that ultimately need different clock references.

 

In short - Core platforms are far more dependent on the quality of their external clock references than Atom platforms. On Core boards results are quite good.

 

For those willing to try to replace their RTC crystal sources: there are a few very nice OCXO's out there with mega-low phase noise at the right frequency - see the SC-cut version of this http://www.futureelectronics.com/en/Technologies/Product.aspx?ProductID=OX4020ALZ2832768RALTRON4008404 (would need surrounding circuitry to work properly, but pretty good). I'm not sure it'd be as useful as a clock replacement on a Core board.

 

Any of the hardware mods only really shine when software issues are sorted (e.g. interrupt management, SMM calls https://en.wikipedia.org/wiki/System_Management_Mode, etc), e.g. getting the most out of something that times your bits better is dependent on not having a lot of other crap in the way  EDIT I should add that this is where Atom boards on good OSs tend to shine; the system latencies are crap but tuning audio latencies is often easier - just a different way to skin (that part of) the proverbial cat.

 

The SoTM mods are interesting, it'd be good to unpack why better clocks work on a board running async mode.

Edited May 20, 2016 by rmpfyf

[Guest rmpfyf]

@@Chanh -

 

A decent quality crystal is hard to beat and better still when it is powering by an isolated/seperated fast clean source. To achieve the best outcome, quality of the power source, implementation, and quality of that crystal must be at equal weight. No point getting a super pricy OCXO (last I checked the Pulsar were $950 each) and anticipate it will provide audiable sonic improvements when powering its with the poor quality supplies.

 

This has been mentioned at length previously (and is kind of obvious). $950ea is the extreme end of things - Pulsar is around USD$500 and there are lower-cost options than this.

 

In term of phase noise of a crystal, your comments are rather disappointing given the level of knowledge you were trying to demonstrate....! 

 

Let's try to avoid these sort of comments, Chanh.

 

For your information, phase noise of a particular crystal will improve relatively to the quality of its own PSU and the surrounding environment of that particular crystal's placement.

 

Yes mate, obvious, and as mentioned earlier - I run my OCXO's on their own power and, to be clear, in their own environment. Not least because it draws over 1A with a nice inrush when the oven's coming on, and not least as the size of the thing makes soldering it to where a 5mm crystal used to live impractical. 

 

Also... we're not all stupid, and implicitly understand that putting something designed to reduce noise in a low-noise environment right in the middle of a high-noise environment defeats the point of the exercise. 

 

Furthermore, phase noise will get worse as its natural frequency gets higher. For example - a 12.xxMHz crystal will certain to perform much much better in term of the unavoidable phase noise against a 48.xxMHz and more so better then a 98.xxMHz. Of course the fabrication's quality of a crystal and the method in which it is built will improve the so called errors or phase noise. Note - it is inevitable that phase noise will exist in any particular crystal at any frequency, in principle the phase noise are more prone at higher frequency.

 

Chanh, without trying to take you to school here - 

 

There are many factors that affect phase noise and many different characteristics of phase noise that are favourable to audio applications. Inherent increases in phase noise at higher frequencies (a) isn't really relevant on a motherboard upgrade when there's only one frequency to replace, (b) the phase noise difference between the crystal being replaced and the the OCXO in it's place are a few orders of magnitude different at frequencies of interest, and © external logic can be used to cut frequency down to a multiple of source, with relevant improvement in phase noise.

 

Lastly, I am confused with your observation "(it's also been amusing to see the PC just boot faster under a better clock source)"?!? Did you mean it would boot faster under the identical frequency crystal replacement or a higher frequency crystal? If you have not changed any other parameters, other than replaced a better quality crystal at the identical clock frequency, then it is truly amusing. I wonder how does this simulate on principle? This reminds me of a practical test we did between a Corolla vs a BMW 650i, both were consistently travelling at 80km/h over a precise distance of 30km. Somehow, it felt much faster with the BMW than the Corolla.

 

Same PCH frequency, verified in dmesg. OS tasks don't execute linearly, Chanh, and many interact and have timing dependencies. 

 

If in that 30km example the Corolla and the BMW split into some 100+ pieces each at the start, went into a constantly-changing maze and regrouped at the end - and you gave the BMW a way of understanding how the maze updated that was slightly more time-accurate, it'd get to the end of the maze and regroup sooner. There are (often) millions of CPU instructions executed in just a few seconds of boot time, and many interact with each other with timing dependencies within near-sideband phase noise tolerances of some key timing devices. The example you provide isn't relevant. 

[TP1]

@@Tasso - PPA's mods fit a range of boards with a bunch of generic chipsets, I'm just not sure they cover Atom chipsets. Actually I'm pretty sure they don't - the NM10 Express chipset on the motherboards you mention has some pretty specific clock requirements (32.768kHz crystal, pages 56, 74 and others of http://www.intel.com/Assets/PDF/datasheet/322896.pdf). This crystal drives the Real Time Clock (RTC) on a relevant Atom board, and from that derives a number of other system timings, but bear in mind it's a kHz reference driving a few MHz sources - not ideal. There are other inputs required; PCIe and LAN may have their own oscillators (100MHz and 50MHz or some multiples thereof - usually 25MHz, e.g. http://www.intel.com/content/www/us/en/embedded/products/networking/ethernet-connection-i218-gbe-datasheet.html). 

 

 

 

Wrong - I spoke to the man himself and  he has told me that there had been failures with other boards. This is the difference between theory and actual hands on experience. 

 

 

 

 

Any of the hardware mods only really shine when software issues are sorted (e.g. interrupt management, SMM calls https://en.wikipedia.org/wiki/System_Management_Mode, etc), e.g. getting the most out of something that times your bits better is dependent on not having a lot of other crap in the way  EDIT I should add that this is where Atom boards on good OSs tend to shine; the system latencies are crap but tuning audio latencies is often easier - just a different way to skin (that part of) the proverbial cat.

 

The SoTM mods are interesting, it'd be good to unpack why better clocks work on a board running async mode.

 

 

Correct. But there is more to the SQ than latencies.  Motherboards with identical chipsets from different manufacturers have shown similar results with latencies yet sound significantly different.  This is not really surprising since the electrical properties of the  server  directly affect SQ.

Edited May 21, 2016 by Tasso

[TP1]

 

@@rmpfyf

 

Yes mate, obvious, and as mentioned earlier - I run my OCXO's on their own power and, to be clear, in their own environment. Not least because it draws over 1A with a nice inrush when the oven's coming on, and not least as the size of the thing makes soldering it to where a 5mm crystal used to live impractical. 

 

 

 

 

 

  I dont want to keep repeating myself about the importance of corroboration in a development thread such as this but without it,  we can go back and forth incessantly without resolving anything.   If you want to prove a point, please do so.

[proftournesol]

<mod post> How about we have no 'point proving' in the thread, it isn't a competition. <mod post>

[AccuTidal]

...

Edited July 31, 2016 by Chanh

[Guest rmpfyf]

@@Tasso, as mentioned previously though I'll make it clear -

 

• I run an OCXO as clock reference into my PCH. This replaced a TCXO, which replaced the original crystal.
• This has provided a great benefit in SQ on my rig. That's my opinion and the opinion of those that come and listen to my rig.
• It's provided a great benefit on all my motherboards. That's a sum total of three different boards.
• I don't use PPA's clocks (not least because they're too expensive for me, and because trialling one someone else supplied didn't beat what I had), I use something different - happy to provide details for those interested. 
• PPA sells his clocks to chipsets and not individual motherboards - can only go off what's on his website (http://ppaproduct.blogspot.com.au/2014/02/clock-upgrade-module-for-main-board.html) - i do like what he's done with sourcing, integration and explanation though. If your emails are true, PPA should be less disingenuous before charging people USD$400. None of the comments you mention are cited on his product sale page. Are you suggesting they should be?
• At no stage has it been suggested - by anyone other than you - that SQ is solely about latencies.
• This won't work similarly on an Atom build (the very bit you've quoted). Different frequencies, different clock domains/architectures. Much like suggesting differences in capacitors are important from a Thin-ITX board perspective - it's true on all boards, of course, and more so anywhere where all power regulation is done on the same PCB.  
• This isn't guesswork. It's practical experience.
• What works on one and not the next motherboard isn't guesswork either. It's engineering. Happy to go down that rabbit hole with anyone that wants to. 
• The above is provided for free to those interested. 
• This isn't a pissing contest - don't make it one.

 

More than happy to have anyone point out where something suggested can be bettered, though we'll all need to do better than incessantly demanding point proofs.

[TP1]

<mod post> How about we have no 'point proving' in the thread, it isn't a competition. <mod post>

It's not about competition but about readers being able to rely on what's being presented in a DIY thread. Most people seem happy to show their achievements and provide others with part numbers, procedures etc . But if I am missing the point , I'm happy to let it go.

Sent from my iPad using Tapatalk

[joelha]

Personally, I don't need proof or evidence. If someone states that they have an experience which, from their post, appears to be credible, I'd like to read it.

 

Just one audiophile's opinion.

 

Joel

[proftournesol]

It's not about competition but about readers being able to rely on what's being presented in a DIY thread. Most people seem happy to show their achievements and provide others with part numbers, procedures etc . But if I am missing the point , I'm happy to let it go.

Sent from my iPad using Tapatalk

 

This isn't a scientific paper, its people writing about their experience with their hobby. Surely the overall result depends on the resolution of the total system including the room effects? There are too many variables to make this a reliably verifiable reproducible experimental study that can end with a dogmatic statement regarding the 'best' music server?

[Guest rmpfyf]

All good @rmpfyf (Ric).

Pulsar must have lower their prices since?. Afterall, it was a couple years back when I was heavily involved in any clocks, i2s..... Back than it was either 350Euro or USD550, which equal out $750 each, Australian. $950 was a typo. Pity Acko is no longer selling them on his new website. Otherwise, I could post the link here.

Regardless, I don't mind funding a trip to Victoria, for you to enlighten me with what you had achieved? I will bring my atom, All in One Server for your evaluations, along with my JZolla if interested. Pity my DDDAC is too heavy, weights approx 45kg and too fragile for the long haul.

Let set up a gtg mate? Tasso may be able to join, too? Let do this rather fleshing out individual's intellectual muscles!!

Can you please advise at soonest the feasibility of a home turf gtg? Alternatively, We (I) will welcome you with any of your mate (Statman) for a first class hospitality here if you could bring what you can achieved to enlighten us here?

Sent from my iPhone using Tapatalk

 

I was looking at a Pulsar Clock for my DAC (though Acko's site too!) Ended up speaking to Pulsar direct, they think they'll end up restarting their project shortly. Finally ended with with a different clock solution altogether.

 

Feasibility of a home turf gig? Zero, Chanh, we're renovating and are a young family. Which is partly why I don't have your JZolla, I simply (like some others, I'd imagine) can't afford it. As mentioned - it's an impressive achievement. I've tried the unmolested JS-1 in place of my HD-Plex 100W LPSU and it's a great step up. A JS-1 also costs more than my entire HTPC, so horses for courses. My HTPC does different things to yours by necessity - has to run Netflix (wife) etc. I've zero doubt here that your system sounds great. I've also zero doubt it owes at least an order of magnitude more than mine. I focus on other areas of development and can afford to share what I learn - in good spirit, I try to do so here. On that - I learned a LPSU is good, a better LPSU is better still and a OMG LPSU (as you have) is better still. My affordability simply stopped at "good". It'd be very gracious of you to fly over with yours, though it's not necessary to prove a point. 

 

I'm also not sure that big iron downstream of bog-standard Vreg stages is the best/ultimate way of doing PC power, but that's another debate.

 

There's really nothing intellectually muscular about it. Nothing you, I or anyone else does in this is new. We are all to various ends passing on applications of others' intellect as stuff we've learned. 

 

FWIW, my reference isn't someone else's HTPC, it's a modified Wadia CD transport into a very special amp feeding some 50's fullrange drivers around a hour north of Sydney. If you're familiar with that scene you'll know who's system this is straightaway. I'm pretty confident I'll never match that system. It's owner is correctly confident I'll try, and has a lot of patience for my endeavours, for which I'm grateful.

 

For what I can do - we can talk. To this end, Chanh, you've got my number, Tasso too, and it'd be good to speak. What can seem adversarial over written text can make for better translation in person or by voice.

[TP1]

This isn't a scientific paper, its people writing about their experience with their hobby. Surely the overall result depends on the resolution of the total system including the room effects? There are too many variables to make this a reliably verifiable reproducible experimental study that can end with a dogmatic statement regarding the 'best' music server?

 

 

Wait till we start comparing it all to vinyl......   .   

[Doncentric]

@@Tasso - PPA's mods fit a range of boards with a bunch of generic chipsets, I'm just not sure they cover Atom chipsets. Actually I'm pretty sure they don't - the NM10 Express chipset on the motherboards you mention has some pretty specific clock requirements (32.768kHz crystal, pages 56, 74 and others of http://www.intel.com/Assets/PDF/datasheet/322896.pdf). This crystal drives the Real Time Clock (RTC) on a relevant Atom board, and from that derives a number of other system timings, but bear in mind it's a kHz reference driving a few MHz sources - not ideal. There are other inputs required; PCIe and LAN may have their own oscillators (100MHz and 50MHz or some multiples thereof - usually 25MHz, e.g. http://www.intel.com/content/www/us/en/embedded/products/networking/ethernet-connection-i218-gbe-datasheet.html). 

 

 

 

Totally wrong. I am horrified to think there is a motherboard out there that derives it's main clock from the RTC clock.

[proftournesol]

Wait till we start comparing it all to vinyl......   .   

 

I remember going to an Antipodes Audio demo evening that PMG put on and even though the NZ guy was confident that his server sounded pretty good (and it did) he was happy to concede that even he was amazed at how much better Warwick's Air Force 1 sounded. That doesn't mean that we can't enjoy digital sound though

[davewantsmoore]

Changing the clock is overrated. Far and away the best thing for  sound quality improvements is improving the quality of power  that gets to the clock, CPU and so on.  I dont intend  wasting too much more time fiddling with software tweaks.

 

Indeed.  IMHO .....   The only reason to mess with the clock, is to it can be powered separately.    The only reason to power it separately is so it's power supply doesn't touch anything else (cos it's full of noise).

 

ie.  the improving the quality of the clock is unhelpful....   you just have to make sure the clock (power supply) isn't allowed to cause harm to something else.

[davewantsmoore]

A decent quality crystal is hard to beat and better still when it is powering by an isolated/seperated fast clean source. To achieve the best outcome, quality of the power source, implementation, and quality of that crystal must be at equal weight. 

 

I don't really understand why that would be.    Can you explain how the quality of the clock itself is important?   (what actual effect does it have?)

 

 

For your information, phase noise of a particular crystal will improve relatively to the quality of its own PSU and the surrounding environment of that particular crystal's placement. 

 

Thanks .... and sorry if I have misunderstood .... but to me this doesn't answer anything.   Yes, the clocks power supply will affect the absolute phase noise of the clock.

 

.... but why is the phase noise of the clock critical in this situation?

Edited May 21, 2016 by davewantsmoore

[davewantsmoore]

I did it to my Cambridge Audio 752, well to be more precise, Joe Rasmussen did it.

Ease of such a mod all depends on the hardware. At least with the Oppo board design, many people are doing clock mods so there is a common guide and pool of experience to work from.

 

But this is a clock which feeds the DA converter, yes?! ..... that's a bit different, as it's obvious why the clock is important there.

[Guest rmpfyf]

Totally wrong. I am horrified to think there is a motherboard out there that derives it's main clock from the RTC clock.

 

Well, we've the start of a discussion here.

 

There's no notion of a 'main clock'. There are many clock sources on any motherboard.

 

I've written only what you've quoted, @@Doncentric - the relevant Atom chipset has an RTC XTAL input only, no other direct crystal inputs, there are some signalled inputs from other peripheral devices that have their own clock sources. The chipset appears to have no 'main clock' of it's own bar the RTC, which is different to a Core chipset (where what's in market needs a 25MHz or 24MHz source). Point being I don't think that particular mod (replacing the chipset clock input) on a Core motherboard is replicable on an Atom board. You could replace the RTC reference, but - as already stated - I don't think the gains would be as great (if at all). 

 

If I'm wrong here please let me know (it's been over a year since I've played with an Atom board.  I've linked the appropriate dev guide if it helps).

 

Some have had success replacing the USB chipset clock (where it exists) as well, I've not had time to try this. For the cost involved a dedicated USB card would be comparable in cost and likely better in performance. 

Edited May 21, 2016 by rmpfyf

[TP1]

I don't really understand why that would be.    Can you explain how the quality of the clock itself is important?   (what actual effect does it have?)

 

 

 

 

To clarify, are you saying that the clock quality is not critical or not as critical as the clocks power supply?