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

[Johnseye]

12 hours ago, tboooe said:

Well I got the Tera Grand Cat 7 cables in 3 ft lengths and used them from my network bridge to my Server PC and for the direct connection between the Server PC and NAA PC.  I cannot be 100% and I doubt I could consistently tell the difference between my previous Blue Jean Cat 6 cables but I think the Tera Grands have a bit better extension at the higher frequencies.  At this point, I am inclined to keep them in my system if for nothing more than peace of mind.  Thank you for the tip.

 

I had been using these cables for a month or so with both my server to switch and from FMC to mRendu.  My server is on a different circuit than my other equipment which can cause a ground loop.  This is exactly what I think happened when using this cable.  I was getting a low hum from my speakers and when I swapped this cable out with a standard cat5e the hum went away.  This is repeatable.  I have no doubt the cable is contributing.  It's not the cause, that's the different circuits and a grounding issue. It is amplifying the hum however.  The only thing I can visibly see being a difference is the metal connectors on the Tera Grand.

 

I then pulled it from the FMC to mRendu and swapped it with a Blue Jean cat6a.  There was no ground loop concerns with this connection.  I "think" I can hear a difference, but it's subtle and I can't be sure.  I'll need to test it a lot more.  Question is, what is this cable doing?

[Johnseye]

4 hours ago, seeteeyou said:

FYI - no need to bother with any MLC or even TLC SSDs since they couldn't even hold a candle to SLC ones according to many audiophiles in Taiwan, they also recommended SuperSSpeed S301 in addition to Intel X25E

 

http://www.myav.com.tw/bbs/showthread.php?threadid=349951&pagenumber=483

http://www.tweaktown.com/reviews/4790/supersspeed_s301_hyper_slc_120gb_solid_state_drive_review/index.html

 

They're very affordable (< $50 for 60GB and < $75 for 120GB) but importing them from China might cost quite a bit of international shipping

 

http://www.supersspeed.com/productAll.asp?sid=93
https://item.taobao.com/item.htm?id=39329474666
https://item.taobao.com/item.htm?id=39360536141
https://www.facebook.com/SuperSSpeedUS/

 

Why would anyone use a legacy drive over an M.2 unless the board wasn't M.2 capable?

 

[Johnseye]

On 4/3/2017 at 1:55 AM, romaz said:

 

The Pachanko SATA Reference cable results in a smoother and yet more textured presentation and very noticeably attenuates the slight glare present with even the X25-E when connected to the inexpensive SATA cable.  It makes the much brighter glare from the Samsung SSD more tolerable and when I briefly went back to my Mac Mini with the OS on a PCIe NVMe SSD, the contrast between the two is even more stark.  I would say avoid NVMe if at all possible.

 

While the Toshiba hard drive when connected to the Pachanko SATA cable and powered by the LPS-1 also sounded very good, I have completely ruled it out due to vibrations.  It would reside on an aluminum tray right next to where my clock board will be placed and even at idle, this drive will likely be spinning and so I can imagine how that could negatively impact the clock board.  Furthermore, as the drive spins, it is audible which is a no go for me.

 

Are you recommending to avoid NVMe because of one test with a Mac Mini using an NVMe drive?

When I hear descriptors such as smoother and textured vs stark I begin to think of comparisons between tube and SS amps where they are described very similarly.  What if someone prefers a stark sound that may be equated to clear and transparent?  Not trying to ruffle any feathers and all this testing you're doing is great, but it's very subjective.  Sometimes there's no other option but to be subjective and take someone's opinion, but as you make recommendations for others to not use a drive type because it doesn't mesh with your preferred sound signature you assume others want that same sound.

[Johnseye]

On 3/25/2017 at 9:15 PM, romaz said:

 

Based on what I could find, the Intel X25-E SSD, in particular, caught my attention.  These drives are no longer in production but they are still readily available and I was able to secure a brand new 64GB drive for only $80 (originally $800 when they were first released).  Because they are based on SLC NAND, they have 10x the number of write/erase cycles as the cheaper MLC or TLC NAND based SSDs and so they should last a very long time.  Some have suggested that SLC also sounds better than MLC or TLC and this may be due to better controllers since SLC SSDs are generally "cost no object" drives designed for data critical servers.

 

If you believe this is the case have you tried new Professional or Data Center series drives that come in both 2.5 and M.2 flavors?  One would expect even better controllers from current generation drives designed for "cost no object" data critical servers.

 

[Johnseye]

On 3/29/2017 at 10:10 PM, romaz said:

 

Clean power is always important but line resistance is maybe even more important only because the PSRR of well-built components is so high these days.

 

To be oblivious to what good power cables can provide with respect to line resistance is to be oblivious to a whole other world of dynamics hidden within your electronic gear trapped by poor cables.

 

This exact argument has been the center of the speaker cable debate.  I agree that it directly translates to all components.  The highest quality materials for a device will eliminate any question as to whether a lesser quality material diminishes performance.  Most companies, especially those which mass produce for large markets, will select materials that meet specification requirements.  Beyond that only introduces higher cost and lower margins for little benefit.  For us consumers, especially those of us who want to get the maximum performance with cost being no object, mass market products aren't good enough.  The question for us becomes, what is the point of diminishing to no return.  I think we can all agree that these tweaks offer small degrees of improvement over macro changes such as speakers/headphones, preamp, amp.  It's been somewhat taboo to manipulate the materials of these devices, whether that's due to warranty or not wanting to experiment on such high cost devices.  What if you were able to dramatically improve the performance of your speakers, preamp or amp by introducing higher quality materials?

 

On another note...

Here's a device I came across that you may have already seen.  I don't know much about it, but it looks interesting as a power measurement tool.

 

http://www.quarch.com/products/hd-programmable-power-module

 

This HD Power Module is a flexible power margining and measurement tool and a significant upgrade on the XLC module. This version has 12V and 5V (switchable to 3V3) outputs and is ideal for SAS, SATA and PCIe based devices. The outputs can be set automatically when used with advanced Quarch fixtures.

Each rail can be programmed from 0 to 120% of its nominal voltage. The voltage and current of each output can be simultaneously recorded to provide accurate power consumption measurement.

The module allows programming of arbitrary voltage output waveforms which can be used to simulate power-up ramps, power failures, noisy power sources and much more. Patterns can be created over a range of microseconds to minutes, allowing a wide range of testing options.

A high-speed record feature provides an oscilloscope-like function to record the output voltage, current and power consumption of the attached device.

[Johnseye]

1 minute ago, seeteeyou said:

 

Why wouldn't? If you don't like my peaches, don't shake my trees.

 

Anyone could feel free to compare any drives at one's own expenses, please go ahead to test them out and report back

 

http://www.addonics.com/products/indm2.php

 

Until then, does it make sense to stick with something relatively affordable that's tried and true to boot? It's easier to assume that new ones are superior to legacy drives but someone will have to pay for the hardware.

 

I was simply linking to previous experiences from fellow audiophiles while anyone could share field-tested experiences on the Internet these days. With that said, I would be more than interested in a comparison between SATA and M.2 drives.

 

I was under the impression that this was a search for the best quality sound components, at least all these A to B comparisons seem to be focused on what sounds better.  I'm looking at building a new media server and before I do want to ensure I build it with quality components.

 

It's not that I don't like your peaches, but you made a point not to even bother with MLC or TLC based on some people's opinions of I'm not sure what, noise? performance?  So I thought I'd ask why mess with any 2.5 drive when M.2 can run at lower power.  Maybe someone has done the comparisons with some validation, or has taken measurements with something like the Quarch device I linked.  I haven't but would like to know.

 

I don't think SSD read speed performance is an issue for playing music so it really comes down to noise and that's related to power unless I'm missing something.  Even when you look at consumer, pro and data center SSD's, when you remove performance as a variable you get down to features such as encryption and temperature.  The only relevant variable I can see is power consumption.  Power consumption will only impact the performance of the device processing the data.

[Johnseye]

30 minutes ago, romaz said:

 

Last I checked, this is an opinion forum meaning statements made by posters are subjective.  All of my comments are indeed subjective and unless I've specifically indicated that I've brought out measuring equipment, my descriptors are based on what my ears tell me.  There are very few absolutes in high-end audio, only personal preference and to take my comments as anything but an opinion would be ridiculous.  It seems I've somehow ruffled your feathers because our preferences differ.  Let's just agree to disagree.  I'm totally cool with that.

 

Regarding M.2, this is only a connection spec and doesn't really say much.  With M.2 storage devices, you have to specify SATA vs NVMe because they operate on different buses.  Check the block diagram of your motherboard and you will find that M.2 SATA drives will likely go down a SATA III bus whereas NVMe goes down a PCIe 3.0 x4 bus.  NVMe drives are much lower latency which is a good thing but you are misguided if you believe NVMe drives are low power.  The goal of anyone looking for an NVMe drive is generally throughput but this comes at the expense of power and noise.  Look at the specs of any NVMe drive and you will find that they generally consume anywhere from 4-6 watts.  As 3.3V devices, that's approaching 2A of current draw.  The legacy Intel X25-E SSD that you seem to have a low opinion of has a typical draw of only 200mA.  That's nearly 10x more current draw.  That's quite a difference.  

 

The other disadvantage of M.2 is that these devices are bus powered.  The PCIe 3.0 bus operates at either 12V or 3.3V but NVMe drives generally utilize only 3.3V.  The 3.3V rail on any motherboard is the noisiest rail there is.  Just check the performance specs of any ATX PSU and you will see how much more ripple there is on the 3.3V rail.  More than that, just check the transient response of even the very best ATX PSUs when under load and see how they struggle to maintain their voltage at 3.3V.  Even the EVGA 1.6kW Titanium ATX PSU that is highly regarded by respectable audiophiles here on CA struggles at 3.3V.  This is the voltage rail that generally brings many PSUs to its knees.  

 

While SATA has it's issues, my SATA situation with my build is unique.  I am using a SATA to PCIe adapter card meaning my SATA connection directly connects to my CPU on a very uncrowded and low latency PCIe 3.0 X 8 bus and so my latency is much lower than a typical SATA drive.  While this PCIe card operates on this noisy 3.3V bus, there is at least some noise isolation brought about by the installation of an ultra low noise linear voltage regulator and high-quality capacitors.  Furthermore, this adapter card's clock is being replaced by a very high end clock that, at least in theory, should help minimize transmission errors and reduce the need for resends.  Lastly, because I am using this "legacy" SSD, I can externally power it with my LPS-1.

 

Having taken the time to have compared and listened, I feel comfortable with the path I am taking.  If you prefer the high frequency energy brought about by NVMe, I'm happy for you that you know what you like.  

 

I intended to play devil's advocate because opinions sometimes become gospel without proper validation.  I meant no offense to you or your methods, and appreciate your input.  Please don't take offense.  The point I made about NVMe was only because you had done one test, came to a conclusion and as a result people may begin to discredit NVMe as an option. I think we should be careful when making broad statements after conducting one test, that's all.  As I mentioned, I'm not concerned about SSD performance for music playback.

 

I should have specified, but the M.2 drives I was referring to are SATA.  Yes they use the 3.3v rail but the drives run at around 80-100 mW as opposed to the 2.6W of the X25-E when active.

 

I'm in search of building a better server, so I'm trying to separate opinion from fact when doing this.  I'm also trying to look for the qualities I believe are important as you are.  I was drawn to your post because like you I have an sms-200 and microRendu.  I would describe one as brighter and the other as smoother.  Very subjective statements but the best a non-scientist can offer.  Same as how you described an NVMe drive.  Do you like bright or smooth?  Forward or laid back? Apples or oranges?  It's up to the listener right?

 

We can really get into the weeds when evaluating this equipment to determine if there's a difference in sound, especially considering how subtle the changes are.  I'm on this journey to learn and improve my system so it sounds the best to my ears.  My feather's aren't ruffled, I'm just trying to keep the playing field level as I read and learn.

[Johnseye]

39 minutes ago, Ralf11 said:

A thread on building an ultimate or near ultimate media server would be worthwhile (even If I personally just keep my mac mini...).

 

I want to partly disagree with another stmt above by modifying it: for "those of us who want to get the maximum performance with cost being no object, mass market products [often] aren't good enough."

 

I agree about the diminishing returns observation you made after that, but there is another issue - that a mass-market component might be quite good enough if, say, circuit topology is all-important and some quality of the component is unimportant.  I'm sure there are other examples.

 

Agreed, mass market products can make a product good enough and some can do it at even a lower cost than one would expect.  I guess my point was that we go through all this effort of tweaking the smallest impacting things like a SATA cable with higher gauge, when we could be missing something a lot bigger.  I never see anyone posting about how they opened up their McIntosh amp and changed out their cabling.  Now I did see Jeff (Jeff's Place) swap out all his Dueland cables with the Western Electric stuff.  He also has a Listening Bias page as a disclaimer which I appreciate.

 

[Johnseye]

22 minutes ago, romaz said:

 

I often do seek out enterprise-grade gear as I do believe they are often more robustly engineered and built to a finer tolerance.  My main problem with the latest generation stuff is that they prioritize speed meaning low current draw and, therefore, low noise is much less of a priority.  Sometimes the improvements are so good that its worth the extra draw but I have to be convinced of it since my experience, in many cases, has been that they are not.  

 

It's clear the path I'm taking is counterintuitive to how many view a music server should be built and so reader beware.  While many are looking for the fastest there is, I'm looking for the slowest that I can get away with.  While there are those that look to Data Centers for the next best thing, I am hunting through E-Bay in search of NOS SSDs and lowly compact flash cards.

 

I think different requirements are in consideration between consumer, pro and data center equipment.  I'm not sure which is best for music reproduction, it probably depends on the component.  If you consider a motherboard, there is a lot more you can tweak with a gaming board vs. enterprise board. One is built to tweak while the other is built for reliability.  My only concern is whether it impacts the sound for better or worse.  I don't think something should be discounted based on age or target market.  Hopefully you discover something good in the ebay bins and we all benefit.  You certainly have invested a lot in your power back end and nothing has held you back.

[Johnseye]

1 hour ago, romaz said:

Obviously, they support that SATA cables do matter even though the data these cables transmit are buffered into RAM.  As I have previously posted, with cables, I believe the lower the line resistance, the better, and this is what shorter cables offer.  Because cables also act as antennas for noise, it would make sense that shorter cables are also less noisy.

 

Why not use M.2 and eliminate the cable all together? 

 

[Johnseye]

31 minutes ago, romaz said:

 

I suspect your SATA SSD might consume 80-100mW while idle but not during writes or reads.  That would be unprecedented if it did and if indeed it's true, then please share which drive you use so I can purchase it immediately.

 

Intel claims that the X25-E consumes 60mW while idle and 2.4W during sequential writes.  While 2.4W doesn't sound impressive, what caught my attention was how it performed in this benchmark:

 

http://www.tomshardware.com/reviews/intel-x25-e-ssd,2158-7.html

 

This benchmark suggests during workstation level disk I/O, the X25-E consumes only 1.1W.  Green Leo previously wondered how the X25-M compared which is based on MLC NAND and it did well but not as good as the X25-E.  It has been stated numerous times by others that SLC consumes less power than MLC and this supports that claim.  What is important with respect to noise, however, isn't power consumption but current draw.  1.1W at 5V equates to less current draw and, therefore, less ground plane noise than the same 1.1W at 3.3V which is the voltage that M.2 devices operate at.  

 

Assuming your M.2 SATA SSD consumes 80-100mW at idle, at 3.3V, that equates to 24-30mA of current draw.  With the X25-E, which is a 5V device, 60mW of consumption equates to only 12mA of current draw.  Based on the benchmark performance of 1.1W of power consumption, that equates to about 200mA.  That's a pretty good number.

 

At the end of the day, it's about how it sounds that matters and not how much current it draws nor am I convinced that the X25-E is the best there is.  I will wait to compare against compact flash before making a final decision but thus far, to my ears and with this particular build, the X25-E is what has sounded the best.

 

 

I was basing those numbers from Intel's site. I can list some models if you'd like.  Comparing what they state for some m.2 drives and the x25-e. For what their spec literature is worth, they state active. 

 

That x25-e is a 32 gig drive which makes a difference.  Even if you have the 64G drive that's not much space and the power consumption will be slightly higher.  The review you pointed to also compares the 32G x25-e  to a 64G Samsung drive which does very well and beats it in some tests. They even have a streaming read power test where it beats the x25-e by almost half the consumption with twice the storage capacity. Did you happen to test that Samsung? 

 

Newer Samsung m.2 drives are also rated pretty well in power bench mark tests. Plus there's the benefit of not using a cable. 

 

I get the 5v vs 3.3v argument but if nothing else is going on with this server, meaning its services have been disabled to the point where it's only streaming music I don't how much it matters. There's little to nothing else on the bus. I'd personally prefer no cable especially since it appears there's some kind of impact from one. 

 

I'm curious to hear your compact flash results. 

[Johnseye]

On 3/8/2017 at 7:23 PM, romaz said:

LAN Adapter Card - Intel E1G42ETBLK Dual Port PCI-Express Server LAN adapter ($145)

So my solution to Lee's statement was to bypass the dual Intel LAN ports built into the Gigabyte motherboard with this card. They incorporate Intel's slightly older but highly regarded server-grade LAN chipset, the E1G42ET. In contrast to their newer I350-T2 chipset, this one consumes less power (2.8w vs >4w) but is also devoid of Intel's integrated power saving feature. Because it is PCIE, it has a lower latency pathway to the CPU and each LAN port has it's own independent path. Moreover, instead of 2 separate clocks for each LAN port on the Gigabyte motherboard, this card uses only a single clock which leaves me with 2 free clocks. To my delight, Lee has told me he can also replace the switching regulators on this card with SOtM's ultra low noise voltage regulators and further add capacitors. This would be as close to an audiophile LAN card as I can think of and may possibly allow me to remove my reclocking switch from the direct path and reposition it between my router and music server.

 

 

 

Not sure if you bought this card but the BLK is a bulk sku.  You can get the same card for about $30.

 

On 3/12/2017 at 5:48 PM, romaz said:

I will now be abandoning this Gigabyte motherboard. Before sending it off to SOtM for clock modification, I felt it was important to build it first to make sure my selected components were compatible and I'm glad I did. Unfortunately, none of the newest boards that incorporate Intel's 1151 chipset is capable of SATA in pure IDE mode (only AHCI). Consequently, I am finding that it is not possible to use a CF card as a boot drive which is a deal breaker.

 

I will have to go backward to an older motherboard that utilizes an 1150 chipset and I have found several that I like although I presently have my eye on an Asrock server-class enhanced mini-ITX motherboard, the MT-C224, that has even fewer unnecessary ICs such as onboard audio and can still be powered from a single 12V lead from my SR7.

 

Using this older chipset will also force me to abandon my 7th generation Intel 7700T but the advantage of this server-class motherboard is that it can utilize a Xeon class CPU and ECC RAM. This motherboard is specifically compatible with an even lower power dual-core Xeon, the E3-1220L V3, with only a 13w TDP and yet still has a fairly generous 4MB of L3 SmartCache (meaning the entire cache is usable for a single core).

 

 

I'm curious, with a max of 1.5 GHz will the E3-1220L V3 be able to handle running Roon?  It looks like you're not interested in, or are giving up any ability to upsample for the lowest possible power spec.

 

I'm hesitant to even go down to the 35w 7700T because I'd like this server to be able to upsample.  That puts things at 65w and I'm tempted to give the Ryzen a try as it's 8 cores.  This is where I'm at in my search for optimal components.  My desire to upsample causes me to compromise.

 

Other than the X25-E hard drive what other components have you finalized for your build?

 

10 hours ago, romaz said:

 

Feel free to share which M.2 SATA drives you like.  My intention isn't to debunk them.  If you like them, that's all that should matter and there may be some who have no choice but to use an M.2 drive and so the information could be helpful.  

 

I definitely agree, with only one exception that I can think of, the best cable is no cable and for some time, I, too, had assumed mSATA or M.2 and especially M.2 NVMe would sound better than standard SATA.  While I understood there would be increased noise issues, I assumed the massively improved latency of NVMe would win the day but that has not been my experience thus far.  I suppose it's like overclocking a 16-core CPU to stream music files -- that type of speed and power just isn't necessary and the noise consequences are just too great.

 

With regards to the benefits of not having to use a SATA cable with an M.2 SATA or mSATA drive, like with all things, my recommendation would be to use your ears and decide what you like better in your system.  If you have zero plans to address your SATA cable issues, maybe you'll find that there isn't much difference or that you prefer the "cable-less" benefits of M.2 SATA or mSATA but having specifically compared an mSATA SSD vs a standard SATA SSD that was independently powered by a separate PSU in my NUC (even when using a cheap SATA cable), my preference was for the independently powered SATA SSD.

 

The reason for this, I believe, is even in the absence of other noise causing devices on the 3.3V bus, you still can't escape the fact that this is inherently the noisiest bus on most motherboards.  First, because it is generally the noisiest rail on many ATX PSUs and with some of the highly regarded ATX PSUs I've looked at, depending on the load, it can be more than 2x noisier than either the 12V or 5V rails.  Second, you simply can't escape the noisy switching regulators that are buried in your motherboard.  Unfortunately, any off-the-shelf motherboard is going to be a nest for noise and to bus power any device is seldom going to have positive SQ results.  Lastly, for every watt of power consumed by a 3.3V device and a 5V device, the 3.3V device will draw 1.5x more current.  It takes much less for a 3.3V device to generate noise.

 

As for 32GB or 64GB drives, yes, these are small volumes but the intention for these drives has always been to use them as OS drives.  In my testing, OS drives have the greater impact on SQ compared to music storage drives and this would make intuitive sense.  OS drives are always active and the noise they generate is random, which I believe is the worst type of noise since it's very difficult to adjust to noise that is randomly coming and going.  

 

Since many software players will buffer to memory (Roon buffers 1-2 tracks in advance, as an example), music storage drives will generally go idle and stay idle for longer stretches of time and so their impact on SQ is potentially less, that is provided that the storage drive you choose has very low current draw while idle.  The latest 2TB Samsung 960 Pro M.2 NVMe, for example, draws the same amount of current while idle than a 2TB Western Digital Blue SATA III hard drive draws while actively writing (about 340mA)!  In even more stark contrast, many compact flash cards while idle consume only 100uA.  That means that In this particular example, while idle, the Samsung 960 Pro M.2 NVMe SSD draws 3,400x more current than a compact flash card.  

 

With regards to that benchmark that had the Samsung SATA II MLC SSD pitted against the X25-E, the Samsung consumed 1/3 less current during sequential reads but they were essentially neck and neck during the workstation-type I/O benchmark which is the more important benchmark for an OS drive.  While I did locate a used Samsung SATA II MLC SSD that I could have purchased to try, I wasn't curious enough to want to spend the money or the energy to try it, especially since there are enough reports out there by people I trust (including Phil Hobi) who feel that SLC drives generally sound better than MLC drives.

 

Here's an example of an M.2 drive, and the most comparable in size to the X25-E at 48G.

https://ark.intel.com/products/94438/Intel-SSD-E-5400s-Series-48GB-M_2-80mm-SATA-6Gbs-16nm-TLC

 

Here's a 128G NVMe with a 100mw active rating

https://ark.intel.com/products/94910/Intel-SSD-E-6000p-Series-128GB-M_2-80mm-PCIe-3_0-x4-3D1-TLC

 

I could be comfortable with a 48G drive at bare minimum.  32G pushes it too close for my liking.  All my music is NAS stored so this drive is just for OS, Roon, JRiver and HQ Player. 

 

Unless I'm missing something, which is very possible, these drives have low power requirements.  I don't own them or have any skin in the game.  I'm just looking to make the right decision when I build my next server, just like you.

 

That Samsung 960 Pro is an example of a very power hungry drive.  A better example of a Samsung is the 850 Pro.  I think you have the 850 EVO but I believe the 850 Pro has even lower requirements.  Not sure why you chose the X25-E over the Samsung in that review you posted, but it looks like you trusted SLC better even though there was less power draw during reads.  Tradeoffs.

 

[Johnseye]

14 minutes ago, lmitche said:

The new optane nvme sticks are going to be an interesting.  I'd definitely get an optane compatible motherboard for a new build.  The optane memory will be used for cache against a local drive.

 

At first look I don't think it would help much for streaming.  I'm not even sure it would be beneficial for upsampling.  It seems to target desktop applications providing a storage buffer between the drive and processor for faster access to common calls to application storage.  I can see its benefit for a lot of purposes, but just don't see any big gains for computer audio.

[Johnseye]

10 minutes ago, ElviaCaprice said:

Not everyone is looking to upsample, especially those of us with Chord DAC's.  Low power, low impedance server is our goal.

 

No, not everyone is looking to upsample and not everyone has Chord DACs, but some people are looking to upsample.  I wasn't implying that everyone was trying to upsample, only that it wouldn't be possible when selecting components with only low power in mind.  Just something to consider when building a server.

 

There are people out there building high powered servers in order to upsample to DSD 512 without a hiccup.  It would be interesting to know how much noise is generated by not only having the components necessary but also pushing the system to do that upsampling.  What's the trade off in sound quality?

[Johnseye]

19 minutes ago, Ralf11 said:

my question is whether everyone should looking to upsample...

 

I think that's subjective and not a matter of everyone or no one.  If you like what you hear when you upsample then go for it, if you don't then that's that.  I personally don't do it right now.  I tried for a bit and wasn't overly impressed.  I have read a lot of positive experiences when upsampling to DSD512 but my DAC can't do that.  I'd like to hear what it's like and will eventually buy a DAC that can support it, which is why I'm considering server components that can.  I personally prefer PCM over DSD but that could just be because my DAC isn't as good at converting DSD as PCM.

 

or was that tongue in cheek Ralf?

[Johnseye]

Just now, tboooe said:

I was checking out the Ryzen as well and it seems to fit my preference for a fanless system.  However, as I understand it that chip does not include any sorts of graphics so I think you will need to add a graphics card.  No big deal as these are cheap unless you want to use one for CUDA offload with HQP.  However, even with a low end card, I am concerned that I am introducing more noise and issues?  This is why I am leaning back towards an intel cpu.

 

Some mobo's will have onboard graphics, but what's the impact I don't know.  It does separate it from the processor which would be a positive when considering processor performance.  Unless I want to wait a year or more for the kinks to be worked out I suspect mobo manufacturers are going to need some time with the new architecture.  I've been reading about bugs and I abandoned AMD years ago because of issues on the bleeding edge.  Another issue is the lack of micro sized boards currently.

[Johnseye]

1 hour ago, JohnSwenson said:

I have covered this fairly extensively in my series at Audiostream.

http://www.audiostream.com/content/qa-john-swenson-part-1-what-digital#35WEoHuMXmJ00tIX.97

http://www.audiostream.com/content/qa-john-swenson-part-2-are-bits-just-bits#Fz4mY1iS5fqvIUkV.97

http://www.audiostream.com/content/qa-john-swenson-part-3-how-bit-perfect-software-can-affect-sound#6yFiDB63CIzx5f3V.97

 

John S.

 

Yes this explains what we're all chasing in our playback chains. Thank you @JohnSwenson

 

One interesting piece of information makes me question whether Intel's new Octane will help with reducing memory access noise. 

 

"A major part in this memory access noise is what is called cache performance. Every processor includes a “cache memory”, this is memory in the processor which operates in sort of a shadow mode. Once the processor accesses a particular memory location, it stores that data in the cache, so the next time it tries to get that location, it doesn’t have to access main memory. There is only a fairly small amount of cache memory available, so an access does not always get the value from cache, this is called a cache miss. Cache is used for both instructions and data.

A program can influence the cache performance in many ways. One of the easiest to understand is just plain size of code. If the main loop of a program is small enough to fit entirely in cache, the program doesn’t have to do any main memory accesses for its instructions."

[Johnseye]

9 hours ago, Johnseye said:

 

Yes this explains what we're all chasing in our playback chains. Thank you @JohnSwenson

 

One interesting piece of information makes me question whether Intel's new Octane will help with reducing memory access noise. 

 

"A major part in this memory access noise is what is called cache performance. Every processor includes a “cache memory”, this is memory in the processor which operates in sort of a shadow mode. Once the processor accesses a particular memory location, it stores that data in the cache, so the next time it tries to get that location, it doesn’t have to access main memory. There is only a fairly small amount of cache memory available, so an access does not always get the value from cache, this is called a cache miss. Cache is used for both instructions and data.

A program can influence the cache performance in many ways. One of the easiest to understand is just plain size of code. If the main loop of a program is small enough to fit entirely in cache, the program doesn’t have to do any main memory accesses for its instructions."

 

*Optane

[Johnseye]

7 hours ago, romaz said:

So, I found a way to externally power an M.2 SSD (either SATA or NVMe) provided that you have a free PCIe slot and it's an inexpensive solution:

 

http://www.addonics.com/products/ad2m2s-px4.php

 

If you click on the Specifications tab, you'll see that this board offers the option of PCIe bus power although it specifically uses the more desirable 12V rail (and not the 3.3V rail) from the PCIe bus.  Should you connect a 5V SATA power connector to this card, even better, the card reverts to external 5V power.  Really pretty clever.

 

The only question, is this a better way to go?  If you have no desire to invest in better SATA cables, it could be.  M.2 SLC SSDs are available but they are still current hogs drawing up to 2A in current and are prohibitively expensive:

 

http://www.addonics.com/products/indm2.php

 

It's possible the new Intel M.2 SSDs could work well.

 

Thanks for finding this, and for your help with discussing component capabilities. This is a great option for M.2.  I just ordered one of the newer Intel M.2 drives.  Unfortunately none of the mobos I've been looking at are just exactly perfect.  I had wanted an ITX board with on board 12v but it looks like it will need to be a microATX with a PicoPSU adapter.  Not sure how the adapter will impact things if at all.  This does open up the door to two PCIe slots and the possibility of using the Addonics card along with the Intel NIC.  It also allows an Optane ready board in the event that proves to a beneficial architecture.

 

I did end up going with the i7-7700T so will see if 35w is enough power to upsample.

[Johnseye]

A question for @romaz or anyone familiar with connecting a mobo to an external LPSU such as the SR7.  With an onboard 12v DC connection it's fairly straight forward.  I thought the PicoPSU adapter could be used but it appears this might be a PSU itself, although it's hard to tell.  Is this used to connect a 24 pin ATX to an SR7 or is there another kind of adapter?

[Johnseye]

3 minutes ago, ted_b said:

The PICO PSU takes 12V (I assume, I had a 12V one and a 19V one and used my Hynes both ways, dialing up or down the voltage internally on the Hynes) and delivers it to the peripherals, etc via the 20-24 pins.  At the other end is a female dc input.  Plug the Hynes in there, or find a spot on the case where it attaches to a dc plug.

 

Ok then the PICO PSU isn't exactly a full PSU, but more of an adapter.  Then it should work connecting any mobo to a Hynes.  Did the mobo spec or the total device power requirement determine whether to use 12v vs 19v?

[Johnseye]

7 minutes ago, romaz said:

 

The Pico PSU is a DC-ATX adapter that takes a 12V input.  HDPlex makes one also that takes a 16-24V input.  The HDPlex is better designed and has better specs.  By the account of many here on CA, it is the better sounding converter as it uses better grade capacitors and inductors.  It also has greater headroom.  Unfortunately, neither of these converters can be considered high-end without modification but at the very least, they're not very expensive either.  

 

These adapters are necessary for any ATX motherboard because the ATX power spec calls for specific voltages (+12V, +5V, +3.3V) as well as a -12V rail for legacy RS-232 and PCI ports and a 5V standby rail to provide soft power to the motherboard when you have your machine turned off.  Since no typical LPSU will have all these rails, these adapters become necessary and the 24-pin ATX connector that is used will then have the necessary pin layout to feed the motherboard these specific voltages.

 

An alternative to using such an adapter would be to buy an ATX PSU.  With the adapter, you have the voltage regulators within the adapter itself.   With an ATX PSU, all the regulators are within the PSU chassis.  Aside from HDPlex and Teradak, all the ATX PSUs you can buy today, as far as I'm aware, are of the switching type.  Switching ATX PSUs are much less expensive, smaller in size, have higher efficiency and run cooler.  Linear ATX PSUs have the potential to have a lower noise floor and lower output impedance but they come at a greater cost, generate more heat, and presently max out at 820w.

 

If you buy a thin mini-ITX or mini-STX motherboard, then the regulators for the 5V and 3.3V rails are integrated within the motherboard.  NUCs and Mac Minis fall into this category also.  This is why you can feed these motherboards a single 12V lead from something like a Paul Hynes SR7 or Uptone JS2.  Unfortunately, because most 12V supplies max out at 10A, such motherboards will rarely contain multiple high-speed (and high-power) slots.  Generally, the most you will find is a single PCIe 3.0 x 8 slot and/or an M.2 PCIe 3.0 x 4 slot.

 

Assuming I went with the HDPLEX 160W DC-ATX would I need Paul to make one of the rails of the SR7 a 19v?  If I go with the Pico it stays 12v but the sound quality could suffer?  Is 12v enough to power a miniATX, or does it depend on the total watts used, or what's being powered by the different voltages?

 

I'd prefer to avoid the ATX PSU as I'd at least like to benefit from the low noise of a Hynes LPSU.  The mini ITX and STX boards are very limiting as you know and I don't want to limit myself or paint into a corner if possible.  Sacrifices are being made here, but I'd like to make as few as possible.

[Johnseye]

3 minutes ago, romaz said:

 

If you go the HDPlex route, you might as well get their 300W adapter.  It uses better parts even if you don't need 300W.

 

With the HDPlex, you would ask Paul for a 19V rail.  How much power is enough will depend on the draw of the components you select.  The SR7 utilizes a 250VA mains transformer and can output 125 watts continuously and 480 watts transiently.  If you have needs beyond this, it can be done but will be very expensive.  Paul has talked about building an ATX PSU but to have it perform to the level of his SR7, I am guessing it would easily cost more than $4k.  

 

As good as the SR7 is, some of its performance will be negated by the DC-ATX adapter and further limited by the switching regulators within any motherboard you buy.  For sure, there will be compromises.

 

I'll have to kick this around with Paul, he provided me a spec to power what I thought I was going to be building, plus my sms-200/mRendu and turntable.  Pretty sweet opportunity if I can limit myself to a single PCIe and M.2 which is where I'm starting to head.  There is the HDPlex 400w ATX LPSU option, but again more sacrifices.

 

Paul's asked me to determine the total wattage rating for the server rail which I'm struggling with.  I can easily add up the watts from the mobo's components, although the memory may be difficult as it isn't clearly stated.  How is it possible to determine what the wattage requirements are of the mobo itself?

 

6 minutes ago, romaz said:

I have come across a purpose-built audiophile class motherboard with linear regulators, short-signal paths, a low power CPU and has very good clocks.  It also runs its lightweight OS off an electrically quiet mini-SD card and can run Roon and HQP.  It can also be powered by the full glory of an SR7.  Pretty much everything you would want.  It's called the microRendu or sMS-200.  

 

Yup except they have their limitations too don't they, otherwise we wouldn't be investing this kind of effort in a server.  I think we'll get there one day.  It's an evolution, and considering John S's posts in Audio Stream were almost 4 years ago, it's a slow evolution.

[Johnseye]

6 minutes ago, romaz said:

 

Here's a place to start:  http://www.coolermaster.com/power-supply-calculator/

 

Thanks Roy

[Johnseye]

55 minutes ago, romaz said:

 

Understood, but both the mR and sMS-200 present us with an ideal model to try and emulate.  Good low power X86 processors with large secondary caches exist now.  No need to go to non-X86 CPUs and resort to emulation:

 

https://ark.intel.com/products/95441/Intel-Core-i7-7Y75-Processor-4M-Cache-up-to-3_60-GHz

 

The problem is the lack of suitable motherboards.  This is why I believe you need one of these types of endpoints if your goal is ultimate SQ since no single box server will ever have all the ideal characteristics that audiophiles desire.

 

I think motherboard manufacturers are seeing a market for audiophile quality components and are including some in their onboard sound, but won't make an audiophile specific board.  It's a niche market at the level we're looking to perfect.  We'll likely start to see more appliances with specialized boards and components, but it might be a while longer before someone builds a motherboard or even a prebuilt PC with a purpose designed board where they'll allow installation of any OS.  I hope it happens, but I think there are challenges to making it profitable.