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

Peter_T · March 24, 2017

On 3/23/2017 at 3:39 AM, seatrope said:

.....HQplayer refused to see it as a valid NAA. I also tried using a DHCP server on the HQplayer computer without bridging, which resulted in the same thing.....

I use a direct connection of my SOTM200 to my server/HQplayer-NUC and dhcpserver to give the SOTM an IP-address. Under W10 and W2016-core : HQplayer immediately recognizes the SOTM. The problem you are having may be system dependent?

romaz · March 24, 2017

7 hours ago, lmitche said:

Roy,

The SATA card is plugged into the PCIe 3.0 x16_1 slot of my Asus z170m plus motherboard. From the rear of the motherboard that's the slot closet to the CPU. While I can't find any documentation suggesting that it is or isn't PCH connected, I doubt it is. This also makes me wonder if I should move the pcie x1 Adnaco card to the PCIe 3.0 x16_2 slot. Feel free to take a look at the documentation on this board and let me know what you think. Any input is appreciated. Thanks.

Larry,

I'm pretty sure your PCIe 3.0 x16 slot bypasses the PCH. I have yet to see a block diagram for a motherboard where this isn't so. I would be interested to know if placing your Adnaco card on your 2nd PCIe x16 slot results in an improvement.

lmitche · March 24, 2017

5 hours ago, romaz said:

Larry,

I'm pretty sure your PCIe 3.0 x16 slot bypasses the PCH. I have yet to see a block diagram for a motherboard where this isn't so. I would be interested to know if placing your Adnaco card on your 2nd PCIe x16 slot results in an improvement.

Roy, the SATA SAGA continues!

Moved the Adnaco to the second PCIE x16 slot putting the nic on a x1 slot. Nic only used for Tidal, so no big deal. Definitely sounding great, female sibilance gone. Had to turn down the volume by 1db, so the image is strong. ATTO disk benchmark runs at 175mbs up from 120mbs.

So far so good. Tomorrow I'll go back to the m.2 SATA port and compare and report.

Things are sounding so good further tweaking just seems futile. That's good as I can't think of any other hardware tweaks to do.

Will start Audiolinux testing with native DSD next week.

romaz · March 25, 2017

28 minutes ago, lmitche said:

Roy, the SATA SAGA continues!

Moved the Adnaco to the second PCIE x16 slot putting the nic on a x1 slot. Nic only used for Tidal, so no big deal. Definitely sounding great, female sibilance gone. Had to turn down the volume by 1db, so the image is strong. ATTO disk benchmark runs at 175mbs up from 120mbs.

So far so good. Tomorrow I'll go back to the m.2 SATA port and compare and report.

Things are sounding so good further tweaking just seems futile. That's good as I can't think of any other hardware tweaks to do.

Will start Audiolinux testing with native DSD next week.

Wow, Larry. This is exciting stuff.

I'm looking forward to knowing what happens when you go back to M.2.

austinpop · March 25, 2017

40 minutes ago, lmitche said:

Definitely sounding great, female sibilance gone.

Larry, anything that reduces female sibilance in your life is a good thing.

Carry on - I have no technical insights - just silly Friday evening humor.

lmitche · March 25, 2017

Rajiv, that's funny. Actually it's more like essing. Sssssssssspanish Harlem.

romaz · March 25, 2017

On 3/13/2017 at 11:58 AM, Superdad said:

Whoa! 4 amps for 16GB of RAM?! That calculator you are using is not to be believed... If anywhere near that much current was drawn by RAM (maybe magnetic core memory from the 1950s?) we would not have laptop computers and the traces on RAM modules would just burn up...

Just sayin'...

It appears this claim of 4 amps of draw for 16GB of RAM just might be accurate after all.

Based on how popular Apacer RAM is in the audiophile community, I decided to order some of their memory to try out. This company is headquartered in Taiwan but I was fortunate to be able to speak with one of their representatives who is based in California and she was kind enough to send me specs on their memory. What is impressive is that for each of the memory modules I was interested in, she was able to provide me a PDF file consisting of 15 pages of specs for each memory module. The spec sheet goes into incredible detail about such things as the pin assignment for all 240 pins of the ECC UDIMMs I am interested in. Anyway, the spec sheet also included current draw and here are some comparisons:

2GB ECC DDR3 SDRAM UDIMM (1.5V)

Active standby current - 279mA

Operating burst read current - 675mA

Operating burst write current - 711mA

Burst refresh current - 1611mA

8GB ECC DDR3 SDRAM UDIMM (1.5V)

Active standby current - 738mA

Operating burst read current - 1728mA

Operating burst write current - 1431mA

Burst refresh current - 2340mA

It's interesting that their 8GB module draws more current during burst reads compared to burst writes. What is impressive is just how much current a single 8GB module draws when idle. 738mA at standby is more than what many SSDs draw when actively reading or writing!

I don't have specs on their 16GB modules but I suppose it's possible it may have a peak draw in excess of 4A. Of course, at 1.5V, that's only about 6w but still, that's a lot of current drawn. I could see how a 16GB RAM drive will have significant advantages with respect to massively superior latency but there certainly will be noise consequences as well. I know in my own testing in the past (where a RAM drive was not utilized), 4GB of RAM definitely sounded better than 8GB of RAM. It's too bad I can't get by with 2GB of RAM like the microRendu or sMS-200.

This is another example of why I struggle to believe that even the most slimmed down music server can compete with something like a microRendu or sMS-200 when it comes to noise or impedance.

greenleo · March 25, 2017

10 minutes ago, romaz said:

It appears this claim of 4 amps of draw for 16GB of RAM just might be accurate after all.

...

This is another example of why I struggle to believe that even the most slimmed down music server can compete with something like a microRendu or sMS-200 when it comes to noise or impedance.

Hi Roy,

I have always been trusting you in the figure of 4A. The point is the V for the DRMA is only 1.2V and the output of 16GB RAM is only 5W--not a lot for a desktop. As a reference, the operating voltage of DDR was 2.5V.

There are something not clear to me and I hope that other members may chime in.

Is the key source of noise from current, voltage or power or ...?

- 4A (in terms of current ) is lot but 5W (in terms of power) is insignificant compare to a 800W PSU. From your figure, the typical noise in voltage for a PC is a few mV. Current, Voltage and power are in different dimension and I find it hard to have uniform approach to assess the importance of them.

- Any approach or method to tell us which will become the important factor under a given situation. I guess you mentioned that 4GB sounds better than 8GB (hence the smaller the size of the RAM, the better) and Alex mentioned that RAM disk as the file disk sounds best (hence the more the RAM, the better).

- Have you compared the SQ of music files play through RAM and SDD/HDD? I am wondering if the SDD is not accessed at all, will the SQ be affected by the SATA cable?

Finally, in your last paragraph, "can compete" should be "can't complete", right?

Best Regards

romaz · March 25, 2017

37 minutes ago, greenleo said:

Is the key source of noise from current, voltage or power or ...?

- 4A (in terms of current ) is lot but 5W (in terms of power) is insignificant compare to a 800W PSU. From your figure, the typical noise in voltage for a PC is a few mV. Current, Voltage and power are in different dimension and I find it hard to have uniform approach to assess the importance of them.

You need to look at noise in relative terms. A few mV is a few thousand uV. When you have an HDPlex ATX LPSU that generates 3-5mV of ripple noise vs an SR7 that generates <4uV, that's a factor of about 1000x more noise.

With respect to what creates noise on a ground plane, it will be a component that draws more current and not more voltage. P (watts) = V (volts) x I (amps). For a component that consumes 12 watts, for example, you would much rather have a 12V component that draws 1A than a 1V component that draws 12A. Even though they both consume 12 watts, the component that draws 12A will generate much more noise. Just look at the noise values of any ATX power supply and you will notice it is always the low voltage rails (especially the 3.3V rail) that have the greatest amount of ripple and that is because for an equal amount of power consumption against a higher voltage rail, a lower voltage rail will consume a proportionately larger amount of current.

It's easy to think that 6 watts isn't much when your PSU is capable of supplying 1,000 watts but when it is a 1.5V device (RAM) that is consuming 6 watts, that means 4A of draw. If this was a 12V device consuming that level of power (like a 3.5" hard drive), that's only 0.5A of draw. There should be no confusion here about which of these components is spewing more noise into the ground plane.

37 minutes ago, greenleo said:

Any approach or method to tell us which will become the important factor under a given situation. I guess you mentioned that 4GB sounds better than 8GB (hence the smaller the size of the RAM, the better) and Alex mentioned that RAM disk as the file disk sounds best (hence the more the RAM, the better).

I am not alone in this assessment and it should come as no surprise why less RAM should sound better than more RAM from a noise standpoint. But what sounds better overall will depend on more than just noise. With more RAM and with faster RAM, you have the potential for less latency and when you incorporate something like a RAM drive that results in less disk I/O activity, it's possible in some systems, the improved latency could offset the increased noise. It's also possible that if your system sounds best when incorporating oversampling with HQPlayer (which benefits from more RAM), the improvement in resolution may be well worth the increased noise. Of course, there are also ways to mitigate noise downstream although it's always best to avoid noise altogether when possible. As always, it is a balance of compromises that we each must make.

38 minutes ago, greenleo said:

I am wondering if the SDD is not accessed at all, will the SQ be affected by the SATA cable?

Yes, absolutely. If the SSD or HD are "not accessed at all", then where does the stream originate? I believe that anything that is directly in the signal path has the potential to imprint on the signal and this imprint can persist all the way to the speakers. WIth a SATA cable, it's not just about the metallurgy that will impact the transmission characteristics of that cable but also the the effectiveness of the shielding that is incorporated, the impedance added by the connectors, solder, dielectric, etc. Basically, everything in a circuit will have a sonic property. While digital is about 1s and 0s, a digital waveform, which can be viewed on an oscilloscope, very much has analog properties with respect to amplitude and rise time and it is my opinion that noise and impedance can impact these properties.

Here is an analogy that I gave on Head-Fi about "bits are bits" and the detriment of latency (although it applies equally well to noise). It's a bit simplistic but it helps me in my own understanding:

"A good analogy would be a pie delivery man who sets out to deliver an apple pie à la Mode to a certain customer on the far end of town. Obviously, with a pie and ice cream that can melt, timing is crucial but on his way to this customer's house, he encounters numerous roadblocks and detours that force him to take several bumpy back roads. Not only does the delay melt the ice cream but the beautiful apple pie is now an apple crumble. Of course, the bits are all still there but there is now no way this delivery man can properly put all those bits back together in a way that exactly resembled the original cake and when the customer takes a bite, while he'll know he's bitten into an apple pie with the essence of ice cream flavor, the enjoyment won't be the same, at least not in the way that the creator of the pie intended. It's not just the ingredients but the quality of the presentation that matters (signal integrity)."

1 hour ago, greenleo said:

Finally, in your last paragraph, "can compete" should be "can't complete", right?

Read it again. This is exactly what I meant.

seeteeyou · March 25, 2017

SSD Reviews
http://www.ssdreview.com/review/solid-state-ssd/

Lowest power consumption SSD?
http://forum.notebookreview.com/threads/lowest-power-consumption-ssd.589021/

SATA Power Management - SATA-IO
https://www.sata-io.org/power-management

StorPortSetPowerSettingNotificationGuids routine
https://msdn.microsoft.com/en-us/library/windows/hardware/hh451513(v=vs.85).aspx

Aggressive Link Power Management (ALPM)
http://www.fit-pc.com/wiki/index.php/Aggressive_Link_Power_Management_(ALPM)

Lowest power consumption SSD
http://www.silentpcreview.com/forums/viewtopic.php?f=7&t=63293

Quote

Most SSD power consumption reviews are completely wrong. Here's the deal: Some SSDs support DIPM, Device Initiated Power Management. This is not an automatic hardware feature, but something that has to be activated by a driver. Most SSD reviewers run automated test benches in either a scripted Linux Live CD (or usb stick) or some other means where they don't install drivers that are compatible with DIPM. This means that those drives that benefit from DIPM are usually not recognized. To add insult to injury, all reviewers without fail don't seem to be able to read the specs and see a huge discrepancy between the quoted power consumption figures and their measurements. Surely, if you see that the Postville Intel SSDs quote a power consumption of 75mW idle/150mW active and you are measuring 0.6W, something must be wrong?. No, they never investigate. Of course I have been an annoying user and wrote to most reviewers that made this error, but they either respond with 'nah, don't care' or 'can't change test bench, won't change reviews'. At the least, remove power consumption figures altogether if they're plainly wrong...

So, here's the low-down:
- Intel-based 5x and 3xnm drives are the thriftiest, with 75mW idle power consumption and on average much less than their quoted 150mW. Peak consumption is somewhere in the 1-2W range.
- Closely followed by 2xnm Intel drives that do 100mW/200mW. Peak consumption is a bit higher still.
- Phison and Samsung based drives mostly idle at 200-250mW (Corsair Nova, Samsung 470)
- Intel Marvell-based drives idle somewhere between 200 and 600mW, but numbers are hard to distill from the many flawed reviews and I haven't had one to test yet. Other Marvell-based drives usually don't support DIPM and idle higher than Intel, although this has changed since C400 (e.g. the Crucial M4).
- Sandforce drives don't support DIPM, idle between 0.6W (SF-12xx) and 1W (SF-2281), also depending on capacity

Do those ALPM registry hacks work for both HDDs and SSDs?

greenleo · March 25, 2017

3 hours ago, romaz said:

You need to look at noise in relative terms. A few mV is a few thousand uV. When you have an HDPlex ATX LPSU that generates 3-5mV of ripple noise vs an SR7 that generates <4uV, that's a factor of about 1000x more noise.

With respect to what creates noise on a ground plane, it will be a component that draws more current and not more voltage. P (watts) = V (volts) x I (amps). For a component that consumes 12 watts, for example, you would much rather have a 12V component that draws 1A than a 1V component that draws 12A. Even though they both consume 12 watts, the component that draws 12A will generate much more noise. Just look at the noise values of any ATX power supply and you will notice it is always the low voltage rails (especially the 3.3V rail) that have the greatest amount of ripple and that is because for an equal amount of power consumption against a higher voltage rail, a lower voltage rail will consume a proportionately larger amount of current.

It's easy to think that 6 watts isn't much when your PSU is capable of supplying 1,000 watts but when it is a 1.5V device (RAM) that is consuming 6 watts, that means 4A of draw. If this was a 12V device consuming that level of power (like a 3.5" hard drive), that's only 0.5A of draw. There should be no confusion here about which of these components is spewing more noise into the ground plane.

I am not alone in this assessment and it should come as no surprise why less RAM should sound better than more RAM from a noise standpoint. But what sounds better overall will depend on more than just noise. With more RAM and with faster RAM, you have the potential for less latency and when you incorporate something like a RAM drive that results in less disk I/O activity, it's possible in some systems, the improved latency could offset the increased noise. It's also possible that if your system sounds best when incorporating oversampling with HQPlayer (which benefits from more RAM), the improvement in resolution may be well worth the increased noise. Of course, there are also ways to mitigate noise downstream although it's always best to avoid noise altogether when possible. As always, it is a balance of compromises that we each must make.

Yes, absolutely. If the SSD or HD are "not accessed at all", then where does the stream originate? I believe that anything that is directly in the signal path has the potential to imprint on the signal and this imprint can persist all the way to the speakers. WIth a SATA cable, it's not just about the metallurgy that will impact the transmission characteristics of that cable but also the the effectiveness of the shielding that is incorporated, the impedance added by the connectors, solder, dielectric, etc. Basically, everything in a circuit will have a sonic property. While digital is about 1s and 0s, a digital waveform, which can be viewed on an oscilloscope, very much has analog properties with respect to amplitude and rise time and it is my opinion that noise and impedance can impact these properties.

Here is an analogy that I gave on Head-Fi about "bits are bits" and the detriment of latency (although it applies equally well to noise). It's a bit simplistic but it helps me in my own understanding:

"A good analogy would be a pie delivery man who sets out to deliver an apple pie à la Mode to a certain customer on the far end of town. Obviously, with a pie and ice cream that can melt, timing is crucial but on his way to this customer's house, he encounters numerous roadblocks and detours that force him to take several bumpy back roads. Not only does the delay melt the ice cream but the beautiful apple pie is now an apple crumble. Of course, the bits are all still there but there is now no way this delivery man can properly put all those bits back together in a way that exactly resembled the original cake and when the customer takes a bite, while he'll know he's bitten into an apple pie with the essence of ice cream flavor, the enjoyment won't be the same, at least not in the way that the creator of the pie intended. It's not just the ingredients but the quality of the presentation that matters (signal integrity)."

Read it again. This is exactly what I meant.

Thank you Roy.

Now I see the key rationale is to reduce the noise in the ground plane. Hence smaller current is more beneficial. The balancing of the optimal amount of RAM is pretty intricate. I saw members who stated that OS in RAM sounds best which means more RAM is needed. With the rationale above, probably the benefit of Imitche's approach of using 1 HDD only is explained--even though the conventional wisdom was to use one for OS and the other for the music files. I guess two HDD means two SATA cables means picking up double portion of noise.

I find the learning curve of CAS very steep and a lot are still experimenting. Thank you again for the explanation.

Finally, I miss the word "struggle" in my initial reading and hence the misunderstanding. In this case, Would an Android tablet sound the best? Just a wild guess. Is it worthwhile to built an single purpose audio tablet? I am not joking but am really wondering.

lmitche · March 25, 2017

FWIW - I run my two 4gb sticks of DDR4 in single channel mode. It definitely sounds better than a single stick or two sticks in dual channel mode.

Memory usage never tops 1.7gb here.

seeteeyou · March 25, 2017

Android tablets sounded interesting, though Roon Server must be running off something with x86 processors.

THE most energy efficient x86 devices out there with a reasonably powerful processor should look pretty much like Intel Compute Stick

http://www.intel.com/content/www/us/en/compute-stick/intel-compute-stick.html

Another one with Pentium N4200

http://www.up-board.org/upsquared/

Less powerful ones (when compared to Core M processors) with Atom

http://wiki.minnowboard.org/MinnowBoard_MAX

https://buildazure.com/2016/08/30/intel-joule-iot-is-an-iot-powerhouse/

austinpop · March 25, 2017

@romaz

Fascinating, as always, but just so I'm not losing context - you think these things matter so much even when we're talking about a music server connected to an endpoint like sMS-200, rather than a music player directly feeding a DAC?

Just wondering if some of your past experiences - like 4gb vs 8gb - were in the context of a music player, whereas the music server is "once removed" from this, no?

Or is that what you want to revisit with your new build?

I think I lost the thread in these arcane minutiae of PC architecture.

SwissBear · March 25, 2017

7 hours ago, austinpop said:

@romaz

Fascinating, as always, but just so I'm not losing context - you think these things matter so much even when we're talking about a music server connected to an endpoint like sMS-200, rather than a music player directly feeding a DAC?

I would be interested to have this confirmation too.

Regarding I/O noise, it will also be interesting to check if Ryzen, the new AMD processor, which offers an integrated I/O hub, including USB ports as well as SATA3 ports directly connected to the CPU, as well as a reduced chipset (X300) to come, will be able to offer reduced noise for its I/O.

See: https://www.guru3d.com/articles_pages/amd_ryzen_7_1700_review,14.html

romaz · March 26, 2017

13 hours ago, seeteeyou said:

SSD Reviews
http://www.ssdreview.com/review/solid-state-ssd/

Lowest power consumption SSD?
http://forum.notebookreview.com/threads/lowest-power-consumption-ssd.589021/

SATA Power Management - SATA-IO
https://www.sata-io.org/power-management

StorPortSetPowerSettingNotificationGuids routine
https://msdn.microsoft.com/en-us/library/windows/hardware/hh451513(v=vs.85).aspx

Aggressive Link Power Management (ALPM)
http://www.fit-pc.com/wiki/index.php/Aggressive_Link_Power_Management_(ALPM)

Lowest power consumption SSD
http://www.silentpcreview.com/forums/viewtopic.php?f=7&t=63293

Do those ALPM registry hacks work for both HDDs and SSDs?

Thanks, you always provide great links. I had seen some of these already.

As a rule, during music playback, most believe any form of power management that saves power is a bad thing and so I am not sure that I would be in favor of ALPM. During music playback, you want your system in a ready and responsive state at all times and not having to awaken from some low power state at inopportune times since that will represent a form of latency. In this regard, you want a storage medium that doesn't draw much current during random reads or writes if you are talking about an OS drive or during sequential reads if you are talking about a music storage drive. Low current draw when in an active idle state is also good since you don't want your storage drive still drawing large amounts of current once your music track has already been buffered into memory. This is another reason why I believe the OS drive makes a larger impact on SQ than a storage drive because unlike a music storage drive that can afford to go idle once a track is buffered into memory, the OS drive is generally active all the time.

I think the general consensus is that speed and power = noise and slower and less powerful is generally less noisy. Others have reported that SATA II sounds better than SATA III and so the best sounding SSDs would seem to be the ones that are no longer in production. 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. This Intel X25E is also a SATA II drive and so its typical current draw when idle is a minuscule 8mA and during routine "workstation type I/O" is only 220mA. These numbers represent less draw than an Apacer SLC compact flash drive.

My thinking is that this Intel X25-E has a shot at sounding better than compact flash not only because it draws less current but also because is has much lower latency. Very soon, I will be making the following comparisons as OS drives: Intel X25-E 64GB SLC SATA II SSD vs 32GB compact flash type 1 vs Toshiba 1GB 5400 SATA II hard drive vs Samsung 850 EVO 2TB SATA III SSD. I also have a Samsung 850 Pro 4TB SATA III SSD but there's no point comparing that one as it is sure to be the noisiest. Each drive will connect to my SATA-to-PCIE adapter card via a Pachanko Reference SATA cable and powered by an LPS-1.

One thing I am indeed curious about is the possibility of using the Toshiba 5400 SATA II hard drive as both an OS and music storage drive similar to what @lmitche does. Only hard drives have the option of being large enough and yet electrically quiet enough to serve as a both an OS drive and music storage drive. While Gordon Rankin has suggested that using one drive for both OS and music storage will result in playback latency issues resulting in more errors, with Roon, because tracks are buffered into memory, I am not anticipating too much downside using one drive as both an OS drive and music storage drive. It turns out there are still quite a few 2.5 inch SATA II hard drives in production today and this particular model consumes only 300mA during active reads and writes, lower than even a WD Green or Blue.

romaz · March 26, 2017

12 hours ago, lmitche said:

FWIW - I run my two 4gb sticks of DDR4 in single channel mode. It definitely sounds better than a single stick or two sticks in dual channel mode.

Memory usage never tops 1.7gb here.

Thanks for sharing this, Larry. I have actually ordered two 2GB sticks because I wasn't sure if single channel or dual channel would sound better and so you have now removed the guesswork for me.

romaz · March 26, 2017

10 hours ago, austinpop said:

@romaz

Fascinating, as always, but just so I'm not losing context - you think these things matter so much even when we're talking about a music server connected to an endpoint like sMS-200, rather than a music player directly feeding a DAC?

I think some of these questions still need to be answered. With my current 2014 Mac Mini, which utilizes Apple's least powerful configuration (i5 at 1.4GHz with a TDP of 15w, 4GB of DDR3, Wifi removed, GPU disabled, all SSDs and hard drives removed, OS off SD card), at idle, this machine consumes under 5 watts but when active, it can consume up to 15 watts. It's hard to do better than this and yet compared against my mR or stock sMS-200 (when powered by my SR7) and using the direct connection, using either endpoint still sounds better (way better).

With my new build which will have as many of the bad clocks in the signal path removed as possible, my inclination is that my sMS-200 Ultra especially will still sound better largely because I am forced to use this HDPlex DC-ATX converter which will significantly handicap my SR7 in terms of noise and impedance. While this server will be necessary to do the heavy lifting, I just can't imagine it can ever be as quiet or quick and nimble as either the mR or sMS-200. At least, that's my hunch.

romaz · March 26, 2017

3 hours ago, SwissBear said:

Regarding I/O noise, it will also be interesting to check if Ryzen, the new AMD processor, which offers an integrated I/O hub, including USB ports as well as SATA3 ports directly connected to the CPU, as well as a reduced chipset (X300) to come, will be able to offer reduced noise for its I/O.

See: https://www.guru3d.com/articles_pages/amd_ryzen_7_1700_review,14.html

This AMD Ryzen is interesting for some of the reasons you mentioned but all the Ryzens I have read about seem to be geared for power use with 8-cores and overclocking in mind. At present, their lowest power CPU, the 1700, has a TDP of 65 watts. If they come out with a "mobile" version that consumes less than 10w that incorporates a large cache, I think that would raise an eyebrow for sure but this doesn't seem to be where they want to go. Their mantra is "Ryzen Powers. You Dominate." That is opposite of what I am trying to accomplish.

romaz · March 26, 2017

11 hours ago, austinpop said:

@romaz

Fascinating, as always, but just so I'm not losing context - you think these things matter so much even when we're talking about a music server connected to an endpoint like sMS-200, rather than a music player directly feeding a DAC?

Sorry, I believe I didn't read your post properly and so I failed to answer your question. With the standard way of connecting an mR or sMS-200 to a router first, I don't think these things matter as much. With the direct connection, while transparency to the recording has increased, so has transparency to the qualities (and deficiencies) of the music server itself. As long as I continue to hear improvements, I will probably continue to push along.

SwissBear · March 26, 2017

6 hours ago, romaz said:

This AMD Ryzen is interesting for some of the reasons you mentioned but all the Ryzens I have read about seem to be geared for power use with 8-cores and overclocking in mind. At present, their lowest power CPU, the 1700, has a TDP of 65 watts. If they come out with a "mobile" version that consumes less than 10w that incorporates a large cache, I think that would raise an eyebrow for sure but this doesn't seem to be where they want to go. Their mantra is "Ryzen Powers. You Dominate." That is opposite of what I am trying to accomplish.

Hi Romaz,

Thank you for your answer. I have probably misunderstood you goals. I thought you were up to a mix between performance and quietness, in the same way as SGM with their SGM 2015. Thanks for the precision.

lmitche · March 26, 2017

10 hours ago, romaz said:

Thanks for sharing this, Larry. I have actually ordered two 2GB sticks because I wasn't sure if single channel or dual channel would sound better and so you have now removed the guesswork for me.

Hmmm, I can't find any 2gb DDR4 memory availability, Guessing you ordered ddr3?

greenleo · March 26, 2017

10 hours ago, romaz said:

...

This is another reason why I believe the OS drive makes a larger impact on SQ than a storage drive because unlike a music storage drive that can afford to go idle once a track is buffered into memory, the OS drive is generally active all the time.

I think the general consensus is that speed and power = noise and slower and less powerful is generally less noisy. Others have reported that SATA II sounds better than SATA III and so the best sounding SSDs would seem to be the ones that are no longer in production. 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. This Intel X25E is also a SATA II drive and so its typical current draw when idle is a minuscule 8mA and during routine "workstation type I/O" is only 220mA. These numbers represent less draw than an Apacer SLC compact flash drive.

My thinking is that this Intel X25-E has a shot at sounding better than compact flash not only because it draws less current but also because is has much lower latency. Very soon, I will be making the following comparisons as OS drives: Intel X25-E 64GB SLC SATA II SSD vs 32GB compact flash type 1 vs Toshiba 1GB 5400 SATA II hard drive vs Samsung 850 EVO 2TB SATA III SSD. I also have a Samsung 850 Pro 4TB SATA III SSD but there's no point comparing that one as it is sure to be the noisiest. Each drive will connect to my SATA-to-PCIE adapter card via a Pachanko Reference SATA cable and powered by an LPS-1.

Hi Roy,

Any reasons that an OS drive is always busy in Windows? Given that an audio PC just runs a few processes, after a short time, all the files to be run or be read would already been cached in the RAM, would it? Also, by disabling the page file, there is no need in reading OS disk again. The OS disk should not be busy.

Not sure if Windows needs to write something secretly into OS disk. If not, then the OS disk activity may be reduced to to zero and the influence of the OS disk may be reduced to minimal.

This is a question that has puzzled me for a long time.

Would you mind to tell us were to get the Intel X25E SSD?

lmitche · March 26, 2017

"I think the general consensus is that speed and power = noise and slower and less powerful is generally less noisy." - Roy

"You think these things matter so much even when we're talking about a music server connected to an endpoint like sMS-200, rather than a music player directly feeding a DAC?" - Rajiv

Aren't these two questions related? First off we are all using two box designs with one box doing, heavy (hqplayer), or heavier lifting (SMBD server), the second converting from ethernet to usb, or fiber to usb each powered by an LPS-1. My fiber to USB box uses very little power 5V DC@ 0.35A maximum for the R1USB30 itself. So, I think we are really talking about the same music sever - endpoint, high power - low power architecture,

With galvanic isolation in place, through either fiber or ethernet, I don't think that the power usage of an isolated server machine makes any difference, but that the cleanliness of the power does.

To me the mystery is why dicking around with something like the sata connection has such a huge impact. I don't get it and can only guess we are exposing the impact of timings differences between various connection points by varying this parameter. What do you think?

lmitche · March 26, 2017

28 minutes ago, greenleo said:

Hi Roy,

Any reasons that an OS drive is always busy in Windows? Given that an audio PC just runs a few processes, after a short time, all the files to be run or be read would already been cached in the RAM, would it? Also, by disabling the page file, there is no need in reading OS disk again. The OS disk should not be busy.

Not sure if Windows needs to write something secretly into OS disk. If not, then the OS disk activity may be reduced to to zero and the influence of the OS disk may be reduced to minimal.

This is a question that has puzzled me for a long time.

Would you mind to tell us were to get the Intel X25E SSD?

GreenLeo,

I have spent hours chasing down random disk writes in Windows 10, finding the source, and researching a way to kill the processes associated with this random activity. There are logging processes all over Windows. In the Windows 10 optimization thread you will find various scripts that many of us have contributed to prevent this random disk writing behavior.

Left on it's own, WIndows makes promiscuous use of the hard disk and will write many megabytes of data per second despite zero user activity and a disabled swap disk. Just run the resource monitor available of the performance tab of the task manager. Once in the resource monitor watch the disk activity tab.

Bring a barf bag.

Despite my efforts, I still have 10s of thousand of bytes of logs flowing every second. Very annoying.

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

Recommended Posts

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Create an account or sign in to comment

Create an account

Sign in