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

[flkin]

Thank you for your kind words @romaz and for your very interesting thoughts. It's not easy to find high end servers within reach so it was important to get the data point out and share. I would love to have a chance to listen to the new forthcoming SGM Evo when it does come out or Adrian's One server too - if they ever come to Thailand.

 

On the matter of convolution, you rightly mention that there are pros and cons to the approach which I may have glossed over. I would like to add that I believe generating convolution files is a bit of an art form and is not a technical solution. My final room corrections are slight and the overall room response curves after adjustment no where close to flat. But I think this is a correct approach. Remove the sound of the room from the signal is the target not removing the character of the system. 

 

Thierry from Home Audio Fidelity whose service I used, offered me some room correction files that attempted to remove cross talk between the right and left speakers. It is apparently possible to reduce the impact of each speaker into the opposite ear. Ultimately I found that the sound wasn't natural and duller and decided against using it. But it shows there are many different approaches to convolution and it may take some back and forth with the "convolution artist" to get it right.

 

On the matter of the Fink Faun 2.16x heat management, it runs hot for sure as the ambient temp in Asia is quite high (CPU temperature as per Bios is 77C, motherboard at 50C ). But the temperature is roughly the same whether the unit is playing or at rest (but still on). It's due to the idea of the CPU running full constantly to stop cores running at different frequencies that may cause latency issues. I'm planning to add a USB fan on top of the streamer to increase air flow when I am not using it to reduce the standing temperature. And turn it off when I play music.

 

Oh and there are 3 transformers in the 2.16x not 2. Two of them are wired in parallel and used for audio path devices. The third one is for non-audio path uses.

 

[FredM]

@flkin and @romaz thank you both for the excellent posts. I enjoy reading your journeys and really appreciate that you share your experience with these comprehensive write ups!

 

Roy, you mentioned a new PinkFaun streamer. Recently at the Xfi event they introduced the Scion streamer:

 

 

[seeteeyou]

1 hour ago, romaz said:

Moreover, the power supply has an energy storage capacitance of >800,000uF.  This is more storage capacitance than my Soulution amplifier!  There was a time when I thought something like this was overkill but not any longer.  There's a reason why a Paul Hynes SR7 sounds better than an SR4 for a tiny component like a tX-USBultra that barely draws a few watts and according to Paul, it has to do with having plenty of overhead so that the supply is always operating within its optimum power band.

 

Overkilled PSUs worked so well for these audiophiles in Taiwan to the point where they just powered a single 5V or 12V rail with a 1500W industrial SMPS (with relatively loud cooling fans) like this, Mean Well used to be the winner but recently Cosel turned out to take the cake

 

https://en.cosel.co.jp/product/powersupply/PBA/

http://www.stsd99.com/phpBB3/viewtopic.php?f=61&t=2674

http://www.my-hiend.com/vbb/showthread.php?5404-走向超值而極緻的-HI-FI電腦訊源（連載）/

 

Used ones are very affordable these days so it wouldn't hurt too much to give them a try, some of them really liked the (unmodified) Sony UBP-X700 + Cosel PBA Series combo.

[marce]

Except SMPS's don't work like linear when providing low power, they can go into current saving mode and create noise in the audible frequency range.

[lmitche]

8 hours ago, romaz said:

@flkin, thank you for taking the time to report your findings in great detail.  Your data points, especially against the Antipodes CX+EX, are of great interest to me personally.  Ultimately, it is how one component compares to another component that I find the most valuable perspective and you did that very well!  

 

Based on your initial favorable comments about the Pink Faun 2.16 some months ago,  I had phoned Pink Faun in the Netherlands to try and better understand their philosophy.  I want to say that I spoke with Jord but I can't remember for sure.  Regardless, we spoke for nearly 1.5 hours and I came away intrigued by their approach.  Here were my main takeaways:

 

1.  Because upsampling with HQP is a key component to their 2.16 server, they believe that more power (and not less power) is better because "faster" = lower latency.  This appears to be Sound Gallerie's approach with their SGM 2015 (and now, their new SGM Evo) and so if a comparison is to be made, ideally, it should be against the Evo.  Having heard how good the SGM 2015 sounds, I think they have made a good case for this "more is more" approach.

 

2.  Unlike everyone else, Pink Faun feels AMD sounds better than Intel which is why they went with the specific 8-core AMD CPU, chipset and board that they did.  Based on your own favorable comparative assessments of the 2.16, it would appear that their decision to go in this direction has been well founded.

 

3.  The 2.16 can incorporate up to 6 independently powered OCXOs depending on the buyer's preference (CPU, chipset, LAN, USB, SPDIF, I2S, etc).  Wow.

 

4.  The 2.16 has the potential for up to 20 linear rails of power (depending on the needs of the user) coming off of 2 large transformers.  Moreover, the power supply has an energy storage capacitance of >800,000uF.  This is more storage capacitance than my Soulution amplifier!  There was a time when I thought something like this was overkill but not any longer.  There's a reason why a Paul Hynes SR7 sounds better than an SR4 for a tiny component like a tX-USBultra that barely draws a few watts and according to Paul, it has to do with having plenty of overhead so that the supply is always operating within its optimum power band.  Unless it's a typo, it looks like the new Sound Galleries Evo will have a storage capacitance of 3.76 million uF.  Let the server PSU wars begin.

 

5.  The 2.16 consumes 100 watts at idle and runs fairly warm and so it requires a well ventilated space.  Because the OCXOs need to be kept heated to sound their best and because it can take a long time to heat up the OCXOs, the 2.16 should ideally be kept constantly running.  I don't think the 2.16 is going to win any "eco" awards.  This is a legitimate concern and probably the biggest downside of the 2.16 for me since my equipment is stored in a cabinet.

 

6.   Pink Faun is presently testing their own low-power mini streamer based off an ASUS Tinker Board which is even smaller (and potentially lower latency) than a NUC.  This streamer will have both USB and SPDIF output capability and can be powered by a 5V/2A supply.  Color me intrigued on this device.  At the time of my conversation with them, there was no word yet on how good this sounded.

 

Now, taking into account your own findings, it would appear the 2.16 is the best server you have heard/owned to date and you certainly have put it against some tough competition.  Here are a few of my own recent findings and comments and so perhaps, I can add to the body of information that could help the community at large.

 

@lmitche was kind enough to configure a low power NUC for me with his optimized version of headless AudioLinux running on it (thank you, Larry!).  The model I purchased (NUC7CJYH) cost me $120 and incorporates my preferred J4005 2-core Celeron with a 10w TDP and a 4MB L2 cache (2MB of cache per core).  My previous NUC had only 1MB of cache per core.  This particular model has no eMMC drive and so once AudioLinux boots into RAM via a USB stick, Larry configured it so that I can remove the USB stick and so this is a completely diskless NUC endpoint running RoonBridge.  Where my previous NUC's OS consumed 5.8GB of memory, this more compacted version of AudioLinux consumes <3GB of memory.  Most would assume that this NUC should sound better than my NUC and indeed, that is correct.  The improvement is considerable and so kudos to Larry.

 

During this time, I also had access to SOtM's latest sMS-200ultra Neo.  When I last owned an sMS-200ultra, it was before I owned the Ref10 and so I felt it was important to revisit the latest incarnation of the sMS-200ultra to see how it would compare to the NUC.  As a similarly small device designed from the ground up for audio, I always felt these types of devices should rule the world.  WIth the sMS-200ultra Neo connected via Habst cable to my Ref10, even though I no longer had an original sMS-200ultra on hand for direct comparison, I very much liked what I heard and it was indeed better than what I recall my sMS-200ultra sounded like.  Where I felt the original sMS-200ultra sounded incisive, airier, and more detailed compared against my microRendu, there was a beguiling smoothness and naturalness to the microRendu that was also very pleasant.  What is interesting is that the ultraRendu has gone more in the direction of the sMS-200ultra and in some ways, sounds even more detailed and more mechanical whereas the new Neo has gone more in the direction of the original microRendu as it has given up some of its incisiveness for a harmonically richer and more natural sounding presentation.  In isolation and when powered by my SR7, my immediate inclination is that the new Neo is a winner and it definitely is.  However, in comparison, I still found my NUC to be better and not by a small margin.  If I were to provide 2 descriptors that sum up what this NUC offers, especially when powered by a DR SR7 rail, it is "presence" and "energy."  It was this same presence and energy that drew me to the Innuos Zenith SE and now draws me even further with this NUC.

 

But here is where things get interesting.  As I previously reported, I found my best SQ when this NUC was used purely as a Roon endpoint (renderer) while server duty was taken over by a separate server.  While this NUC was capable of full server and endpoint duty and while playback occurred smoothly, SQ sounded strained.  While my Zenith SE filled the bill very nicely as a RoonServer, when I placed SOtM's new sNH-10G network switch between my RoonServer and Roon Endpoint, it no longer seemed to matter what I used as a RoonServer.  In other words, my unoptimized Mac Pro with its noisy 12-core Xeon, 64GB of RAM, dual GPU, and 1TB of PCIe SSD storage sounded just as good.  If there was a difference, it was too small to worry about it.

 

WIth my 2nd NUC in hand, I now had the luxury of trying my original NUC as the RoonServer.  Even though I couldn't hear much difference between Zenith SE and Mac Pro as RoonServer, would this NUC running AudioLinux in RAM result in an improvement?  Of course, with only 8GB of memory, there isn't much room for any kind of a Roon database and so I configured this NUC for Tidal streaming only through Roon.  Even with the lower latency you get by running your OS completely in memory, there are definitely latency penalties by using a Celeron vs a 12-core Xeon when it comes to all the tasks that RoonServer is responsible for.  For example, browsing through my Tidal library occurred much more slowly through the NUC.  Nonetheless, it is SQ that matters and here is what I found.

 

With my original NUC powered by a 12V rail from my SR7 and functioning as RoonServer and with my new "Larry" NUC powered by a DR 12V rail from my SR7 and functioning as RoonBridge, transients had better snap and seemed faster.  However, with my Mac Pro functioning as RoonServer, there was better dimensionality with richer, deeper and more natural tone.  In this instance, the Mac Pro was sounding better overall than my NUC as RoonServer.  How could this be?

 

A couple of thoughts...

 

1.  I no longer have SOtM's sNH-10G switch in my possession as that switch is now with Rajiv.  While SOtM thought they would be shipping this switch by now, they are having difficulty procuring enough chassis and so it remains unclear when their switch will be available and so I am presently using TLS's very fine switch but I don't think this switch isolates as well as SOtM's switch and so this may be accounting for some of the differences I am hearing.

 

2.  When it comes to the heavy lifting required by RoonServer, I believe Pink Faun's philosophy is the correct one.  More power = faster = lower latency.  I continue to argue that with a "one trick pony" RoonBridge device, blazing speed and power are unnecessary and even unwanted and so I am very intrigued to see what Pink Faun's new Tinker board mini streamer will sound like but when it comes to RoonServer and especially if the user plans to upsample with HQP or employ DSP room correction, the more power the better.  The challenge with a high power RoonServer is they are more difficult to power well (just look at what went into the 2.16) and so there will be cost, energy, and heat considerations to take into account.

 

I believe there is growing consensus from the likes of Antipodes and Pink Faun that for ultimate SQ, you need to distribute tasks between multiple machines.  In my current situation (without SOtM's switch), it would seem that both server and renderer matter and that attention needs to be paid to both which is an unfortunate bursting of my bubble, however, I continue to strongly believe that the renderer has the greater overall impact.  If I were to assign a level of importance, in the absence of upsampling or convolution, I would say the renderer has at least 80% impact while the server has at most 20% impact.  This bodes well because a good renderer costs a lot less than a good server and so I am hoping this observation stands the test of time.

 

For those who plan to upsample or employ DSP-based room correction, I feel you have less choice.  If SQ is your top priority and if finances permit, then it would seem that the 2.16 should be on your short list but then so should Sound Galleries' offerings.  Otherwise, you might just find this cheap AudioLinux-based NUC to be the worldbeater that the audiophile community has been waiting for.

 

A final word on convolution.  I have explored this with AudioVero's Acourate convolution software and I expect to explore it further because I do feel it has merit but there are consequences.  With the help of Uli Brueggeman from Germany, I was able to map my room and at various seating positions, we were able to determine my room's frequency response and where my nodes are.  He created idealized convolution profiles for me that I plugged into HQP and while I was able to effectively flatten my frequency response, it seemed to occur at the expense of harmonic richness and depth.  In other words, altering the audible frequency response of my room had an impact on the harmonic frequencies that was akin to trading accuracy for emotional impact.  The biggest negative was the soundstage became flatter.  The good news is that Uli is capable of creating graded profiles and so it's possible I will come across a profile I like.  The reason I bring this up is to highlight the pros and cons I have found with room correction.  If you're going to do it, definitely do it in the digital domain because with analog EQ, there will be resolution penalties but even in the digital domain, there are challenges.  If you employ it at the server level, then you will require a powerful server.  If you buy a device specifically designed for digital room correction like a Lyngdorf, DEQX, or Kii Three, then you're stuck with the DACs that are built in to those devices and thus far, I have not liked what I have heard.  If at all possible, I feel it's better to optimize the room rather than to digitally shape the sound but I also realize that this isn't always possible.  

 

Hi Roy,

 

Your observations about the dirty side 12 core Xeon  Mac vs. 2 core NUC with Celeron j3455 was a reminder that the Celeron line is usually de-featured compared to Pentium and Core CPU lines. Below are the features left out of the j3455. It seems logical that lack of Speed Shift and turbo mode makes the j3455 more stable from a latency standpoint and along with Audiolinux  translates into the fantastic SQ we are hearing. I have added the j4005 missing features as well.

 

These same features in a Core or Xeon CPU may enhance SQ from Roonserver, which is designed to run in a processor built to support server functions.

 

In the next few days I will setup my core I7 upsampling machine to compare SQ vs. the J4005 server to validate your findings. As always we only have our observations as a guide.

 

I'd like to compare an i7 Quad core NUC as well.

 

Best,

 

Larry

 

----------------------------

From Intel Ark j4005 specs:

Advanced Technologies

• Intel® Optane™ Memory Supported ^‡No
• Intel® Speed Shift TechnologyNo
• Intel® Turbo Boost Max Technology 3.0 ^‡No
• Intel® Turbo Boost Technology ^‡No
• Intel® vPro™ Platform Eligibility ^‡No
• Intel® Hyper-Threading Technology ^‡No

From Intel Ark j3455 specs

Advanced Technologies

• Intel® Optane™ Memory Supported ^‡No
• Intel® Turbo Boost Technology ^‡No
• Intel® vPro™ Platform Eligibility ^‡No
• Intel® Hyper-Threading Technology ^‡No

[rickca]

8 hours ago, FredM said:

Roy, you mentioned a new PinkFaun streamer. Recently at the Xfi event they introduced the Scion streamer

Looks like there is no USB output on the Scion, just coax, AES/EBU and optical.  I've sent a request to Pink Faun for technical details and availability/pricing info.

[auricgoldfinger]

Shipping goods from China is about to get (much) more expensive (think Ghent Audio and eBay parts and components).

 

https://www.nytimes.com/2018/10/17/us/politics/trump-china-shipping.html

 

[Confused]

18 minutes ago, auricgoldfinger said:

Shipping goods from China is about to get (much) more expensive (think Ghent Audio and eBay parts and components).

 

https://www.nytimes.com/2018/10/17/us/politics/trump-china-shipping.html

 

Not for Brexit Brits. ?

[auricgoldfinger]

2 minutes ago, Confused said:

Not for Brexit Brits. ?

 

Given the global nature of CA, I should have been more precise.  ☺️

[Superdad]

1 hour ago, auricgoldfinger said:

Shipping goods from China is about to get (much) more expensive (think Ghent Audio and eBay parts and components).

 

https://www.nytimes.com/2018/10/17/us/politics/trump-china-shipping.html

 

 

As much as I know that Trump's trade-war policies are wrong-headed and bad for all parties, on this particular issue (the Universal Postal Union treaty) I agree that it should be overhauled.  A package of up to 4.4 pounds can be mailed from China for $5 (and 1 pound or less is about $1.50 converted), yet even with my volume discounted commercial rate a 3 ounce package from California to anyplace overseas costs us $13.40 to mail via First Class Mail.  Oh, and 4 pound package to Asia? Over $50!  

 

Recently I have negotiated us onto a program for amazing rates via FedEx (been automatically upgrading overseas orders to FedEx at no extra charge), but what is troubling us now is that the flow of goods through the Chinese customs offices has slowed WAY down.  We receive a fair number of orders from China each month and lately even FedEx delivery is taking 12 days to deliver.  

 

So far the tariff issue has not impacted my business in either direction, but I know it will soon.  

[auricgoldfinger]

8 minutes ago, Superdad said:

what is troubling us now is that the flow of goods through the Chinese customs offices has slowed WAY down.

 

I have been seeing anecdotal reports that Chinese customs officials are holding up shipments from the U.S., so you are not alone.

[romaz]

18 hours ago, flkin said:

Thank you for your kind words @romaz and for your very interesting thoughts. It's not easy to find high end servers within reach so it was important to get the data point out and share. I would love to have a chance to listen to the new forthcoming SGM Evo when it does come out or Adrian's One server too - if they ever come to Thailand.

 

On the matter of convolution, you rightly mention that there are pros and cons to the approach which I may have glossed over. I would like to add that I believe generating convolution files is a bit of an art form and is not a technical solution. My final room corrections are slight and the overall room response curves after adjustment no where close to flat. But I think this is a correct approach. Remove the sound of the room from the signal is the target not removing the character of the system. 

 

Thierry from Home Audio Fidelity whose service I used, offered me some room correction files that attempted to remove cross talk between the right and left speakers. It is apparently possible to reduce the impact of each speaker into the opposite ear. Ultimately I found that the sound wasn't natural and duller and decided against using it. But it shows there are many different approaches to convolution and it may take some back and forth with the "convolution artist" to get it right.

 

On the matter of the Fink Faun 2.16x heat management, it runs hot for sure as the ambient temp in Asia is quite high (CPU temperature as per Bios is 77C, motherboard at 50C ). But the temperature is roughly the same whether the unit is playing or at rest (but still on). It's due to the idea of the CPU running full constantly to stop cores running at different frequencies that may cause latency issues. I'm planning to add a USB fan on top of the streamer to increase air flow when I am not using it to reduce the standing temperature. And turn it off when I play music.

 

Oh and there are 3 transformers in the 2.16x not 2. Two of them are wired in parallel and used for audio path devices. The third one is for non-audio path uses.

 

 

Lots of advancement in the digital server front this year which makes it painful for some of us "cutting edge" early adopters since the best server of today will likely be a solid mediocre choice in a year's time.  I know this lesson far too well.  Nonetheless, direct comparisons of upper echelon servers happen very rarely and so your contributions are much appreciated.

 

Regarding DSP room correction, you said it well.  It is a blend of science and art to do it properly.  I enjoy visiting different venues to listen to music because each venue, whether it be Carnegie Hall, Alice Tully Hall, or the Sydney Opera House, has its unique acoustical charms.  Never would I desire to attend a concert in an anechoic chamber.  Having said that, I believe DSP room correction has its place and it would appear that you are approaching it as I would -- very judiciously.  

[romaz]

18 hours ago, FredM said:

@flkin and @romaz thank you both for the excellent posts. I enjoy reading your journeys and really appreciate that you share your experience with these comprehensive write ups!

 

Roy, you mentioned a new PinkFaun streamer. Recently at the Xfi event they introduced the Scion streamer:

 

 

 

The Scion could be it, Fred.  I can't read Dutch but it looks like they are using supercaps in their power supply.

[diecaster]

7 hours ago, Superdad said:

 

As much as I know that Trump's trade-war policies are wrong-headed and bad for all parties....

 

Whoa, there. You may not understand this, but his policies are intended to make trade more fair. In every case, he put tariffs in place to combat tariffs placed on us or to combat some other unfair trade practice. China is a very aggressive trading partner that is being unfair on several fronts. We should not be afraid of near term pain if we get long term relief. 

[romaz]

11 hours ago, lmitche said:

Hi Roy,

 

Your observations about the dirty side 12 core Xeon  Mac vs. 2 core NUC with Celeron j3455 was a reminder that the Celeron line is usually de-featured compared to Pentium and Core CPU lines. Below are the features left out of the j3455. It seems logical that lack of Speed Shift and turbo mode makes the j3455 more stable from a latency standpoint and along with Audiolinux  translates into the fantastic SQ we are hearing. I have added the j4005 missing features as well.

 

These same features in a Core or Xeon CPU may enhance SQ from Roonserver, which is designed to run in a processor built to support server functions.

 

In the next few days I will setup my core I7 upsampling machine to compare SQ vs. the J4005 server to validate your findings. As always we only have our observations as a guide.

 

I'd like to compare an i7 Quad core NUC as well.

 

Best,

 

Larry

 

----------------------------

From Intel Ark j4005 specs:

Advanced Technologies

• Intel® Optane™ Memory Supported ^‡No
• Intel® Speed Shift TechnologyNo
• Intel® Turbo Boost Max Technology 3.0 ^‡No
• Intel® Turbo Boost Technology ^‡No
• Intel® vPro™ Platform Eligibility ^‡No
• Intel® Hyper-Threading Technology ^‡No

From Intel Ark j3455 specs

Advanced Technologies

• Intel® Optane™ Memory Supported ^‡No
• Intel® Turbo Boost Technology ^‡No
• Intel® vPro™ Platform Eligibility ^‡No
• Intel® Hyper-Threading Technology ^‡No

 

Your points about the varying architectures of different CPUs are well taken, Larry.  I suspect certain CPUs are better choices for servers that need to do heavy lifting and multitasking while other CPUs are better choices for single-minded renderers.

 

What is interesting is that the very best servers/renderers I have heard so far are based on X86 CPUs and not ARM-based CPUs.  Is it because X86 CPUs are inherently lower latency due to their design and perhaps larger memory caches?

[romaz]

10 hours ago, rickca said:

Looks like there is no USB output on the Scion, just coax, AES/EBU and optical.  I've sent a request to Pink Faun for technical details and availability/pricing info.

 

When I spoke with Pink Faun, they told me they would be using ASUS Tinker Boards for their new streamers.  Here are technical specs on these boards:

 

https://www.asus.com/us/Single-Board-Computer/Tinker-Board/

 

As you can see, there is a SPDIF header on the board in addition to USB 2.0 ports.  There is even an integrated 24/192 DAC and so Pink Faun has options regarding what they can offer with this platform.

 

Where I have questions is with the CPU.  They utilize ARM processors in an SoC architecture and thus far, I have been underwhelmed with SQ from devices that used ARM processors.  Not that they're bad but the best I have encountered have used X86-based CPUs although I am happy to be proven wrong.

[austinpop]

Let’s keep politics off of here please.

 

[Superdad]

8 minutes ago, austinpop said:

Let’s keep politics off of here please.

 

My apologies.  And you will note that earlier I even marked my own post--about international postage costs--as OT.

[lmitche]

53 minutes ago, romaz said:

Is it because X86 CPUs are inherently lower latency due to their design and perhaps larger memory caches?

Roy,  I don't know enough about the differences between Intel and ARM cpu designs to offer an informed opinion. My conjecture is that ARM CPUs most often operate in mobile environments and Intel designs in more complex power and connectivity environments on desktops and in datacenters, with laptops an exception. If it is correct that Intel cpus are designed for harsher environments it wouldn't surprise me if there are engineering differences that harden the Intel design. Perhaps those translate into a SQ advantage?

 

The Asus Tinker board looks a lot like a Raspberry Pi. I have almost zero experience with this class of machine with audio except for the old SMS-100 and a short time with the original microrendu. Both are greatly surpassed by today's standards. It will be fun to learn more about the Asus solution over time.

 

We can't discount the impact of a low latency OS in either environment. I don't know of a low latency OS in the ARM world, but haven't done much, if any research into this topic.

[jabbr]

1 hour ago, lmitche said:

...

We can't discount the impact of a low latency OS in either environment. I don't know of a low latency OS in the ARM world, but haven't done much, if any research into this topic.

 

The ARM world is trending toward SoCs which can have network optimized hardware built into the same chip as the CPU. The Marvel Armada series is designed for network ops and I've used for audio streamers:

 

Solid-Run ClearFog has both an SFP input as well as RJ-45

Solid-Run MacchiatoBin has a set of SFP(+) input (10Gbe) as well as M.2

GlobalScale EspressoBin has an M.2 which allows for WiFi 802.11ac

 

Mine run with Realtime linux kernels... (4.18.12)

 

These pups are designed for low latency network I/O ... and cheap  

[seeteeyou]

The term low latency could be used quite casually, sometimes they might simply use that for marketing purpose without stating what the value is or maybe in other cases how a particular value is obtained.

 

Here's the latest from Mellanox with hundreds of nanoseconds

 

http://www.mellanox.com/page/press_release_item?id=575

Quote

Mellanox will demonstrate the industry-low, sub 300ns latency performance of its recently introduced 10/40Gb Ethernet switch using Corvil's market-leading latency management system, CorvilNet.

 

http://www.mellanox.com/blog/2018/05/ethernet-storage-fabric-part-2/

Quote

Other Ethernet switches often produce latencies in the tens of microseconds in actual deployments, whereas Spectrum switches have ~300ns port-to-port latency and zero packet loss, regardless of frame sizes and speeds. Furthermore, Spectrum switches are designed with a shared buffer, resulting in maximum micro-burst absorption capacity.

 

http://www.mellanox.com/related-docs/solutions/SB-breaking-the-low-latency-trading-barrier-with-next-gen-intelligent-interconnect.pdf#page=2

Quote

Mellanox Spectrum-based platforms demonstrate sustainable cut-through latency of 300ns with zero frame loss at all frame sizes and at all link speeds. Moreover, this latency is preserved even in mixed environments, combining slower and higher speed ports.

 

Though their latest ConnectX-6 EN (Ethernet adapter) turned out to be much higher

 

http://www.mellanox.com/page/products_dyn?product_family=268&mtag=connectx_6_en_ic

Quote

ConnectX-6 EN supports two ports of 200Gb/s Ethernet connectivity, sub-600 nanosecond latency, and 200 million messages per second, plus block-level encryption and NVMe over Fabric offloads, providing the highest performance and most flexible solution for the most demanding applications and markets.

 

Then there's always something like this

 

https://blog.metamako.com/how-to-measure-the-latency-of-a-4ns-switch

Quote

We have conclusively shown that the latency of the MetaConnect 16 is very close to 3.95ns per hop.

 

[seeteeyou]

On 8/10/2018 at 3:01 PM, romaz said:

Regarding power supplies for this NUC board, yes, it scales extremely well to power supplies and better than my DFI board.  This board can accept 12-19V.  During my testing with my SR7s, 19V sounds a little better than 12V but DR (double regulated) 12V sounds considerably better than 19V.  The ideal SR7 for this board would likely be a DR 19V which I do not have.

 

 

NUC7i3DN and NUC7i5DN came with only 1.5MB cache per core, and we're still getting 2MB cache per core from NUC7i7DN. However, ALL of them could go as high as 24V

 

https://www.intel.com/content/dam/support/us/en/documents/mini-pcs/nuc-kits/NUC7i3DN_TechProdSpec.pdf#page=51

https://www.intel.com/content/dam/support/us/en/documents/mini-pcs/nuc-kits/NUC7i5DN_TechProdSpec.pdf#page=53

https://www.intel.com/content/dam/support/us/en/documents/mini-pcs/nuc-kits/NUC7i7DN_TechProdSpec.pdf#page=51

 

And then the BEST part is an internal (lower impedance) Molex connector in addition to an external (higher impedance) DC connector, therefore it should be easier to try power cables with higher gauge number / lower resistance.

 

Besides, just wondering if we're able to apply a somewhat similar trick as what Innuos was doing?

 

https://www.computeraudiophile.com/forums/topic/30376-a-novel-way-to-massively-improve-the-sq-of-computer-audio-streaming/?page=313&tab=comments#comment-813712

On 4/26/2018 at 4:25 PM, nvitorino said:

1. ATX only (using a Pico-PSU with a clean 12V input)

2. 12V only

3. ATX + 12V

 

1. and 2. sound similar in our tests. 3 sounds significantly better.

 

We could even run i3-7100U / i5-7300U @ 800MHz (7.5W TDP) easily and maybe DDR-4 could also match that speed if BIOS options weren't too lousy? Same deal with i7-8650U @ 800MHz (10W TDP) so that's equal to the TDP of Celeron J4005.

 

 

66 quids versus 166 bucks, oh well

 

https://www.quietpc.com/akasa-newton-s7d

https://www.amazon.com/Akasa-Newton-Fanless-Generation-NUC38-M1B/dp/B079YXZTQM

http://www.performance-pcs.com/akasa-newton-s7d-for-7th-generation-intelr-nuc-boards-dawson-canyon.html

 

Still £83 versus $140

 

https://www.quietpc.com/akasa-plato-x7d

https://www.amazon.com/Akasa-Plato-fanless-Generation-NUC39-M1B/dp/B079Z1FNJH

http://www.performance-pcs.com/akasa-plato-x7d-for-7th-generation-intelr-nuc-boards-dawson-canyon.html

 

 

https://community.roonlabs.com/t/the-best-nuc-for-rock/45601/35

[rickca]

8 hours ago, romaz said:

When I spoke with Pink Faun, they told me they would be using ASUS Tinker Boards for their new streamers.

Roy, I'm just looking at the Pink Faun specs written in Dutch in the photo from this post

https://www.computeraudiophile.com/forums/topic/30376-a-novel-way-to-massively-improve-the-sq-of-computer-audio-streaming/?do=findComment&amp;comment=886594

 

It says standard outputs are coax, AES/EBU and optical.  It also says something about a configurable I2S/LVDS option.  No mention of USB but I do see there are USB 2.0 ports on the Tinkerboard.

[rickca]

7 hours ago, lmitche said:

The Asus Tinker board looks a lot like a Raspberry Pi.

Here's a nice comparison

https://www.jeffgeerling.com/blog/2018/asus-tinker-board-compelling-upgrade-raspberry-pi-3-b

[FredM]

1 hour ago, rickca said:

Roy, I'm just looking at the Pink Faun specs written in Dutch in the photo from this post

https://www.computeraudiophile.com/forums/topic/30376-a-novel-way-to-massively-improve-the-sq-of-computer-audio-streaming/?do=findComment&amp;comment=886594

 

It says standard outputs are coax, AES/EBU and optical.  It also says something about a configurable I2S/LVDS option.  No mention of USB but I do see there are USB 2.0 ports on the Tinkerboard.

 

Roughly translated it says:

Soon available Scrion

The PinkFaun Scrion is a Roon endpoint

Equipped with a linear supercap power supply

Standard outputs AES/EBU, coax and optical

Expandable with multiple digital output variants such as i2s-TTL or LVDD