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

[Peter Avgeris]

I see here that quite a few guys have experimented with this Startech SFP network interface card. I would just like to add my two cents here.

 

First of all, this card carries the Realtek 8168 B/E etc chipset, which costs really nothing. Cents of a dollar. On top of this, this Startech NIC requires the addition of a SFP module to function. And the impact of these modules to the final SQ has been already demonstrated by a lot of guys. For my set up I need 4 cards, one in my NAS, two inside my audio server and one on my audio endpoint.

 

I find the price of Startech NIC ridiculously expensive, that does not make sense for a massively commercially released product, specially when the heart of the NIC costs cents. So before throwing 1.000+ Euros into the bin, I decided to make a little research.

 

Back in the old days, light transmission was never better that the copper interfaces. I am sure you remember the old optical interface vs. the coaxial interface. It seems that for small distances (and not 100s of meters in length) the copper solution might outperform the optical. So I did a little research to figure out what was better in Startech NICs. It should not be the light interface, to my instinct, it is the chipset that should make the difference, at least up to +90%. So I performed a search in order to find a reasonably priced NIC, with copper interface, good old trusted Ethernet copper connector, of course equipped with the same Realtek 8168 chipset, as this is the beating heart of the Startech adapter.

 

So I found out that there is a ridiculously cheap solution that would outperform many of the highly priced network adapters around. The exact description of the model is TP-Link 3468 v.3, it comes in both tall and narrow mounting bracket and it sells for 6-9.5 Euros. Yes! Less than 10 Euros for a NIC that carries the same beating heart as the Startech device does! I need four cards, so I got some more. Seven pieces in total. I also told of this story to a friend living in Germany. He got the same cards as well.

 

To his surprise (and mine, too!) this card was found to greatly outperform well known commercial applications that cost in excess of 400 Euros per piece. A couple of seconds of a heads-up battle was enough to tell the winner without the slightest hesitation. And if you consider that here we have no extreme power supplies, no extra clocks or oscillators, nothing but the core, then this device has a truly great potential to be the absolute winner! Although Realtek devices are considered to be real crap things for servers (due to lack of drivers, support, etc.), it is found that for audio applications, they really outperform probably everything out there!

 

So, before spending 1000s of Euros in NICs, please do me a favor and spend a mere -20 Euros for a pair of this 'crappy' (!) NIC! I am pretty sure that you will be surprised (at least!). This little thing surprised me to the point of thinking of manufacturing a final product for the HiFi community, equipped of course with truly exotic power supplies and premium oscillators, that would cost less than 100 Euros as a final product, all included! Although my tube amplifiers cost ridiculously high amount of money, with exotic iron double c cores, etc, in excess of 100k Euros, at the same time I am a strong believer of doing your best effort for the lowest price you can.

 

FYI, this lovely little crap is currently breaking in. It is found to be extremely close to the Startech NIC equipped only with the flagship Planet module (all else are clearly inferior!), so damn close that properly breaking it in might tell this the final winner. So damn close! I will inform you further.

 

[Peter Avgeris]

16 hours ago, beautiful music said:

 

Can you post a link and what is the difference between V.2 and V.3?

I really have no idea. Maybe minimal upgrades? v3 is the current production model. I do not know, maybe it could perform worse than v.2...

[Peter Avgeris]

3 hours ago, bodiebill said:

For many no option I guess, but I am now enjoying the best thinkable network topology with no network devices or -connections having an adverse effect on SQ: namely no network at all.

 

I tried many things in order to eliminate the nasty effects of network activity on the sound. RJ-45, fiber, JCAT net card FEMTO, MikroTik and Cisco SFP routers, LPS's to match, and the SQ got better and better. Converging to the SQ that can only be had by... no network at all.

 

I did try this before, over a year ago, but then I had a DAC (Lampizator) that made DSD sound much better than PCM, and upsampling to DSD256 with HQPe/EC was the best achievable. This for me required a network.

 

Now I have a DAC (Denafrips) with glorious PCM and I no longer feel the need to convert/upsample to DSD256. And wtfplay is an gorgeous player for PCM that eliminates all frills and totally concentrates on SQ. In spite of many requests the author of wtfplay has not added remote control via network to his player, and I can understand why. It would lower the SQ.

 

Controlling wtfplay with a tiny keyboard with trackpad and a monitor (a tv) is actually quite easy and user friendly. You can even create a playlist and then play it from the command line, with no need for a GUI. 

  

I could have spared myself a lot of cost (including that for the LPSU's) and much trouble, but it was fun to do. And I may occasionally go back to a network setup for convenience. But the SQ I am now hearing is hard to resist and forgo.

Hello! 

I have read your statement with so much interest. To make the long story short, if you are looking for maximum sound quality, your approach, although understandable and acceptable, to my experience is absolutely wrong. 

On the other hand, overdoing the net way is also completely wrong. The truth seems to be lying in the details and clever moves & decisions. 

I will note some issues. First of all, the magnitude of impact of the VGA is probably enough on its own to make you abandon this idea in a couple of minutes. In all my systems, the presence of an active VGA adapter either on the MoBo or inside the CPU is absolute catastrophe. In addition, splitting the server and playback services in different pcs is a move you can't go backwards. I have been through this numerous times. A fine tune (not to the max) of a two pc setup can't compete against a single pc setup. Not a chance. 

I could mention several more parameters. You can enjoy your music through a single point but when we are talking of sophisticated network builds, we talk of totally different scales of quality. 

With kind regards

Pavgeris

[Peter Avgeris]

5 hours ago, bodiebill said:

 

 

Hi pavgeris. If I understand well you always prefer a 2 box setup (with network) to a one box setup? I am asking as the second sentence seems to say the opposite.

 

 

 

 

You are sooo right. I wrote exactly the opposite, thank you for your attention. 

[Peter Avgeris]

5 hours ago, bodiebill said:

 

 

Hi pavgeris. If I understand well you always prefer a 2 box setup (with network) to a one box setup? I am asking as the second sentence seems to say the opposite.

 

I did not mean to be categorical and remain open to different topologies, but reported the change in SQ between my previous complex 2 box setup and the new 1 box setup.

I did wonder about the effect of the hdmi and keyboard, compared to that of a NIC (and everything behind it).

 

Hard to identify the key reason for the 1 box setup to sound better (to my ears), as so many things are different, including the SW. But given that it does and that so many devices can be removed, makes me prefer it for now. 

 

 

I have no experience with keyboard /mouse but the video adapter is for sure detrimental. 

I am now upgrading my network, I will let you know about my results. 

[Peter Avgeris]

4 hours ago, Downtheline said:

Can you expand on this? I've not come across this report before. 

Very easy! 

Fine tune everything and after you have finished by being able to fully control remotely, take off the vga adapter or alternatively, switch it off through the bios or jumpers. You can't go backwards. It is beyond logic to upgrade all your stuff and leave vga working! 

[Peter Avgeris]

On 5/23/2020 at 12:07 AM, dctom said:

Thanks for this.

Thought I would give the TP-Link 3468 v.3 a try as it is such a low cost. Unfortunately it seems to create a hardware conflict with my Z390 phantom MB and OS. I am running audiolinux are you using windows?

 

 

 

 

 

No. I am using Euphony Linux. Not a conflict in 4 systems! 

[Peter Avgeris]

7 hours ago, dctom said:

Thanks - have got it to work now - however , strangely, I need to start up with ethernet cable disconnected, then plug it in when computer is up and running. Otherwise the system will not recognise my internal NVMe internal library storage?

So strange! 

Do you notice anything better with this little card? In my systems it was a clear winner in every aspect. 

[Peter Avgeris]

5 hours ago, MagnusH said:

Here is a little USB tweak I made to a cheap TP-Link network card. Nothing to fancy, I just cut the 3.3v connection to PCIe and patched in a linear voltage regulator board and a connector to get external power (currently using a battery pack, 4*AA). It made a nice improvement to the sound though, calmer and more fluent. 

 

But I would like to take this further, maybe a better clock? Or cut the ground connection to the PCIe and make the card floating (will the card work after?). Suggestions are welcome 🙂

I see that you have made a nice little mod on this board.

It is fantastic!

Did you cut only one trace on the pcb? If I ma correct there exist 4x 3.3V connectors on PCIe interface.

I also see that you desoldered the LEDs. I hope that you got really nice sound out of this mod...

[Peter Avgeris]

2 hours ago, MagnusH said:

I also just ordered a 25Mhz NDK NZ2520SD for $6 (but shipment $30 from Japan), going to connect that to the regulator board and then the clock output to the little black chip's clock input and cut the connection to the old clock. Lets see if it makes any difference (the clock has very low phase noise).

That's so nice.

Please inform us of the upgrade as soon as you complete the process.

I had uploaded a little article about this card about a month ago. Did you already have this hard or maybe you read that post and you decided to get it?

If you read the article and then you decided the upgrade, then I am very curious/interested of your findings. I hope I'm not the only one here... Personally I liked this little card a lot!

 

Thanks for your update.

[Peter Avgeris]

14 hours ago, Nsxturbo said:

 

Great report!

 

I recently received my 2nd Buffalo BS-GS2016 to add to my 1st one, which currently is just running on a 12v linear supply and connected to a EtherRegen clocked with the Mutec REF10. I’m actually liking the Finisar transceivers over the Planet Tech’s.(more air and delicacy without loss of detail or dynamics.)

 

Looks like I’ve got more experimenting to do besides the 2nd linear supplied Buffalo😉 I might as well try replacing one of the Buffalo’s 25Mhz clocks with a sCLK-EX 25Mhz synthesized output fed by the REF10, since I already those two pieces in the stable. That will give us all another data point!

I have tried to find the BS-GS2016 model but it is absolutely unavailable in Europe. Not a single shop has it available, so I got a pair of the same but with 8 ports, without SFP, as I am not interested in SFP at all. I got the BS-GS2008 whoch should normally be the same but with fewer ports (also not needed). I am expecting to receive them soon so I will let you know of my impressions.

[Peter Avgeris]

11 minutes ago, Nsxturbo said:

Crazy thing about the availability, because I’m in the US but actually got my last one on EBay from Italy! Go figure.

It is out of stock throughout the planet. It seems that the HiFi maniacs have gone crazy with this Melco switch. The S100 model is like the BS-GS2016. Not being interested in SFP at all, I think that 2008 is still preferred, inside they are the same but with fewer ports and (logically) less noise. I will try it and will let you know further.

 

FYI, I have been notified by the greatest network guru in Greece that the audiophiles *MUST* purchase two net switches and *NEVER* one. Stacked switches is definitely the way to go if they are used in a HiFi system. A local friend had purchased 1x S100 and when he tried to stack a second identical machine, he was shocked in awe! One machine is not enough, as dual switches help cancelling out the induced noise instead of trying to filter it. Noise filtration can't be effective because HF always finds a way to pass through. This is what I have been informed by the greatest network guru here. He also told me that anyone who buys very expensive switches does not know what he's doing. It is preferable to stack two cheap machines instead of getting one very expensive unit. This is what exactly he told me about this issue.

[Peter Avgeris]

13 minutes ago, Nenon said:

 

I am not sure about that. The Melco board is exactly the same as the BS-GS2016 board. But it's installed in a chassis that does not have opening for the top 8 ports, so you only see the bottom 8 ports. 

People who have been tinkering with the Buffalo BS-GS2016 before the Melco existed have noticed that using the bottom ports sounded better than the top. That's how the Buffalo BS-GS2016A version appeared for audio. My guess is this was the reason Melco did not expose the top 8 ports at all. I have not seen an 8-port Buffalo switch for audio... the Japanese audiophiles prefered the 16-port Buffalo with plastic caps covering the top 8 ports. 

The other thing Melco suggests is to use the first 4 ports on the Melco (the first 8 ports on the Buffalo) for your streamer and the second half of the ports for the rest of the network. They also configure the first half of the ports as 100Mbps, but that does not work well for me. The second half of the ports are 1Gbps, and they suggest those for the rest of your network. 

This particular Broadcom chip must be playing a big role of the sound signature of the Melco/Buffalo. Haven't seen a block diagram to understand the switch fabric architecture, but the isolation between some of the ports, might be playing a big role here. 

 

The Broadcom chip is the most expensive part in that switch. It's almost $80 on Mouser if you buy it in the hundreds / thousands, which makes me think it is at least half of the cost of the entire switch. It does not make sense that Buffalo would be using the same 16G chip for their 8 port switch and sell it for half the price, or nearly the cost of the 16G chip itself. That's not how these companies make their profit. We are not talking about audiophile switches here. Buffalo makes small business switches in large quantities at consumer prices, not overbuilt/overpriced switches for audiophiles. 

 

Peter - remove the heatsink of the switch, clean up the thermal paste and check the Broadcom chip part number. Send it to me, and I will confirm if it is the same. It would be nice if it is, but I highly doubt that. 

Yes, I will do it, no problem. I am curious too. The price of 2008 is not far away from 2016. It might be a different chip by Broadcom, it might be the same, it might be a better one in terms of sound quality, nobody knows unless he compares heads up. Like in the case of Realtek network adapter, which is an absolute crap in terms of drivers but it outperforms all Intels I have tried so far. You never know in audio 🙂

[Peter Avgeris]

3 minutes ago, Nenon said:

 

I bought some spare Buffalo switches while they were still in stock, and I think I can try 6 switches in parallel. Not practical and need a bunch of cables, but I will give it a go just for the sake of it. The problem is they need at least 3 week to break-in properly as they sound pretty bad the first couple of weeks. 

The 16-port model is out of stock and probably will never be in stock again, as most likely the company plans to release another model. Who knows?

[Peter Avgeris]

3 hours ago, austinpop said:

 

LOL ... you are "preaching to the choir!" If you look back on this thread over the last 3+ years, you'll see this has been explored extensively.

I do not intend to claim any royalties out of your discovery.

Don't worry!

Just wanted to emphasize something I have been told about...

[Peter Avgeris]

I would just like to add my two cents here...

 

Please do not consider my writings as an offense but as a starting point for further experimentation and critical thought. I need to double-underline this, as I have no intention to 'treat' one's mind or to persuade anybody in favor of my thoughts.

 

I have been in the audio cables business for several years as manufacturer of audio cables. Together with my partner we had made in depth analysis and theoretical approach to the plethora of the phenomena that impact sound waves transmission through conductive material. Both lab measurements and theoretical models of transmission.

 

After having spent more than 100k Euros in this arena, two things I would like to focus on:

 

1. Don't lose your valuable time and money searching for silver conductors. All silver conductors, regardless of structure, treatments, mixture with other metals, etc, sound *worse* than copper conductors but please be careful: all other parameters being equal. When you compare silver vs copper with all the parameters of A-B comparison being identical (diameter, insulation, tempering, crystalline structure, etc), copper *ALWAYS* wins and silver always loses. This of course means that you may experience a silver cable performing better than a copper cable but when you examine the conductors under identical denominators, then copper always wins.

 

2. Don't spend a cent for cryogenicaly treated parts and accessories. They are so bad for audio. They take out the soul and magic of music. I'm a mechanical engineer as you know, with gathered experience in this field for aprox. 20 years now. Cryogenic treatment does exactly the opposite of what we are looking for, in audio reproduction. Several guys here know who I am. For those that don't, I need to write that I manufacture audio reproduction components where cost is not considered an obstacle. In my components, a single cryo treated item or part or whatever is enough to destroy their flavor and their signature. I have evaluated each and every component in the past and under no circumstances I accept to install cryo treated parts, even if this is a custom order or even if this has the impact of cancelling the order. In my amplifiers I am using an IEC socket that costs around 50 cents of a dollar only for the reason that it sounds best among all. I hope that this might sound as something strange to your ears. It's true. 100% true.

 

The debate copper vs silver has a long history. Please don't start over another debate here. After having spent so much in research and manufacture in the field, with this notice I would just like to declare my findings so as you benefit. I know that the vast majority here will not take my comments into account, they won't have a second thought because they also have their own judgements. My intention is that the few that anticipate life through a thoughtful approach, now have a hint to think over their own findings and judgements once more. Under a critical thought and point of view only!

[Peter Avgeris]

59 minutes ago, RickyV said:

 

Mmmm wouldn't a pure copper and gold plated IEC inlet be better than a 50 cent one?

The 50 cent ones are usually brass. 

Yes! Of course.

But all the pure copper AC inlets seem to be in conflict with the parameter #2 of the same text. And cryo makes sooo bad...

[Peter Avgeris]

Hi guys.

I owe you something here.

Today my Buffalo BS-GS2008 switches arrived. I took one of them and disassembled it.

I took a pic of the chipset. Here it is.

Nenon should reply here...

 

[Peter Avgeris]

So they are the same series chips, one is for 16GE and other is for 8GE networks, as the page on Mouser designates.

[Peter Avgeris]

Hi again. 

What I would like to state with my writing regarding copper vs silver is far away from trying to persuade somebody to follow my path. This is the reason I tried to make it as clear as I can. Follow your ears, follow your instict but try to be open and thoughtful, nothing else. 

Regarding copper & silver, I am not referring to third party products. I am referring to designing and manufacturing the cables from scratch in specialized laboratory, with specialized equipment under very strict and controlled conditions. The cables were not extruded but casted continuously through custom designed carbon nozzles with temp gradient of around 1 deg C above and below solidification temperature. The average crystal length was around 400 meters in both cases. No dielectrics, no insulation. Bare conductors. This was the last judgement of copper vs silver. The theoretical results came in line with findings through listening tests. 

This is my final statement here regarding copper vs silver. I have no intention to dig deeper, just to pass on my own findings to you. We live in a free world where everybody is free to do whatever he wants! 

 

Regarding the dc-dc converters, I am unable to read the part numbers as it is impossible to read the letters. A friend has an iPhone with pro lenses, I will call him tomorrow for further assistance with pictures. 

[Peter Avgeris]

1 hour ago, Superdad said:

 

So that is a Monolithic Power Systems MPQ8632 synchronous buck converter. That series uses the same 3x4mm QFN package for all models from 4A to 12A. Based on the value of the inductor next to it, that one is either the 4A or 6A version.  

 

 

Don't worry about it Peter, at least one of them has been read above.

 

BTW, seems your web site (exoticaudiodesign.com) is down. I hope China is not blocking access to Hong Kong sites now. 

 

Cheers,

--Alex C.

Hello Alex. 

The website is not down due to China's control in HK. It is going to be completely redesigned with all models and series. 

[Peter Avgeris]

On 6/16/2020 at 3:29 AM, Superdad said:

 

My eyes certainly popped out when I saw the size and complexity of your phono preamp.

 

 

Thats a LOT of wire and magnetics for just a PHONO PREAMP! 

Very impressive...

Hello!

Thaks for your comments.

The first picture is the phono preamplifier. MM only. For MC cartridges MC step up transformers are required.

The last picture is a power amplifier manufactured eight years ago. EML20A > EML300b mesh > United Electronics 311

All iron stuff is of course calculated, designed and manufactured in house.

[Peter Avgeris]

On 6/16/2020 at 10:03 AM, RickyV said:

 

OMG @Peter Avgeris this must be in the guinness book of records for the biggest phono stage ever. And the first picture is the preamp?? Or is it all phono amp? 

Hello from Athens.

The first pic is the Exotic 3-chassis MM Phono Preamplifier. You are correct.

I have already designed the next level of MC Step Up Transformer. It is one chassis with 4 pairs of SUTs, it is plugged on AC mains and is made out of 24 iron pieces for the stereo set.

[Peter Avgeris]

5 hours ago, John769 said:

 

 

Just wondering if the various net card comparisons you made also included s/w using Windows or Server OS?  In particular with the intel based cards versus the TP-Link, bearing in mind you preferred the latter.

Hello!

The comparisons were made on Euphony Linux. Regarding comparisons under Windows, unfortunately I have no news.

Under Euphony on my own system, all Intels I have tried (82574, i210, i350-T2, i350-T4, i219-V, X540) were clearly inferior with the results coming in less than 10 seconds.

I am now playing with this small adapter and have plans to replace the crystal oscillator with OCXO some time in the future.

[Peter Avgeris]

I have plugged the Buffalo switches that I have received two days ago.

After having them on fire for aprox. 4 hours I decided to make some listening tests.

The Control Point PC was connected on the NAS. In between them I placed the two switches in series. NAS routes the packages to the Control Point PC, so here I did not follow the 'less is more' approach.

No LPS, no special cables, no special equipment. A=nothing, B=2x BS-GS2008

I went A>B and B>A several times.

With the double switch there was always more clear sound and better definition but something was missing. Music had lost some of its thrill & excitement. OK, I thought some sort of burn in would be required.

After 3 hours I decided to try the polarity of the power supplies. Here in Europe, the Schuko power plugs have been designed in a wrong way and the phase/neutral conductors are not explicitly specified. All the good things that were missing are now back, on top of course of all the good aspects.

To sum it up: once you go from A to B, you can't go backwards from B to A.

I do not know if Melco switch does miracles but for the 140 GBP I paid for the pair, I ma more than simply satisfied.

As soon as BS-GS2016 is discontinued, I cannot see a reason not to go for the BS-GS2008.

Just my two cents!