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

[Nenon]

26 minutes ago, dminches said:

I thought this could be accomplished by running the ATX off group 1 and the CPU off group 2.

Absolutely.

I am just assuming that P8 (12V) --> EPS might be a cleaner DC than P8 --> Hdplex DC-to-DC --> EPS.

[Nenon]

1 hour ago, Chopin75 said:

Sorry, I am just as confused here. You have 2x 19V/8A PS, each supplying part of the CPU, but then you stated it is actually using 12V x2, 12V from each PS ?? Then what happens to the 19V? Is it that the 19V is shared for both the MB and CPU (12V to CPU and 7 V to MB?), so you have 12V x 2 to CPU and 7Vx 2 to MB, if that is even possible??  The simplest is of course 19V to MB and 12V to CPU, as suggested above. Thanks for clarifying!

 

 

He has:

Keces P8 (19V) --> HDPlex 800 Group 1 --> EPS (12V) and ATX (3.3V, 5V, 12V)

Keces P8 (19V) --> HDPlex 800 Group 2 --> EPS (12V)

 

The HDPlex is converting the 19V to 12V /for the EPS connector(s)/.  Look at the diagram he sent, it makes perfect sense (to me). 

[Nenon]

54 minutes ago, bodiebill said:

If the CPU of my minimized endpoint PC only draws, say, 1.5A (as it seems to do), why would it care if the PS is able to offer more?

 

If the CPU draws 1.5A, I would feed it with a LPS that can provide 8A, or it least 6A. That from my experience sounds a lot better. It's what other people have reported too. Oversizing the power supply feeding the CPU makes a difference. Don't ask my why... I have a theory but not worth getting into it. 

[Nenon]

5 minutes ago, bodiebill said:

 

Thanks Nenon, I tried the same setup except using the HDPlex normal 12V/10A output instead of its (variable) 12V/2V LT3045 output. Indeed I think I prefer the 12V/10A. Better slam, dynamics.

 

Downstream I now need to power:

A. motherboard with 19V

B. CPU with 12V

C. iFi micro iUSB3.0 with 9-12V

 

When my Paul Hynes arrives I will have the following power linear power supplies:

P1. Paul Hynes SR4: 9, 12, 15 or 19V / 2A

P2. SBooster BOTW ECO MkII: 19V / 1.75A

P3. HDPlex 200W: 12 or 19V / 10A or 5, 9, 12 or 15V / 2A (LT3045 based)

P4. MPAudio DLS-HPULN (LT3045 based): anything up to 15V / 5A (to be powered by anything up to 17V)

 

In addition I have three Studer900 LPSU's (12V, 15V and 19V).

 

Currently:

P2 => A

P3 12V/10A => B

P3 12V/2A => P4 => C

 

For best SQ, what to power with which?

 

Of course I will try, compare and report back, but it would be nice to know your and other's expectations...

 

 

This is hard to answer or even give you any recommendation. I have only used the SBooster and HDPlex from your set of power supplies at very early stages of experimenting with music servers. I don't have them anymore and would not consider them in my system again. I have not tried the SR4 or the MPAudio. 

 

Instead of trying to map a power supply to component, let me try tell you what I think in general. There are two factors - power and quality. Typically the best quality power supplies cannot provide a lot of power. 

 

If 'quality' was the only factor, it would have been a really easy recommendation:

You put the best quality power supply you have on the USB. The second best goes to the CPU. The third best goes to the motherboard. 

 

If 'power' was the only factor, it would have been a really easy recommendation too:

You put the most powerful power supply you have on the CPU. The second most powerful goes to the motherboard. The third most powerful goes to the USB.

 

The problem is when you have to balance between those two factors and introduce a dozen other factors - then it gets really complex. You would have to experiment and see what works best. A good starting point for those experiments would be to:

• Put the most powerful power supply on the CPU.
• From the remaining power supplies, pick the best quality one and put it on the USB.
• From the remaining power supplies, try the best quality power supply on the motherboard; but you may also try the most powerful, or something in between. 

Good luck with the tests and let us know what works best. 

 

Oh, and one more note - I am increasingly starting to think that building a computer with multiple rails of the same power supply and wire accomplishes more natural sound overall. Look at my Building a DIY Music Server thread for example of what I mean by that. 

 

 

[Nenon]

I have two things to report:

 

1. For those of you following my 'Building a DIY Music Server' thread, you may know that I switched the AsROCK Z390 Phantom Gaming-ITX/ac to the ASUS Z390-I ROG Strix Gaming Intel LGA 1151 mini ITX motherboard. I compared the two, and I liked the ASUS better. It goes on the top of my recommendation list for Intel Core ITX motherboard. 

 

2. The Intel Core i9-9900K CPU does not support ECC RAM. Some people here have built their server with AMD Ryzen CPU mainly because of that. However, while reading the AsROCK motherboard specs, I noticed the following under memory:

- Supports ECC UDIMM memory modules (operate in non-ECC mode)
I found that quite interesting. Having some Apacer ECC RAM on hand, I decided to try if the ECC would work on the ASUS motherboard. It did! I am running the ASUS motherboard with ECC RAM (operating in non-ECC mode). And it sounds very good. I will do more back and forth comparisons with ECC vs. non-ECC (both from Apacer), but I really like what I am hearing. 

[Nenon]

5 hours ago, Superdad said:

Amazing to see all this discussion of power supplies without a single mention of one of nearly the most consequential factor, output impedance. Computers are a very “bursty” load, and given adequate current and even just average noise performance, the supply with the lower broadband output impeadance will always win.

 

Sure we did! We just did not call it with the technical term low 'output impedance'. I called it 'quality'. And absolutely - the low output impedance is one of the main things that makes a LPS 'good quality'.

 

On 1/16/2020 at 12:51 PM, Nenon said:

 

This is hard to answer or even give you any recommendation. I have only used the SBooster and HDPlex from your set of power supplies at very early stages of experimenting with music servers. I don't have them anymore and would not consider them in my system again. I have not tried the SR4 or the MPAudio. 

 

Instead of trying to map a power supply to component, let me try tell you what I think in general. There are two factors - power and quality. Typically the best quality power supplies cannot provide a lot of power. 

 

If 'quality' was the only factor, it would have been a really easy recommendation:

You put the best quality power supply you have on the USB. The second best goes to the CPU. The third best goes to the motherboard. 

 

If 'power' was the only factor, it would have been a really easy recommendation too:

You put the most powerful power supply you have on the CPU. The second most powerful goes to the motherboard. The third most powerful goes to the USB.

 

The problem is when you have to balance between those two factors and introduce a dozen other factors - then it gets really complex. You would have to experiment and see what works best. A good starting point for those experiments would be to:

• Put the most powerful power supply on the CPU.
• From the remaining power supplies, pick the best quality one and put it on the USB.
• From the remaining power supplies, try the best quality power supply on the motherboard; but you may also try the most powerful, or something in between. 

Good luck with the tests and let us know what works best. 

 

Oh, and one more note - I am increasingly starting to think that building a computer with multiple rails of the same power supply and wire accomplishes more natural sound overall. Look at my Building a DIY Music Server thread for example of what I mean by that. 

 

 

 

[Nenon]

1 hour ago, bodiebill said:

 

A Brennenstuhl power meter arrived today and I did some measurements.

I may need some help here as I am somewhat baffled by the first results. Please note that I only measured my minimal audio endpoint PC with AudioLinux headless (meaning GUI-less) + NAA. (I am not in a hurry to measure my heavy-lifting server as I am happy with its current power supply.)

 

First the audio PC specs:

mobo:    Supermicro X11SAE-M
cpu:    Intel Core i3-6100T (35W TDP)
memory:    2x Apacer D31.23185S.001 4GB ECC DDR4 2666MHz (wide temperature)
PSU internal:    HDPlex 800W DC-ATX converter

NIC:    Intel X520-D1

HD/SDD:    none

 

CPU / EPS

I used a simple 12V SMPS:

- steady state (nothing running): 0.04A

- playing any kind of audio file (redbook, hi-res PCM or DSD) streamed from the HQPlayer server): 0.14A

 

ATX

I used a 19v Chinese linear PS, wattage unknown as I did not have an SMPS that fitted:

- steady state: 0.17A / 30W

- playing music: does not change, always at 0.17A / 30W (does not change)

 

So maximum total current drawn approx. 0.3A. Such low amperage? It does not make much sense to me. But it is consistent with the fact that I used to power the whole PC (ATX+EPS) with a 36W / 1.75A / 19V SBooster without problem.

 

Probably someone here can tell me what I am doing wrong?

 

Your equipment is not capable to catch microbursts. If it was able to update let's say every nanosecond, you would see bursts that might be even hundreds of times higher. 

 

[Nenon]

On 1/19/2020 at 2:46 PM, jean-michel6 said:

A small miracle ! 
 

I have been for several years in this game of building a very good server-player pc system . 
I went through the 626 pages of this thread and exchanged with several of you . 

You can have the detail of my system on my profile .  I have achieved very good musical results with this system.

 

However , I felt it was too complex and I wanted to make it simpler . I made two changes to my system :

 

1- I replaced the quite complex optical switch and optical network by an UpTone etherregen.

 

2- The pc player with its clocks driven by SOTM clock board , SOTM usb pcie card ,SOTM tx usb , SOTM hd usb (usb to spdif ) were all replaced by an upnp/dlna renderer from engineered . A small company in Switzerland who is selling it’s network renderer to Dartzeel , Soulution , CH ,Audiomat ....

https://www.engineered.ch/products/ered-dock/

 

My system now is : 

- server: Super micro x10 sba under daphile with an upnp dlna plug in 

- etherregen switch 

- player : engineered network card 

- 2 very good 12v psu and 2 at 5 volt.

 

Results : It is much simpler than my previous system but the small miracle is that the sound quality is much , much better than what I did get before . It is so much closer to the real performance. It is truly amazing. 
I could write many many things about the sq....
It cost much less than what I had before therefore an amazing move. 
 

Both the etherregen and the engineered card were key in achieving those results. 
I really encourage you to explore this road , the reward is exceptional .

 

 

This looks good. Thanks for sharing. I am planning to buy one and try it in my system.

 

One thing to note here is that @jean-michel6 is using SPDIF. There was a time when I used SPDIF output. What I found interesting at the time was that I was getting better results from devices that output directly SPDIF rather than outputting USB and then converting to SPDIF. There were some really good USB to SPDIF DDCs that did an excellent job when you power them with clean power, use good cables, etc. But I could always hear some digital artifacts of doing that. This is when I discovered the PinkFain SPDIF bridge and their ultaOCXO clock. 

 

5 hours ago, jean-michel6 said:

That is truly an excellent results to have this roughly 2500$ set up challenge successfully a 20 000 Euros drive !

Well, I remember the time when my goal was to make digital source sound better than my friends highly tweaked CD player. It was very difficult. But these days, many of our digital sources are much better than pretty much any CD transport out there. But again, that's a really good sign for the engineered modules. I like what I see and read about them. 

 

I had good results with the Allo DigiOne Signature player. With very good power, ethernet noise filtering techniques, and a Stillpoints Ultra 6 under it, it was sounding pretty good for the price but was never my best source. I guess a DigiOne Signature in the new USBridge SIG board would sound even better. It would be an interesting comparison to the engineered modules. I still don't think any of the Allo products has the potential to sound better than a properly designed and tweaked DIY server. But those engineered modules might have even better potential. I will try them and report back at some point.

 

 

7 hours ago, Dutch said:

Some interesting developments here. And full circle again from lightweight endpoints to heavy servers/endpoints back to the lightweight endpoints again.

 

So there’s the Intel i9/Ryzen7 or 9 servers, Allo USBridge, Optical Rendu and Engineered eRED-Dock options that should all outperform the NUC based endpoints. I had been thinking about replacing my dual server/NUC i7 endpoint for a single i9 machine build but now not to sure. The lightweight solutions make HQ powering stuff so much easier! 
 

Thanks all for sharing! 

 

I've had several Allo, Sonore, and SOTM products in my system, and I did not like any of them enough to keep them. I mentioned my experience with Allo above. The Sonore products were never my cup of tea - I've had a microRendu and ultraRendu, but never had the opticalRendu. The SOTM products also had their limitations. 

 

So far, a highly tweaked music computer sounds best to me. But that is because we don't have the right, properly engineered products from the ground up. My bet would be on FPGA technology rather than RaspberryPis or small mobile chips / CPUs that consume little power. The problem is that the FPGA programming for audio is not mature enough. The top world-class FPGA developers are working in sectors with much more profit (Financial industry, Military, Space agencies, etc.). There is one company that originally wrote most of the FPGA code for the first FPGA DACs, and as far as I am aware it might be somehow related to Rockna / Audiobyte. 

Take a look at the PS Audio DAC - they manage to improve the sound quality by releasing new FPGA firmware every six months. That probably confirms that the FPGA code is not mature enough yet. It will get there one day, and I will be the first to jump on FPGA.

And speaking of Audiobyte, their Hydra Hub looks interesting - https://audiobyte.net/products/hydra-hub. I friend of mine has one on order, so I will get a chance to listen to it in my system. 

 

But back to the engineered boards - I like them and looking forward to test them. Too bad they are not in stock until March. 

[Nenon]

9 minutes ago, RickyV said:

 

You are a busy bee, are you. I really like you are going to test the engineered board. Still a pity it doesn’t output usb. A while back I made a choice to go for usb and invested in that. But we have to see what the outcome will be. 

 

I am planning to integrate it inside a DAC that takes I2S input. I will be using the engineered I2S output with 1/3'' solid silver wires connected to the I2S input on the DAC. I will also test it with SPDIF output but more interested in the I2S interface with the DAC. 

[Nenon]

I have been looking at the Taiko Extreme server with huge admiration to Emile. There is no way on earth a DIY-er like me could ever get to the level of engineering and tweaking Emile has done on his server. And romaz does not help with his "actually took only 30 seconds" comment (to realize how good the Extreme is). Okay, maybe one day I will get an Extreme too. But until then I will keep pushing the limits of my DIY servers to see how far we can go. Yes, I have probably spent as much as a new Taiko Extreme on experiments, many of which were not successful. I realize that. But it's a nice hobby, especially in the winter months in Chicago. 

 

A natural thought in my head would be to try some of the components that are used in the Taiko Extreme. That would be without all the incredible things/tweaks Emile has done, which are many. How much of the performance of the Extreme can we get out of the stock components? 10%? 20%? 50%? 80%? No one knows. Curiosity to try this motherboard and CPUs (Asus WS C621E Sage with 2X Intel Xeon Silver 4114 CPU) kept growing and growing. But there are at least 3 challenges with that:

1. Power - you have to power up not one but two high powered CPU with clean power. There are some linear power supplies that can do that - Paul Hynes Designs and Sean Jacobs have high current linear power supplies. Farad is working on one. Marcin is also working on something. The HDPlex 800W DC to DC can also be a good option in addition to a high current power supply.
2. Passive cooling case for the extended ATX size motherboard. Turemetal UP10 is the only suitable case I know. Expensive but there is a solution.
3. RAM. Taiko Extreme uses custom high quality RDIMM RAM. And this is the biggest problem. We now know how much difference the RAM makes. And I would not want to invest so much into such build only to put some bad quality RAM modules in it. Our favorite manufacturer, Apacer, does not produce industrial wide-temperature RDIMM RAM. And that is the main reason I never bought that Asus WS C621E Sage motherboard.

Well, after long discussions with Apacer, they spec’ed custom wide temperature industrial memory for me and are willing to run one small custom batch of it. It has the same specs as the ECC RAM that sounds best. I am getting it! Taiko uses 12 x 4 GB. I am planning to get two sets. So far, there is no indication that this memory would be produced as a regular product, besides this custom batch, but you never know. However, I will not miss the opportunity.  If anyone else is interested, please ping me on PM.

 

 

[Nenon]

20 minutes ago, auricgoldfinger said:

 

That is going to be quite a project!

 

Indeed! I don't know when I will work on this, but I will start collecting some of the parts now. Probably after the summer...

Right now, I am working on this:

On 2/5/2020 at 4:55 PM, Nenon said:

I have used the AsRock in previous builds. It's a great motherboard. You can't go wrong with either of them. I have a slight preference towards the Asus.

 

Also, speaking of HDplex H5 and Asus ROG motherboards, I have the big brother - Asus ROG Crosshair VII Hero placed in a H5 case (not mounted, just placed on top). This is obviously an AMD motherboard, not Intel. 

 

When I did the build in this thread, I had a lengthy discussion with the now owner of this server, and it was decided to go with what I already knew works best rather than experimenting with new things. The Asus motherboard was a wonderful serendipity. So, time for new experiments with a bigger Asus motherboard now. 

 

The three BIG questions I would like answers for are:

1. Does this big full size ATX motherboard sound better than its smaller sibling I am currently using in their stock versions?

2. How much better Qobuz would sound (if it sounds better) when my JCAT NET Femto NIC is on a regular PCIe slot with no extension cables?

3. Last, but not least, what would be the effects of changing the motherboard clocks with high quality OCXO clocks? That's the part I am most curious about as it's something I have never done before. If time allows, I may even document the clock changing process here, but I am not promising.

 

What I like about the H5 case is that it supports a full size ATX motherboard and allows you to mount all PCIe cards vertically with no adaptors. 

What I don't like about the case is that this plate needs to come out completely every time you want to add or remove a PCIe card. But I haven't assembled it yet, so maybe there is a workaround this. It also feels a little cheaper quality than the Streacom, and I wish the top and bottom plates were thicker. 

 

This won't be a fully documented build like I did before, but rather some comparisons after I do the work. It would probably take several weeks before I have any discoveries I can share. I may post some info on the clock changing process, though. 

 

The first problem I am running into - I am not sure which PCIe slot is connected directly to the CPU and which goes through the chipset. If anyone knows, please let me know. It would save me some time. Ultimately, if I can't find any info I can probably trace the PCB traces on the motherboard and figure it out, but that would be too geeky :). 

 

Stay tuned for more updates.

 

[Nenon]

18 minutes ago, Dev said:

After you are done with your project, it will be very interesting to get a direct comparison with the Extreme (if someone can loan you one 🙂).

 

Let's be clear. I have no expectations to reach the level of the Extreme. I am not even aiming there. Just want to see if this motherboard / CPU / RAM would sound better than what some of us are using. A little change from mini-ITX gaming motherboards and consumer grade CPUs. Most people here build their computers with NUCs or gaming motherboards, which are really good because they are overbuilt for overclocking. But let's shake this trend a little bit and find out why the big guns are using those much bigger and powerful components. I would be happy if that experiment gives an improvement over what we are using. 

 

Also, I should have clarified in my first post. This is RDIMM. You need a special motherboard with Intel Xeon CPU(s) that supports RDIMM. It does not work on gaming / consumer grade motherboards / CPUs. 

[Nenon]

4 hours ago, austinpop said:

@Nenon

 

Thanks for cheering me up with that post. Now that @romaz has achieved moksha and ascended to a higher "I am done" state, we lesser mortals need a new leader to take us forward!

 

In all seriousness, I have been kinda depressed since I read Roy's epic Extreme posts on WBF. As you know, I have a similar build to Roy, also powered by SR-7 DR rails. I thought this thing sounds amazing, and it's depressing to learn that it can be so comprehensively bettered with something I could never afford in this lifetime. 24,000€ - sheesh!

 

But then, on reflection, positive thoughts have re-emerged. One of the tangible outcomes of this thread is when manufacturers pick up on the DIY findings and start delivering products with these findings baked in. This has happened with audiophile switches, USB regenerators, reference clocks and others. But for some reason, this was not the case with music servers and endpoints. The SGM Extreme is the singular product whose philosophy of using extreme compute resources to run an extremely light workload matches our DIY findings. This is something to celebrate, as to me it represents a validation that the approach is sound. I hasten to add that I am not claiming that our ideas drove Emile - indeed it may have been the other way around.

 

The challenge now is to achieve anything even approaching the the Extreme using DIY. Of course, the other ray of hope is that Emile has indicated he wants to bring out follow-on products at a lower price point, which is classic trickle-down at work. Maybe these will be priced at more achievable levels.

 

OK - I feel better now.

It's all for good. Taiko raised the bar. We have some new goals to reach. New challenges. It started to get a little boring here with everyone doing the same builds with NUCs or the AsRock/Asus gaming motherboards and not a lot of fresh ideas.

Actually credit about new ideas goes to @jean-michel6 for pointing out the Engineered board. I haven't tried it but looks very interesting, especially for integration inside the DAC. There are some other new interesting products coming on the market too. The only problem I see is our hobby just got a lot more expensive . 

 

4 hours ago, ray-dude said:

ordered an Extreme

Congratulations! I am sure you made the right decision.

 

4 hours ago, dminches said:

The ASUS mobo is too large for the UP10.

That's a bummer! Thanks for letting me know. I have been looking for the internal dimensions of the UP10, but there is very little info. It did look quite short indeed. One of the members here was planning to get one, and I asked him for more pictures and dimensions. I am not in a hurry, though. Maybe something else will pop up on the market in the next few months. I can probably modify an existing chassis for this. It just creates more work and probably would not end up as good looking as the UP10. Memory was my bigger concern, as that's not something I can do myself. 

[Nenon]

Looking at the HdPlex H5 Gen 2 case in front of me and doing some measurements:

Internal dimensions are: 360 mm x ~350 mm (339 mm by specs, but there is some extra space)

Taiko Extreme motherboard: 305 mm x 330 mm

 

The motherboard would fit inside. I would just need to drill a couple more holes and add additional standoffs. Easy!

 

One CPU should fit with the stock cooling pipes. The other one would need longer pipes to go to the other heatsink. I think that's doable and actually looks much easier than I thought. Of course, I would not know for sure until I have the actual motherboard, but it looks good on paper at least.

 

There would be very little space left for power supply.

[Nenon]

I feel like my message was misunderstood. I tried to explain a couple of times. But let me say it one more time. I am not trying to clone the Taiko Extreme. I cannot even if I wanted to. I just want to try some new components. 

 

1 hour ago, RickyV said:

With DIY we can only do so much, we can buy sr-7’s for every individual rail, we can buy apacer wt ram etc. But what Taiko does is probably more.

Sure, he does! Probably a lot more than we know about. And what we know about is already unreachable by us.

 

1 hour ago, RickyV said:

The ram they use we can not buy.

No, we cannot. But we can buy the custom order Apacer is doing for me. It's wide temperature industrial RAM, similar to the ECC RAM that sounds so much better than commercial grade. I don't know how that would compare to Taiko's custom RAM, but that is the best we can buy, and I would not be surprised if it is as good as any. And so far, it sounds like a one-time opportunity. There is no indication that Apacer would be producing more of it, but you never know. 

 

1 hour ago, RickyV said:

The mobo is it standard or is it modified for Taiko?

Standard.

 

1 hour ago, RickyV said:

Then there is the s.w. Taiko programs the two CPU’s as a server and endpoint configuration, can we do that?

No, we cannot. But I am sure romaz will pop a USB stick with Euphony on his Extreme and get an idea how much worse or better it is. This beast probably needs completely different type of tweaking than NUCs or low powered CPUs. If it sounds good right off the bat, I would be happy to put my effort into software / BIOS tweaks rather than building hardware. 

 

1 hour ago, RickyV said:

They probably have measured all sorts of latency, EMI an noise levels we can not do.

I am sure they have. We don't have their resources. Can't even dream of that...

 

1 hour ago, RickyV said:

The only thing we can do is buy the best parts we can get our hands on and build pc’s. But maybe we have come to a point where that is not enough anymore and have to start measuring (or something).

You are absolutely right on both points. Emile has said multiple times that he is doing a lot of measurements. For the rest of us - it's a trial and error, some common sense combined with a little experience, and hopefully some luck with the trials and less errors. 

 

Again - no hopes that I will build an Extreme. I cannot! Not even trying to. The idea is just to try some new components. I am not aware of anyone here done tests with two powerful Intel Xeon CPUs, industrial RAM, and good quality power supplies. If we were to start from somewhere, we need to pick a motherboard. To reduce the trial and error risks, why not pick the same motherboard the Extreme is using? Emile said he built his own custom motherboard, and this one sounded better. Sounds like a safe bet to me. I am also looking at a couple other motherboards.  And this whole project may change quite a bit by the time I get to it, which is slated for end of the summer as of right now. 

 

The triggering point was the Apacer wide temp, industrial RDIMM memory. That would also be good for other Intel Xeon CPUs / motherboards. You can't use regular DIMMs in multiprocessor configurations, and Apacer does not make industrial RDIMM/LRDIMM. This is my opportunity to obtain some. To be honest, I am kind of surprised for the lack of interest of this memory. I expected there would be some interest. 

 

I've already done several low-powered CPU builds and several Intel Core i9 builds. I have been perfecting my AMD Ryzen 7 3700x build for quite some time, and this is still ongoing - with the full size ASUS ROG ATX motherboard and motherboard OCXO clocks (I posted about), I think I would get a taste of the best I can squeeze out of such AMD Ryzen build. And after that, while I am enjoying my AMD-based build, I will try an Intel Xeon configuration. This lineup makes perfect sense to me :). Hopefully that would be a good learning experience, and other people would benefit from it as well.

[Nenon]

55 minutes ago, ASRMichael said:

I’m really looking forward to you testing the Asus WSC621E Sage with to CPU’s. 

Early stage questions? 

I will try to answer some of these questions. But let me start with a disclaimer again that I am not trying to build a Taiko Extreme. The engineering level and the attention to details on the Extreme is way beyond anything I can do. I am repeating this in case someone sees this message out of context and misses my previous explanations. It's fun to test new stuff, that's why it's a hobby. Some things work, some don't, but we gain experience during that process. 

 

Just now, ASRMichael said:

- Are the two CPU’s powered independently? E.g. could be powered by 2 x Sean Jacobs 5a LPS? 

There are two dedicated EPS connectors, so yes they can be powered by 2 x Sean Jacobs 5a LPS. His 10A LPS would be even better for this. That's what I am planning to use - 2 x 10A rails. BTW, that's not officially in production yet, but he has made some units. 

 

45 minutes ago, ASRMichael said:

- With 2 CPU’s does this mean it can run two operating systems at one time? E.g. can run Stylus on one & run Stylus EP on another?

No, you cannot do that. You can only use one OS. But having 2 CPUs with 20 or 24 cores (40 or 48 threads) gives you a lot of flexibility to isolate processes into their own CPU / cores / threads.

Per Taiko:

Quote

The 2 cpu’s and 12 memory modules are not shared but split into 2 domains, 1 dedicated to the operating system, 1 dedicated to music related services and playback software.

 

 

47 minutes ago, ASRMichael said:

- With so many cores/threads realistically it can isolate most if not all Euphony processes?

That's the idea. Although on a server like that, it's probably worth not to assume that Euphony would sound best and experiment with other operating systems and players too. As you know, I am a big fan of Euphony, and that is what I am using now, but I am open minded. Definitely worth exploring some other options again. 

 

50 minutes ago, ASRMichael said:

- the other question is? Why does the T Extreme have so much RAM?

I have some speculations but no good explanations. Emile has provided some limited info but he won't reveal his secrets, and I don't blame him for that. 

 

Per Emile:

Quote

Addressing memory modules:

One of the largest disturbances on server music playback quality we found is DRAM refresh. The charge on DRAM memory cells weakens over time therefor it needs to be refreshed in pre-set time intervals. During the refresh period all memory operations halt. Memory current draw peaks during refresh, a single memory dim can easily draw 2.5A or higher during the refresh cycle. This is the main reason for DIY community feedback that fewer, or lower speed, dimms sound better, which can be true except you're seriously limiting bandwidth as you have more then just a single memory channel in modern high performance systems.

With the EVO we started focussing on reducing the amount of refresh intervals and refresh period time with quite significant advances in playback fidelity. For this to work you need a higher then average selection of memory chip quality, close chip matching and keep temperatures as low as possible. For the EVO we bought virtually all available consumer level dimms and selected those in which we could get the lowest memory refresh rate / length. As it turned out quality varies from dimm to dimm, so on top of that we started cherry picking the best of the bunch.

For the Extreme we took this a few steps further, these dimms have factory pre-selected A-grade memory modules, all dimm components (memory chips, capacitors, resistors etc) are matched to within a 1% tolerance and selected for low current draw. The side effect of this is refresh rate burst current is reduced by almost 50% and operating temperatures are much lower allowing lower refresh rates. Combined with power supply filtering this allows us to use 12 memory dimms with a very high combined bandwidth at very low refresh noise levels.

Source: https://www.whatsbestforum.com/threads/taiko-audio-sgm-extreme-the-crème-de-la-crème.27433/post-556159 

[Nenon]

1 hour ago, seeteeyou said:

Both Buffalo BS-GS2016/A and Melco S100 seemed to look pretty much the same when we're comparing the PCB inside each unit, unfortunately the clock itself couldn't be seen without removing the heat sink

 

So you are thinking that Melco used the Buffalo BS-GS2016/A with better power supply and maybe a better clock?

 

That would make the Buffalo BS-GS2016/A switch an excellent choice for tweaking. 

Do you happen to know the difference between BS-GS2016/A and BS-GS2016 (without the /A part)? The BS-GS2016 model is $210 on Amazon. Hmmmm.... not that I have spare time for more projects, but that is quite intriguing . 

 

2 hours ago, seeteeyou said:

Since music servers with Xeon / Core i9 / Ryzen would tend to perform so much matter when compared to similar options with Pentium / Celeron / Atom, just wondering if a network bridge with Asus WS C621E SAGE + dual Xeon Scalable + Apacer wide temperature RDIMM + modified JCAT NET Card FEMTO were gonna totally smoke all switches out there

 

https://www.windowscentral.com/how-set-and-manage-network-bridge-connection-windows-10

How about EuphonyBridge  ? 

I have been thinking about this too, but I think networking is completely different than processing music. So what works best in one may not work best for the other. I personally would not go that route though. But I am up for modding a Buffalo switch. I even have a 25Mhz PinkFaun ultraOCXO clock. 

[Nenon]

On 2/12/2020 at 12:34 PM, seeteeyou said:

 

Both Buffalo BS-GS2016/A and Melco S100 seemed to look pretty much the same when we're comparing the PCB inside each unit, unfortunately the clock itself couldn't be seen without removing the heat sink as shown in the second link

 

https://www.hiendy.com/hififorum/forum.php?mod=viewthread&tid=107184

http://my-hiend.com/wp/wp-content/uploads/2020/01/DSC_5221mm.jpg

 

If that heat sink were actually removed, we couldn't really put it back on without messing with the thermal paste and therefore I'd rather stay from that.

 

I'm expecting that both Melco N100 and S100 would share a very similar type of clock from NDK since that should be a good fit under the heat sink of S100, either picture linked below is showing that 24MHz NDK clock for NEC / Renasys µPD720201 where "X17" could be found

 

http://my-hiend.com/wp/wp-content/uploads/2019/04/melco_n100_002.jpg

http://my-hiend.com/wp/wp-content/uploads/2019/04/melco_n100_006.jpg

 

I also have D10 and it's the big brother of D100 as shown in the following picture, 24MHz for NEC / Renasys µPD720210 on the left and 30MHz for µPD720231 on the right respectively

 

http://my-hiend.com/wp/wp-content/uploads/2020/01/DSC_5221mm.jpg

 

They might have scratched the surface of the clock and there's no way to identify the model. However, NZ2520SD looked like a potential candidate to me

 

https://www.ndk.com/en/products/search/clock/index.html

https://www.ndk.com/en/products/search/clock/1190904_1433.html

 

Nowhere close to the performance of megabuck OCXO products like NDK DuCULoN or even MSB Galaxy, though it's still a pretty darn good deal for such an affordable option

 

https://www.head-fi.org/threads/lks-audio-mh-da003.745032/page-20#post-12775721

https://www.arrow.com/en/products/nz2520sd-25.000000m-nsa3449e/ndk

 

It might be somewhat challenging to put Fidelity Audio C5 inside S100 because of that heat sink mentioned above, it's also requiring its own PSU and therefore I'd rather leave S100 as is.

 

OTOH, HFS1155 from JS PC Audio could be another avenue since it's so easy to replace the clock

 

https://www.whatsbestforum.com/threads/sotm-snh-10g-network-switches-x2-the-new-king-of-usb-network-gadget-setup.27758/page-21#post-625621

 

Dimensions of C5 are 100mm × 55mm versus 230mm × 180mm for HFS1155, that Connor Winfield accepts either 5V or 3.3V but C5 needs at least 6.5V to operate

 

http://www.conwin.com/datasheets/cx/cx259.pdf

https://www.fidelityaudio.co.uk/c5-signature-low-jitter-clock-4461-p.asp

 

Of course there's also this 25MHz clock that costs as much as S100 itself

 

https://www.pinkfaun.com/shop/clock/67-ultra-ocxo.html

https://www.pinkfaun.com/shop/sheets/PinkFaunUltra25MHz.pdf

 

However good sounding S100 would seem to be, there's still a lingering question for me. All audiophile switches still have low power CPUs while some supervisor modules from Cisco are much more powerful

 

https://www.cisco.com/c/en/us/products/switches/nexus-7000-series-switches/models-comparison.html

 

Since music servers with Xeon / Core i9 / Ryzen would tend to perform so much matter when compared to similar options with Pentium / Celeron / Atom, just wondering if a network bridge with Asus WS C621E SAGE + dual Xeon Scalable + Apacer wide temperature RDIMM + modified JCAT NET Card FEMTO were gonna totally smoke all switches out there

 

https://www.windowscentral.com/how-set-and-manage-network-bridge-connection-windows-10

 

More about this great finding from @seeteeyou. I got one of these switches and powered it with a LPS. Haven't done any listening tests yet. As already mentioned the Buffalo switch looks identical to the Melco. The layout of the board is exactly the same. Every visible component is exactly the same (i.e. the Melco does not seem to have any special/upgraded "audio-grade" parts). None of that is surprising since Buffalo and Melco are related. 

 

It's super easy to remove the SMPS and install a 12V DC connector. Then you can power the Buffalo switch with a better power supply than the Melco. Of course the Melco switch is in a much nicer chassis, which I am sure is beneficial. The only question that remains open is have they actually replaced the clock or not? It is strange that I could not find any mention of a clock from Melco or any reviewers. I am starting to wonder if they have even done that. 

 

Here is how the Buffalo switch looks like with the heatsink removed.

 

I wish I had a Melco switch to peak under the heatsink. It's super easy to remove but there is thermal paste between the chip and the heatsink. I am glad I removed it, because the thermal paste was of pretty bad quality and not applied properly. I put some fresh Thermal Grizzly Kryonaut thermal paste on mine. 

 

 

[Nenon]

1 hour ago, Dutch said:

Hi Nenon, do you mean you got the Buffalo BS-GS2016/A (or the non /A version perhaps)?

The non /A version but according to one of the posts above it is the same switch with 3 changes:

-LLDP disabled (I always do that anyway)

-Hostname changed

-Half of the ports disabled

and comes with a better Lan cable.

 

 

[Nenon]

7 minutes ago, RickyV said:

Any ideas on its sound yet?

 

Not yet. I would like to let it burn in for at least 200 hours before I do any critical listening. Initial impressions (a quick 2 minute A/B) were that it gave something and it took something away. That was after working for 20 hours only, so I decided to let it burn-in first. In the meantime I am trying to find if the Melco uses a special clock or not. 

 

But I am prepared to replace the clock and will most likely try that regardless of the Melco outcome. 

[Nenon]

I expected to see something here about @romaz latest report on another forum. He reported that the inexpensive Monoprice SlimRun USB 3.0 extender inserted between your DAC and computer sounds better than any USB cable without it. It seems like an easy thing to try for those using USB interface. YMMV (for the moderator @austinpop  - oh, that may not be needed here, actually )

 

Credit goes to @romaz

 

 

 

 

https://www.whatsbestforum.com/threads/taiko-audio-sgm-extreme-the-crème-de-la-crème.27433/post-631854

 

[Nenon]

6 minutes ago, seeteeyou said:

Question for @Nenon since you're planning to build something that's somewhat similar to SGM Extreme. IIRC you're having better luck with Asus motherboards than other choices, did you take a look at their KRPA-U16 for AMD EPYC Rome 7002 Series by any chance? Ryzen / EPYC should be good for your Pink Faun I2S Bridge while having 16 DIMM slots could support 8 memory channels instead of 6 with Xeon Silver 4114

 

No, I did not. And the main reason for that is that processing power is not the only factor. 

An Intel i9-9900K in the ASRock miniITX gaming motherboard sounds different than the same CPU in the ASUS ROG miniITX motherboard, which sounds different than the AMD Ryzen 7 3700X in a ASUS ROG miniTX motherboard, which sound different that the same CPU in the AMD ASUS ROG ATX motherboard. There are almost infinite combinations one can try. Okay, perhaps not an infinite, but large enough number of combinations to spend years of R&D. 

I wanted to start from somewhere, and felt like the motherboard Emile picked for the Taiko Extreme motherboard is a good start. He said he designed his own custom motherboard that did not sound better than this one. Sounds like a good bet to me.  I would probably go with the Xeon Silver 4214 instead of the Xeon Silver 4114 - more cores and more cache. 

 

As far as the I2S bridge with ultraOCXO clock - I am actually planning to use USB with this future build and have been working on modifying the USB interface on my DAC to improve the USB. 

 

I have said that before, but the interface between the music streamer and DAC is one of the most important places that needs special attention. I am a purist and don't necessarily like the idea of generating USB signal from the computer to convert it to I2S inside the DAC. Every conversion like that has the potential to impact the sound quality negatively. Hence, I have been using I2S.

I2S has a lot of problems too and needs an extreme level of attention to details to sound good (that's a very long topic). But having control over my server and my DAC I managed to optimize both to work together really nicely. And as of right now, I2S sounds better than USB in my system. But my DAC is better optimized for I2S input... that's why it sounds better.

I have learned a lot of new things lately - some stuff from the Cerat Kassandra II Ref DAC, others from power supply design and computer design. And I have a very optimistic plan to try to make USB to sound better. Time will show. 

[Nenon]

2 hours ago, dminches said:

Have you picked a case that will fit the ASUS sage mobo?

 

 

5 minutes ago, Johnseye said:

 

I believe it will fit in the HDPlex H5 but @Nenon can say for sure.

 

Yes, that's correct. I am pretty sure it would fit, although I haven't tried it yet. 

[Nenon]

2 hours ago, Dev said:

 

WS C621E Sage is EEB Form Factor - 12 inch x 13 inch

 

For comparison:

ATX Form Factor - 12.0 inch x 9.29 inch

Extended ATX Form Factor - 12 inch x 10.9 inch

 

Also Sage is a dual socket mobo. If you populate both, you need to figure out the cooling and routing heat pipes on the H5 case.

 

 

Planning to use both heatsinks, one for each CPU. 

[Nenon]

22 hours ago, ASRMichael said:

Hi, I thought I was finished today with my network & then I  seen your post “attenuators” thought what is this? I’m new to fiber (today)
 

My setup, do I need attenuators?

 

Edge Router sfp——-- MGB-TLX Planet Mini GBIC LX Module 20km (-40 to 75 C), (both ends)——-1m OM3 Aqua Fibre Optic LC LC Duplex MM 50 125 Patch LSZH Cable——-EtherRegen sfp

 

Any advise would be appreciated?

 

There are so many combinations one can try. If you have the time to experiment, by all means do it ... and please report back your findings. If not, here is what works really good with my etherREGEN:

- PlanetTech MGB-TLX transceivers.

- No attenuators.

- Single mode fiber cable, 10 meters or more. I use 10m even if I only need 1m.

 

That works pretty good for me, and although I have about 50 transceivers I can try, I am resisting that at the moment in favor of other more important tests. 

 

Honestly, the ethernet cable makes a much bigger difference in my system.