Building a DIY Music Server

[ASRMichael]

To Hyper Thread or to not for non upsampling.,
 

Many users do recommend HT but some suggest to deactivate it. 

Anyone want to explain the for & against? 

[Topk]

3 hours ago, guiltyboxswapper said:

Firstly I own the X570 Aorus Master and coming from an Intel based solution (granted a few gens old at that point) your description doesn't match with my own experience tbh. 

 

What your comparisons  did remind me of is when I introduced something which at first seemed worse than before, but it often turned out there was a bottleneck elsewhere that when rectified, then reintroduced the same change made a lot more sense.  But this is DIY and what works for one may not work for another, so take it as it is!

When you say your experience doesn’t match mine, do you mind giving more details on what is your experience i.e. what did you compare with what?

 

Especially since you also mention that the prior solution you compared with was a “few gen old” which would then introduce many different variables.

 

More details on the MB, CPU, hifi system used would be useful to provide additional context and assess this additional data point.

 

[guiltyboxswapper]

19 minutes ago, ASRMichael said:

To Hyper Thread or to not for non upsampling.,
 

Many users do recommend HT but some suggest to deactivate it. 

Anyone want to explain the for & against? 

 

I found an improvement with hyper-threading enabled believe it or not!  Ofcourse you can check for yourself.  With HT disabled if using HQPlayer, OpenMP (the underlying library that does the work distribution) cannot recognise the CPU correctly (for AMDs at least) and therefore cannot internally pin process/thread affinity correctly.   An error appears in HQPlayer's logs related to cpu affinity failing.

 

What this looks like?  Open up top or htop,  set the view to all CPU cores (if top, press 1), and play any filter/modulator choice you like (apart from sinc-* and xtr-mp which distribute the work across all cores anyway), and you'll find that the workloads appears to move between the cores constantly.  Re-enable HT and and retry and you should see the workload "pin" to certain cores instead.  

 

What this resulted in my case?  For a while I had HT disabled and preferred filters that used all the cores, namely xtr-mp.  After discovering this, I now switch between short-mp and xtr-mp depending on the music source.

[Topk]

33 minutes ago, ASRMichael said:

To Hyper Thread or to not for non upsampling.,
 

Many users do recommend HT but some suggest to deactivate it. 

Anyone want to explain the for & against? 

 

Generally speaking, HT is not recommended for SQ when not upsampling. It appears HT might add some layer of processing by creating the logical cores which might be detrimental to SQ. Same with AMD for SMT. But it might be dependent on the OS you use and it’s better to test for yourself and see which one you prefer in your system, your OS, use case etc. I tested on my system with Euphony and disabled SMT.

 

When it comes to upsampling, it appears that  the benefits of using HT outweigh the drawbacks. @Streamfidelity did recommend enabling HT for upsampling and should be able to provide more details. 

[guiltyboxswapper]

12 minutes ago, Topk said:

More details on the MB, CPU, hifi system used would be useful to provide additional context and assess this additional data point.

 

I'll PM as to not pollute.  If you can, try it (x570) again with the HDPlex 500w.  

[Topk]

3 minutes ago, guiltyboxswapper said:

I'll PM as to not pollute.  If you can, try it (x570) again with the HDPlex 500w.  

I don’t think it’s polluting at all, it’s sharing findings with all so that we progress our understanding of audio PC  :-)

[guiltyboxswapper]

1 minute ago, Topk said:

I don’t think it’s polluting at all, it’s sharing findings with all so that we progress our understanding of audio PC  :-)

Kind of, but I have to admit, some systems just will have a different result to others.  For example I've gone rather extreme on the vibration control, so much so my networking gear (apart from the Solarflare cards themselves) are 20m+ out the way and also on vibration control!  There's a degree of "take it with a pinch of salt, and just make whatever work for you" going on here, rather than science IMHO.  Which is fine.

[Topk]

17 minutes ago, guiltyboxswapper said:

Kind of, but I have to admit, some systems just will have a different result to others.  For example I've gone rather extreme on the vibration control, so much so my networking gear (apart from the Solarflare cards themselves) are 20m+ out the way and also on vibration control!  There's a degree of "take it with a pinch of salt, and just make whatever work for you" going on here, rather than science IMHO.  Which is fine.

Completely agree. All disclaimers apply! There’s so much we can do as a community. 

[StreamFidelity]

32 minutes ago, Topk said:

@Streamfidelity did recommend enabling HT for upsampling and should be able to provide more details. 

 

It depends on the purpose. As some have already written, it is highly recommended for upsampling with the HQPlayer. Hyperthreading does not increase the computing power as some think. Instead, the parallel processing is distributed over more (virtual) cores.

 

The manufacturer also recommends hyperthreading for the Solarflare SFN8522 NIC to reduce latencies.

[MarcelNL]

I think there ought to be common denominators that apply to all, but if a system (or the builder) is not yet at a stage where it (he/she) is able to reveal/hear the effects it may not be showing.

 

[guiltyboxswapper]

54 minutes ago, MarcelNL said:

I think there ought to be common denominators that apply to all, but if a system (or the builder) is not yet at a stage where it (he/she) is able to reveal/hear the effects it may not be showing.

 

TBH I think we need to remain open to the fact the result maybe subjectively observed correctly and perhaps different to one another, rather than assume "their system isn't good enough".   It's guesswork, not really a science.

For example I notice virtually no references to vibration control in most people's systems (here or otherwise) especially when it comes to one of the major offenders (the speakers themselves) yet it can really change the end result dramatically.   Same with room treatments (I have quite a bit, but again, dont refer to it).

[MarcelNL]

Oh sure, I was not implying subjectivity is not involved, just trying to say that when all would be equal the results probably are quite similar as they are based on some fundamental differences.

[guiltyboxswapper]

Just now, MarcelNL said:

Oh sure, I was not implying subjectivity is not involved, just trying to say that when all would be equal the results probably are quite similar as they are based on some fundamental differences.

Kind of- lets take a motherboard for example.  I buy one 6m ago of a certain model.  You buy the same one as me 6m later.  

 

In the meantime the manufacturer:

- updated the firmware between the 2 (e.g. different defaults, microcode updates)
- changed build of material components

- also used a different PCB manufacturer for a different batch

 

It's quite easily possible that we could hear different things, even if we base things on an equal recipe.  Unless your someone like Taiko who make it their literal business to track every little change, have baselines etc, there's a bit of "fingers crossed" involved even if we all follow the same build recipe.

[MarcelNL]

My POV was a bit wider, f.e. that having a MB with better power handling improves SQ. or that more cores, more capable PCUs improves SQ aspects. It's clearly impossible to compare low level details yet hovering over the details there are some key design principles and I was pointing at those. (or trying to)

[Topk]

Recommendation for the El Fidelity PCIe power filter

This is another item that I didn’t see much discussion about. This time, the report is based on 2 different systems (sample of 2...): mine, and my friend’s. There’s a favorable report from alpha audio as well.
 

I have already  described my system. For context again: my friend has been in the game for many years (left AS a while back once he felt there was nothing else to learn there), his system is a highly tweaked PC audio with Asus / Apacer / 3900x / x570 / custom made silver DC cables / SLC SSDs in a top of the line Synology NAS powered by LPS, 7 LPS all around, fiber optic, ~10 different switches/routers at his home (not all used at the same time obviously) with custom made Ethernet cables, Jcat XEs, DDC, external disciplined clock, custom network settings, tens of thousands of dollars in various cables, balanced isolation transformer combined with custom made power filtering...etc..etc. He tried everything on the market you can think of including ~10 MBs (including supermicro, Asus, Gigabyte...), all the different storage methods including Taiko’s Optane 900P. You name it. Went down very deep into the rabbit hole.
 

Anyway, to cut a long story short, we both use the EF PCIe power filter. There are 2 connector sides so you should try both sides. What it does is add better separation in the sound. Less muddiness. Easier to follow the different strands of the music. There’s a slight change in tonal balance but it’s not objectionable. I tested up to 3 of those stuffed in the MB, now down to 1 on the short side. 

 

Bought and tried ALL the other EF products and that’s the only one I recommend (eg SATA filter, memory filter, even CPU fan filter...etc.).
 

Considering the very low price it’s well worth trying and would enjoy hearing reports from others. This thing looks almost comical, but it makes a positive  change. Hope this helps someone.

 

Usual disclaimers: in those 2 systems, personal preferences, YMMV etc

 

[guiltyboxswapper]

9 minutes ago, MarcelNL said:

My POV was a bit wider, f.e. that having a MB with better power handling improves SQ. or that more cores, more capable PCUs improves SQ aspects. It's clearly impossible to compare low level details yet hovering over the details there are some key design principles and I was pointing at those. (or trying to)

No harm in doing that at all, just sadly in this game every damn detail matters....

[MarcelNL]

@Topk i have looked at the ELfidelity products but I'm not very partial to the types of caps on them so I passed.

Did you try add some decent caps to your EPS and ATX cables? 

[Topk]

26 minutes ago, MarcelNL said:

@Topk i have looked at the ELfidelity products but I'm not very partial to the types of caps on them so I passed.

Did you try add some decent caps to your EPS and ATX cables? 

I didn’t try that but I’m certain my friend did since he tried pretty much everything you can think of. He is using large caps in the mains power path in his custom power filtering (Mundorf): it’s allegedly something that is unique that he has refined over many years. But not on EPS and ATX, he’s using silver cables (not the Mundorf silver/gold from Nenon). I’m using FIS audio (probably some kind of OCC) and plan to try some Ghent.
 

What did you do exactly and how did it impact the sound? 

 

Yes absolutely the EF so called “rainbow” caps look bad. It’s almost revolting to use this product, and in fact I was thinking of modding one of the 2 I don’t use with the best caps I could find. I even looked into Audio note caps. Yes, this EF thing is not supposed to work. The middle ones are bipolar of increasing values of 10, 100 uF... etc (on top of my mind).

 

BUT my philosophy is to listen without prejudice: proof is in the listening result and that overrides everything else. And the EF helps. YMMV

 

Maybe there’s an opportunity to do a similar product with the best audio caps on the market. 

[MarcelNL]

I haven't come around to add caps to the power cables but will do so, I'm doing incremental steps of canges so I can keep track of which is which. I'm using copper Neotech UP OCC for the wiring because it sounds great and the Mundorf silver-gold alloy presents a squeaky ringing in the upper mids.

[Blackmorec]

1 hour ago, MarcelNL said:

I haven't come around to add caps to the power cables but will do so, I'm doing incremental steps of changes so I can keep track of which is which. I'm using copper Neotech UP OCC for the wiring because it sounds great and the Mundorf silver-gold alloy presents a squeaky ringing in the upper mids.

Hi Marcel,  I have both Neotech UPOCC and Mundorf silver/gold DC cables (JSSG360). The Mundorf takes a long time to run in but once it does its incredibly transparent, pure, 3 dimensional and holographic. Not a trace of any ringing or indeed any impurity in the upper mids. The Neotech is a very nice cable but in my system the Mundorf is better by a considerable margin, once its been in the system 2-3 months. 

[MarcelNL]

Thanks for that insight!

I'll make another EPS cable with the Mundorf and will burn that in separately, I did listen to the analog interlink I made with the Mundorf wire for 2 months

I can see how you arrive at saying it's 3D  and transparent, I found it very 'forward' in the beginning, it all became more homogenous after 4-6 weeks but in a final comparison after 2 months (constantly burning in) I found it too 'ringy'. Perhaps it fares better as power cable. 

[StreamFidelity]

12 hours ago, Topk said:

Anyway, to cut a long story short, we both use the EF PCIe power filter.

 

I don't understand how it works.

 

Audiophonics writes: The AXF-107 module from Elfidelity is a switch mode power supply filter for computers connected in PCI Express or PCI Express 16x and designed to provide cleaner power to all connected cards (sound card, interfaces, GPU ...).

 

My questions:

 

1. Is the PCIe card simply plugged into a free slot and "sucks" EMI interference?

2. How is the power consumption?

3. Are there any recommendations as to which slot to use?

[Topk]

45 minutes ago, StreamFidelity said:

 

I don't understand how it works.

 

Audiophonics writes: The AXF-107 module from Elfidelity is a switch mode power supply filter for computers connected in PCI Express or PCI Express 16x and designed to provide cleaner power to all connected cards (sound card, interfaces, GPU ...).

 

My questions:

 

1. Is the PCIe card simply plugged into a free slot and "sucks" EMI interference?

2. How is the power consumption?

3. Are there any recommendations as to which slot to use?

1. it’s supposed to suck some noise from the power supply. I am not sure about EMI. 

 2. I don’t have this information

 3. since I already use the 2 direct to CPU (not going through the chipset) PCIe slots with the Jcat cards, I just use whatever PCIe slot left on the MB, physically away from the other cards. The EF card is not hot under operation 

 

I’m wondering if the EF works under the same principles that the Jcat USB enhancer (attached)

 

Another tip about PCIe slots: since the Ryzen CPU I use does not have an internal graphics card, I use a passive graphics cards to make BIOS changes but once I’m done, I remove the graphics card from the MB. The MB boots without it, no problem, and I then run headless with Euphony. Removing the graphic card improves noticeably the sound quality. Less power consumption, less heat, less noise...etc. I assume disabling the internal GPU for a CPU with IGPU *might* have SQ benefits as well, if running with a headless O/S (assuming the BIOS allows to boot without it).

 

In the end, the MB has no graphic card, no chipset fan (or any fan), 1 EF power filter, and no SATA, WiFi, Bluetooth, RGB leds, onboard LAN, USB ports (if not used).., etc: all disabled in the BIOS. Only 1 stick of 4gb RAM. I’m also considering physically removing some of these onboard devices on the MB (eg wifi) like others have.

 

 

[MarcelNL]

as far as I can tell the Elfidelity device is simply a bunch of capacitors stuck on a PCIe circuit board. I assume (always dangerous) that it simply filters some of the HF crud flying around in a PC, you can do the same with any bunch of caps attached to various supply voltages.

I predict the effect is small, and it should be fairly easy to filter HF crud a bit more effectively using different caps in different places but preventing HF crud is always best. Some of the HF will be created by the active elements on the MB itself, anything short of designing a new MB with audio in mind will likely fail to decrease it almost completely.

I expect the Taiko ATX PSU will do a great job to prevent most HF as it is designed to operate at a switching frequency that puts all noise in a very high bandwith that is unlikely to affect other components, except the crud created on the MB. Using a clean PSU and a decent MB combined with as much externally powered add ons will probably be the best solution for the near future.

 

The JCat is also one or a few caps, mica, I can only hope they are silver mica at that price tag, probably connected between ground and +5V. BTW; typically a few smaller sized caps in parallel work best as their ESR goes down.

[StreamFidelity]

39 minutes ago, Topk said:

1. it’s supposed to suck some noise from the power supply.

 

Many Thanks. I'll just try it out now. 👍