Jump to content
IGNORED

Building a DIY Music Server


Recommended Posts

13 minutes ago, Topk said:

The next test to validate the core/power theory would be to move up in cores within the 5 series and assess impact on SQ. but it does seem to indicate the 5 series might sound better than the 3 series.
 

I also want to mention that my test was with euphony, not HQ player, and without upsampling. 

I think we'd simply have to try tbh, but this logically still kills the more cores = better for me, as surely even with improvements in the 5000 series the less cores would result in worse sound?    

 

The chiplets (each 4 core unit) on the 5900x are rated at the same clock speeds as the 5800x (each 4 core unit) which suggests on paper they're using the same grade, but there's more cache on the chip too so who knows.

 

I think HQplayer and also active cooling will change the results considerably,  I've seen the same thing with running a slow active fan against a LT3045 10x PSU (considerably better SQ with fan). 

 

Would be interesting if you guys with passive cooling can try some fan cooling outside the chassis, possibly followed by higher vcores. 

 

Also the Keces PSU you use is rated to 8amps and as posted lately the HDPlex 500w offers 2x 12.5a outputs, which when both plugged into the CPU EPS resulted in a big unexpected uptick in dynamics.

Link to post
Share on other sites
29 minutes ago, guiltyboxswapper said:

I think we'd simply have to try tbh, but this logically still kills the more cores = better for me, as surely even with improvements in the 5000 series the less cores would result in worse sound?    

 

The chiplets (each 4 core unit) on the 5900x are rated at the same clock speeds as the 5800x (each 4 core unit) which suggests on paper they're using the same grade, but there's more cache on the chip too so who knows.

 

I think HQplayer and also active cooling will change the results considerably,  I've seen the same thing with running a slow active fan against a LT3045 10x PSU (considerably better SQ with fan). 

 

Would be interesting if you guys with passive cooling can try some fan cooling outside the chassis, possibly followed by higher vcores. 

 

Also the Keces PSU you use is rated to 8amps and as posted lately the HDPlex 500w offers 2x 12.5a outputs, which when both plugged into the CPU EPS resulted in a big unexpected uptick in dynamics.

Our responses crossed. I happen to have the HDplex 500w on the way so I can compare soon vs Keces P8.

 

As for the core theory, I am making the distinction between CPU cores/cache and inherent CPU sound.

 

It appears that the inherent CPU sound / performance gains of the 5 series give them a 4 core equivalent advantage vs the 3 series. However if we move up in cores within the 5 series, will the additional number of cores and additional cache make the sound even better? My prediction is that it will. Unless we have reached the point of diminishing returns. It might also be OS dependent (since euphony likes power and fully leverages all the cores) and system dependent (need a revealing system to hear the difference)

 

As for the 1v voltage and 3.3 locked core speed, the 3900x runs at 3.6 stock speed so while the decrease is small, it is a decrease in speed that could have impacted SQ and also reduces the ability of the CPU to boost. On the workload, all those cores are sitting mostly unused or with very little and very short use (without upsampling), so my theory is that the SQ boost has to do with process latency. But who knows.


I’m interested in your fan and cooling system: what exactly do you do and why SQ improves in your opinion? Passive guys tend to avoid anything that can add acoustic/electric noise and vibrations due to theoretical/potential impact on components and clocks. That includes fans, water circulation etc. Maybe to a fault?

Link to post
Share on other sites
14 minutes ago, Topk said:

It might also be OS dependent (since euphony likes power) and system dependent (need a revealing system to hear the difference)

I don't think its a system issue per say here, as mentioned I can tell the difference between a single DC wire being changed upstream on the network with ease.  The 5800x lowers the power usage considerably too (10%+) running the same workload which in theory should work to its advantage.

 

23 minutes ago, Topk said:

It appears that the inherent CPU sound / performance gains of the 5 series give them a 4 core equivalent advantage vs the 3 series. However if we move up in cores within the 5 series, will the additional number of cores and additional cache make the sound even better?

 

The 5000 series is a refinement not a big step change up according the AMD themselves, so I'm not honestly expecting the 5900x to be much better tbh, but will report back eventually if I can get my hands on it.  I honestly don't think this is about core counts anymore, but some other factor that we aren't aware of.

 

15 minutes ago, Topk said:

I’m interested in your fan and cooling system: what exactly do you do and why SQ improves in your opinion? Passive guys tend to avoid anything that can add acoustic and electric noise and vibrations. 

Simply put, running hardware comfortably within its specification thermally rather than letting it run to its limits can in some cases outweigh the vibrations from the fan itself. This will vary from hardware to hardware.  

Link to post
Share on other sites

Slightly off topic, but for context for non-Euphony users, here is how my cores are used in Euphony while playing music. I do have very short bursts (mostly 1% on the player on 1 core, sometimes larger % on the OS), it just happened to be all zero when the screenshot was made which is how it looks like most of the time because the workload is spread over 16 cores.

 

B24DEB23-9249-455B-BB68-A7B24C023CA8.png

Link to post
Share on other sites
41 minutes ago, Topk said:

As for the 1v voltage and 3.3 locked core speed, the 3900x runs at 3.6 stock speed so while the decrease is small, it is a decrease in speed that could have impacted SQ and also reduces the ability of the CPU to boost. On the workload, all those cores are sitting mostly unused or with very little and very short use (without upsampling), so my theory is that the SQ boost has to do with process latency. But who knows.

 

I'm no longer convinced ultra low latency is all that important either, but well behaving well orchestrated processes (i.e. harmony).  Many love Windows 10 LTSC and that is not a real-time kernel by any stretch of the imagination. 

Link to post
Share on other sites
23 hours ago, Topk said:

First of all, 10900K is 125W TDP, that’s 30% above what HDplex recommends for H5. You have low temps but the 10900k does need power. It can pull up to 225W if I recall correctly.

 

Right now with 3950x, my CPU is un-throttled for speed within the limits of the core voltage which is set manually at a pretty high 1.25V (I’ll explain what I mean by that).

 

All the rest including Precision Boost Overdrive, Asus performance enhancement, CPU clock speed are all on Auto. Which means the CPU is free to boost frequency as needed (well

above 4 GHz) as long as core voltage is no more than 1.25V. And based on Linus Tech Tips tests and others, the 3950x runs around a very low 1V voltage under most conditions / benchmarking tests. Also, keep in mind music is not a CPU intensive task (without upsampling). It’s not like you’re running Cinebench R20 or Prime95.
 

To cut a long story short, my 3950x is pretty much running full speed in a fanless case....

 

On the other hand, the very powerful 10900K is rated 125W TDP which means it draws a lot of power. Intel was very smart in optimizing with thinner die but it’s still 14nm.
 

If I were you, I would first:

1) run all cores - it’s critical for sound quality

2) pick a decent clock speed - start with the stock speed and manually lock it (manual override in the BIOS). If needed you can decrease later on 

3) disable turbo boost (for now) - that can potentially put strain on your LPS 

4) Decrease CPU core voltage. Start with stock voltage and decrease in 0.25v increments, and check for stability 

 

Another way to do it is manually set TDP (if you can on Intel) to 95W or less and let the BIOS adjust everything else. 


Now - you mention you can’t go past boot. So a few more tips:
 

It could be because your CPU is boosting on the 10 cores and exceeds 10A of power usage during boot. In that case, the locked lower voltage might help. The lower fixed TDP might help. Or if you manually lock the power going to the CPU (PPT for AMD), max peak amperage (EDC for AMD) or max amperage (TDC), that should help too. Under AMD terminology:

 

- Package Power Tracking (“PPT”): The PPT threshold is the allowed socket power consumption permitted across the voltage rails supplying the socket. Applications with high thread counts, and/or “heavy” threads, can encounter PPT limits that can be alleviated with a raised PPT limit.


- Thermal Design Current (“TDC”): The maximum current (amps) that can be delivered by a specific motherboard’s voltage regulator configuration in thermally-constrained scenarios.

 

- Electrical Design Current (“EDC”): The maximum current (amps) that can be delivered by a specific motherboard’s voltage regulator configuration in a peak (“spike”) condition for a short period of time.


i don’t know the intel equivalents of the above but that could fix the issue

 

And I agree hyper threading should be disabled (for sound quality) if you’re not upsampling  but I don’t think it’s the root cause

 

I couldn’t boot in euphony (maybe like you if you use it too) and I could see on the Keces P8 the CPU pulling much more than 8A (which is the max for P8), probably boosting the CPU during boot. When I manually locked the voltage to 1.25V, the issue went away.

 

while you’re at it, disable spread spectrum as well since it measurably creates jitter. 

Valuable information. Thank you. Now running 10 cores & turbo on set at 5.3

 

I set PSU setting. Long 110w & short bursts to 150w. 
 

Didn’t try under voltage. Maybe something to try later.

 

For now it’s a case of touch nothing & sit back with a pint or two! 
 

Cheers

Link to post
Share on other sites
23 minutes ago, ASRMichael said:

Valuable information. Thank you. Now running 10 cores & turbo on set at 5.3

 

I set PSU setting. Long 110w & short bursts to 150w. 
 

Didn’t try under voltage. Maybe something to try later.

 

For now it’s a case of touch nothing & sit back with a pint or two! 
 

Cheers

Victory!!! :-)) enjoy the music!!

Link to post
Share on other sites
4 hours ago, guiltyboxswapper said:

I'm no longer convinced ultra low latency is all that important either, but well behaving well orchestrated processes (i.e. harmony).  Many love Windows 10 LTSC and that is not a real-time kernel by any stretch of the imagination. 

For those who want to try real time (RT) vs non real time O/S, it can be done easily with (free) Daphile which is available with both a RT and non RT kernel. I did try both, and happen to prefer the RT kernel.

 

Windows 10 LTSC has actually a lower process latency than a stock windows 10 since it is a lightweight version with fewer processes than the regular version. And many tools like Jplay, AO, process lasso and fidelizer focus on process latency optimization showing benefits in SQ (this is quite the consensus).

 

I stand corrected on the 3900x base clock: it is 3.8Ghz (not 3.6Ghz) which means than a 3.3ghz clock speed is quite a bit lower than stock. So in the event you compared 3900x with a stock 5800x at 3.8ghz (not undervolted / down clocked) then potentially not apples to apples with a 3900x undervolted at 3.3ghz.

 

Lastly, another data point: the 3 series is on the Zen 2 architecture, and the 5 series on the newer Zen 3 architecture. As a result, even if the 5800x and 3900x run at the same 3.8ghz base clock speed, the 5800x (with half the cache) is about 25-30% faster for single thread performance, showing the benefits of the newer architecture. Therefore if single thread is heavily used by the O/S then 5800x actually performs significantly better than 3900x. Lots of variables. 

Link to post
Share on other sites
56 minutes ago, ASRMichael said:

 

For now it’s a case of touch nothing & sit back with a pint or two! 

Fantastic. Interested in any i9 9900k to 10900k comparisons you can provide.

Link to post
Share on other sites
1 hour ago, Topk said:

I stand corrected on the 3900x base clock: it is 3.8Ghz (not 3.6Ghz) which means than a 3.3ghz clock speed is quite a bit lower than stock. So in the event you compared 3900x with a stock 5800x at 3.8ghz (not undervolted / down clocked) then potentially not apples to apples with a 3900x undervolted at 3.3ghz.

 

3900x and 5800x both at same fixed cpu speeds and both at stock voltages so I could do an apples vs apples, I haven't listened to the 3900x undervolted in over a year, it was that awful. 

 

What would be interesting?  Run your Ryzen 3950x at stock with fans cooling the external heatsinks, and see if its any better or worse. 

 

It's complex though as a Keces 8a (? I think) rail to the 12v CPU EPS socket may start to get saturated or worse not even boot up, so its possible that your hearing the PSU not keeping up rather than the CPU itself at its best.  Do try this with the HDPlex 500w though as it can supply 25a provided your mobo has dual CPU EPS sockets.  I will try undervolt the Ryzen 5800x to 1.1v out of interest mind.

 

1 hour ago, Topk said:

For those who want to try real time (RT) vs non real time O/S, it can be done easily with (free) Daphile which is available with both a RT and non RT kernel. I did try both, and happen to prefer the RT kernel.

 

It'll sound different for sure, but this very forum shows many happy Windows LTSC users.

 

1 hour ago, Topk said:

Windows 10 LTSC has actually a lower process latency than a stock windows 10 since it is a lightweight version with fewer processes than the regular version. And many tools like Jplay, AO, process lasso and fidelizer focus on process latency optimization showing benefits in SQ (this is quite the consensus).

 

Yes, but its not really a true real time kernel - and yet it achieves great SQ (with fiddling).  Proves my point really.  

 

1 hour ago, Topk said:

Lastly, another data point: the 3 series is on the Zen 2 architecture, and the 5 series on the newer Zen 3 architecture. As a result, even if the 5800x and 3900x run at the same 3.8ghz base clock speed, the 5800x (with half the cache) is about 25-30% faster for single thread performance, showing the benefits of the newer architecture. Therefore if single thread is heavily used by the O/S then 5800x actually performs significantly better than 3900x. Lots of variables. 

 

I don't honestly think (especially for PCM streaming) its about performance really.  It does help HQP a lot though (AVX2 improvements).

 

Remember Emile himself went through all sorts of Xeon processors (with a whole range of CPU cores and clock speeds) and accidentally came across the 10 core silver Xeon which for reasons unknown, just sounded the best out of the lot.  Yes, the other CPUs he tried were MUCH faster on single threaded performance too.  And yet it didn't seem to matter.

Link to post
Share on other sites
51 minutes ago, Topk said:

On the performance upgrade side, I did try CPU upgrade with, again, 1 variable change at a time (including going from 12 to 16 cores with all other variables unchanged) so that point specifically can actually be concluded based on actual listening test with a sample of 1 :-) .... with same architecture. And the importance of locking the CPU ratio to stock (or above) in my step by step undervolt process 

 

Would be interested to see if this applies to 5000 series.  Again, was expecting to take the 5800x very soon after plugging it back in.  I think we all need to just try it and see rather than assume, as Emile's own posts show there's no strong correlation we can always see from the outside.  I use a JCAT XE USB with SuperCaps + 10x LT3045 PSU, Sablon USB cable.

Link to post
Share on other sites
1 hour ago, Topk said:

Test of impact of power phases and 2 of the best AMD motherboard (MB) available on the market for sound quality

 

The general consensus is that Asus MBs sound better and are favored by most. 

 

However, I noticed 2 very interesting posts from Energy and Seetoyou on the Gigabyte Aorus Master MB. It’s the first MB on the market that has 16 direct power phases. Power phases/VRM down convert and regulate the 12V and provide power to the CPU and chipset. 
 

Therefore, the idea naturally came that the Aorus Master might be the best MB for audio PC since it has the best power phases (along with 6 layer PCB, 2 oz of copper etc.) therefore benefiting from robust, low ripple, stable, more dynamic power to the CPU. Also high quality VRMs overclock better, generate much less heat etc. This is entirely measurable and all proven via benchmarks. In fact the Aorus master is considered one of the best MB for the AMD Ryzen in the PC world.
 

We also all know the critical importance of power in general since music replay is just power transformed into sound.

 

And there’s quite a consensus that power (and cable) specifically to the CPU is critical to sound quality - and that CPU’s power comes from the VRM.

 

Therefore selecting a MB with the best power phases/VRM absolutely make sense (on paper) - all other things being equal (which they never completely are... but at least we can control for the variables as much as practically possible).

 

However that hypothesis is to be tested as rigorously as possible using only 1 variable change at a time.


Some key decision criteria for the MB, in my opinion:
 

1) Chipset fan: no chipset fan since those generate noise and vibration. For AMD, that excludes most X570 chipset (except Asus Dark Hero). Pretty much all B550 qualify.

 

2) Strong power phases/VRM: 


As noted by Seetoyou, per LTT AMD VRM tier list, best MB on paper WITHOUT chipset fans (using process of elimination) are Aorus Master, Asus dark hero, both tier 1 (300A), and Asus B550e, tier 2 (250A).

 

Test of Asus B550E vs Gigabyte Aorus Master

 

Only the MB was changed. Test was performed with 3950x and - again - only variable is MB change. Burn in period for both MB. Same BIOS setup.


Hypothesis: Aorus master is the best MB vs Asus due to the best power phases/VRM on the market.

 

Result:

 

- Aorus master has excellent PRaT: very sharp and well defined transients, excellent timing, however almost in a mechanical kind of way. Excellent instrument separation as well. Excellent control over all frequencies, no bloom or muddiness. Almost like having an upgraded clock. Excellent for electronic music. However sound is tonally much more grey and bland, more shut in, less open and colorful, less compelling and - dare I say - less analogue and less naturally musical. More homogenized as well.

 

- Asus B550E. As soon as I put the MB back, there’s no question. Much more musical, natural, room filling, tonally rich, colorful, more real sounding.
 

Bottom line: the MB absolutely matters for SQ and in this A/B test, in my system and OS, Asus is the way to go. No question and very surprised by the result. This test made me want to check out the Dark Hero (tier 1 with excellent VRM) but I lost interest since B550E sounds so good.


It is unclear what the impact of power phases are - maybe transients and detail - but it is clear in my system/for my taste (usual disclaimers...) what is the best MB taken as *a package* and between those 2, it’s the Asus B550E.
 

The best one overall for AMD Ryzen for Audio might very well be the Dark Hero but that is untested. 

Good info thanks. I purchased my MB after seeing a few posts here also. I believe my Asus Maximum XII Extreme also has 16 power stages & no doublers. 

Link to post
Share on other sites
14 minutes ago, ASRMichael said:

Good info thanks. I purchased my MB after seeing a few posts here also. I believe my Asus Maximum XII Extreme also has 16 power stages & no doublers. 

Definitely looks like a good one!

Link to post
Share on other sites

My previous notes on using the i9-10900K in the GB Aorus Master- no comparison to another MOBO (yet).

I should add that both CPU and DC-ATX supply 0V wire (GND) run  straight through back to the SR7T to minimize loops.

 

 

Link to post
Share on other sites
1 hour ago, ASRMichael said:

Good info thanks. I purchased my MB after seeing a few posts here also. I believe my Asus Maximum XII Extreme also has 16 power stages & no doublers. 

Can I assume the Asus is called Maximus, not Maximum, or are they different?  If Maximus, is this mobo $1500 and cost $160 to ship??  Yikes!

https://www.newegg.com/asus-rog-maximus-xii-extreme/p/1JW-000C-00U79?quicklink=true

Link to post
Share on other sites
9 minutes ago, ted_b said:

Can I assume the Asus is called Maximus, not Maximum, or are they different?  If Maximus, is this mobo $1500 and cost $160 to ship??  Yikes!

https://www.newegg.com/asus-rog-maximus-xii-extreme/p/1JW-000C-00U79?quicklink=true

Hi yes sorry. Maximus £800 delivered UK. You’re paying extra for thunderbolt & dimm.2 card. 
 

I wanted the Dimm.2 card where you can connect M.2 card to it. This goes straight to CPU. However when I tried it today it worked but my pcie lane 2 stopped working. Pity. 

Link to post
Share on other sites
1 hour ago, bit01 said:

My previous notes on using the i9-10900K in the GB Aorus Master- no comparison to another MOBO (yet).

I should add that both CPU and DC-ATX supply 0V wire (GND) run  straight through back to the SR7T to minimize loops.

 

 

Would be really interested if you compare with a high end Asus MB!
 

Such a test is a huge pain to do (even worse if you have a fanless case) because you need to disassemble the copper heat pipes, remount the MB and CPU, put thermal paste wait for burn in etc. I did it but will not do it again, that’s for sure. If your system reacts like mine (which we don’t know), I would say it’s totally worth trying Asus based on my listening test and would be another data point for everybody.
 

Attached showing the superiority of the unique direct 16 phase (14+2) Aorus Master VRM design over any Asus VRM currently on the market (for the sake of clarity, the middle design is typical of high end Asus, 16 (non direct) phases, no doublers). The bottom one from Asrock has doublers.

 

It’s pretty obvious based on the below why I wanted to test the Gigabyte Aorus Master (first diagram) for SQ.

 

42205193-FD8B-4F87-BD71-2DA4606292D5.png

2327B713-26C3-4B36-982E-343F565EFE53.png

9DE896AC-16E9-46B9-B646-C819D4A6215A.png
 

 

Something else I wanted to add is that I would avoid ITX and sub ATX MB designs in general since they tend to have less robust VRM designs. That most probably outweighs potential benefits such as shorter PCB traces.

Link to post
Share on other sites
1 hour ago, Topk said:

Would be really interested if you compare with a high end Asus MB!
 

Such a test is a huge pain to do (even worse if you have a fanless case)...

 

 

I will agree with that! No rush at all  at the moment. The nice weather is on the way here hopefully. It would probably be a bit later on (likely in early autumn).

Link to post
Share on other sites
On 1/11/2021 at 12:57 PM, guiltyboxswapper said:

I have one of the first HDPlex 500w units, which has proven to be very decent (though I don't have anything else to compare it to that's not SMPS!).  I dont use the LT3045 outputs mind.  

 

One of the biggest surprises however is using both 12v outputs (PCIE and CPU EPS) to feed both ATX12v sockets on my motherboard (x570 + Ryzen 3900x).  Instead of placing most of the CPU 12v load on a single rail, its now spread across the 2 rails which clearly results in a obvious across-the-board uptick in sound quality with HQPe running.  It also means I have a good upto 25amps available to the CPU.

 

And as always for me at least, the shortest cables ATX cables win.

The MB has 2 CPU input connectors: 8 pin and 4 pin. Do you connect the HDplex CPU cable on the 8 pin and then one PCIe cable on the 4 pin? My original plan was to use HDplex 500w on 1 CPU connector and Keces P8 on the other CPU connector. But maybe using the 2 HDplex rails is the way to go. Will probably end up with SR7 or Taiko converter/unregulated DYI combo eventually... depending on feedback. 

Link to post
Share on other sites
12 hours ago, Topk said:

The MB has 2 CPU input connectors: 8 pin and 4 pin. Do you connect the HDplex CPU cable on the 8 pin and then one PCIe cable on the 4 pin? My original plan was to use HDplex 500w on 1 CPU connector and Keces P8 on the other CPU connector. But maybe using the 2 HDplex rails is the way to go. Will probably end up with SR7 or Taiko converter/unregulated DYI combo eventually... depending on feedback. 

That work work fine too, as the 8pin EPS standard is actually 2x 4pin of the same layout, so long as both the 8pin and 4pin are EPS sockets ofcourse.  

 

I used a PCIE to EPS adapter cable, and at some point i'll re-arrange the provided cables to avoid that adaptor.  

 

You may get strange grounding issues introducing 2 PSUs (even if on the same rails) to the motherboard so whilst you can try it, it may not work best. 

Link to post
Share on other sites
16 hours ago, Topk said:

It is unclear what the impact of power phases are - maybe transients and detail - but it is clear in my system/for my taste (usual disclaimers...) what is the best MB taken as *a package* and between those 2, it’s the Asus B550E.

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!

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now


×
×
  • Create New...