DIY Project High Performance Audio PC with high quality wiring

[Zauurx]

And with these network cards and these settings, what do they give under LatencyMon?
Especially under the "drivers" tab with the DPC by driver? Really effective?
I'm trying to assess the way to go... 🤔
I have just reached the maximum of the optimization of my H3 with Gigabyte card (used purchase).
With the 2 Intel inside ethernet cards (Ultra Durable ™ 25KV ESD and LAN protection against 15KV overloads), I am amazed.

 

This only concerns the OS .. for the hardware and the power supply I am still very far (PC and DC cables...)

[Zauurx]

48 minutes ago, StreamFidelity said:

 

HQPlayer with DSD 256 x 48, Modulator ASDM7EC

 

Thank you for this feedback. I have (almost) the same result without special hardware (Intel eth inboard on gigabyte Z370N).
It was this tab that interested me / It seems difficult to me to go lower :

 

LanencyMon on 5 min.

[Zauurx]

6 hours ago, StreamFidelity said:

 

These are my results at 5 min. 

Very good .. still a little way for me.

Now, I will be able to concentrate on the rest and switch from DC-ATX to 400W linear PSU ATX.
I will try to find the right cables .. at the right price.

By the way, I don't know the sync-mqa-lp filter. Interesting for upsampler from redbook to dsd?

[Zauurx]

Very professional look and design.

[Zauurx]

And for the ATX or CPU cable, in 65w ... it is better to 18AWG (Silver Plated Copper Wire) ?
It is to directly connect my HDPlex 400W-ATX.

[Zauurx]

On 1/23/2021 at 5:28 PM, StreamFidelity said:

- Receive Side Scaling (RSS)
RSS is enabled by default for best networking performance. It is best to set the maximum number of RSS queues to be equal to the maximum number of RSS CPUs.

 

I did it, without having read this post, for my 2 intel cards integrated in my Gigabyte motherboard.
On the other hand, I chose to limit to 1 RSS queue on the entry card and 2 queues that which goes to the NAA player.
HQPlayer uses core 0 and 3 (ASDM7EC). Only core 1/2 and 4/5 handle network interrupts.
Choice validate after LatencyMon tests

 

Name                                            : Ethernet
InterfaceDescription                            : Intel(R) Ethern
Enabled                                         : True
NumberOfReceiveQueues                           : 1
Profile                                         : Conservative
BaseProcessor: [Group:Number]                   : 0:2
MaxProcessor: [Group:Number]                    : 0:2
MaxProcessors                                   : 1
RssProcessorArray: [Group:Number/NUMA Distance] : 0:2/0

Name                                            : USBridge
InterfaceDescription                            : Intel(R) I211 G
Enabled                                         : True
NumberOfReceiveQueues                           : 2
Profile                                         : Conservative
BaseProcessor: [Group:Number]                   : 0:4
MaxProcessor: [Group:Number]                    : 0:5
MaxProcessors                                   : 2
RssProcessorArray: [Group:Number/NUMA Distance] : 0:4/0  0:5/0

 

[Zauurx]

On 3/10/2021 at 8:27 AM, StreamFidelity said:

 

Interesting. This is somewhat in line with Taiko Audio's philosophy of assigning precisely defined tasks to specific cores. Did you do it with Process Lasso?

 

Yes with ProcessLasso but you can do almost the same with taskmanager.

And I use other utilities ..
Have you noticed how the core 0 is always more in demand ?

[Zauurx]

Yes, and it is especially LatencyMon that intrigued me with this imbalance of interruptions between cores.
I tried to distribute but windows itself is not flexible.

[Zauurx]

37 minutes ago, StreamFidelity said:

... "Unfortunately" the direct CPU power supply from a Keces P8 with 12V is still better than the complete power supply from the Taiko ATX. That can change with a better LPS.

Despite the quality of the Taiko module, the direct power supply by the Keces would be superior (12v CPU) ?
Interesting but it minimizes the contribution of the Taiko.

[Zauurx]

But in the end, has the quest for the lowest latency really been evaluated against a quality ethernet card (with a good buffer). These cards consume more power and generate more heat, so is it worth the risk?
How to evaluate the latency with the final streamer? 

[Zauurx]

3 hours ago, StreamFidelity said:

 

I did not manage to deposit the correct link. Presumably our forum has an aversion to audio “science” review. 😂

 

I know KSTR from German-speaking forums. He writes well-founded. I am always grateful when I understand the technical construct. But at the same time humble, because not everything can always be explained with today's technology. My ears always win.

 

The named sample jumps are wanted by the developer (it was written somewhere in the developer blog).

 

Besides some measurement results, did KSTR "hear" the sampled music? There is nothing about that in the report.

 

I also use MinorityClean as a CPU jitter cleaner because it "sounds" better. Everyone as he likes. 😉

Why this test ? You are (like me) an HQplayer user who does it on the fly... with possibly Roon which switches to 64bits if processed upstream (auto level).
MC user as well (MC203/4). 👍

[Zauurx]

7 hours ago, StreamFidelity said:

Thank you all. The X2522 card runs 24h without dropouts and the SQ got even better....

I appreciate your work, your research... elegant solutions.
I admit to being frustrated by the loss of the left radiator by putting a pci card in the HDPlex box.
Here it is used.

To have an improvement of the SQ, a minimum latency is desirable.
But below a certain level, is the game worth the price?
The Solarflare solution is excellent, elusive but it is quite possible with the same environment and a JCAT Net card to go below 1 ms (and down to 0.05 for TCPIP or Ndis drivers).
It's also a problem of buffer management, software and windows optimization.

Yes, better hardware with better clocks, better power supplies will do better... with a cost !!
A simpler scheme (direct connection to the streamer in IPV6 without going through a switch) with less hardware to optimise, a simple second-hand JCAT Net card and fewer problems to solve.
Of course my i5 gets hotter in DSD 256 / gauss-xla and ASMD7EC... but it does it! 😋

[Zauurx]

@Vivling2021, thanks for sharing information... inventive solutions.
But if I understand the schematic, it's to link the audio server under Euphony, directly connected in USB to the dac.
In the end, it is the USB driver that provides the SQ via the JCAT USB card.
But the quality of the audio processing depends on the latency of Euphony.
This is not a server/client scheme (?)

@StreamFidelity, your stream goes back through a Zyxel switch (with store-and-forward buffer) ?
The final SQ also depends on the switch, its power supply and its link to the streamer.

I left out the switches to test the direct connection to my Allo (which also defines the final SQ).
This solution is not sponsored by Jussy but it works.
I send DSD 256 / gauss-xla or PCM 1.412 to the Pegasus via a USBridge SIG under GentooPlayer.
The server under win10 LTSC is otpimized... as much as I can with the chosen hardware (20 windows processes... for the whole, between 34/37 depending on the set of programs and a driver  latency spread over the 6 cores between 0,03 and 0,07)... and yes, the hardest to stabilize is ndis.sys !
Sometimes, it drifts ! 🤪
Then the ethernet flow is sent in IPV6 without intermediary to the streamer. 
I also chose (listening) a mix of ethernet cable with a MED1005 isolator in the middle.

 

In the end the only switch is the server with the dual JCAT Net card (IPV4 in, IPV6 out)

[Zauurx]

2 hours ago, StreamFidelity said:

In my opinion, a switch benefits the most from
- an improved clock
- a linear power supply
- a galvanic separation

 

This cannot be achieved with a standard switch.

Thank you for this information.

I'm agree.
I also tweaked one of my two Meraki's and then simplified... for now!
I have not added a clock and I think that is the element that may justify putting a switch back into my system.
It's been discussed at length on the French forum I follow.
The more experienced are currently exploring Diretta... I'm sticking with my schema for the moment.

 

You tried a direct solution like mine ? 

JCAT Net : Eth1 --> ROON (IPV4)
                 Eth2 --> (IPV4) HQP (IPV6) --> USBridge

 

[Zauurx]

Yes, interesting but I couldn't find a commercial link and price ?

[Zauurx]

You have a Schumann generator ??  😉

[Zauurx]

I am thinking of updating my configuration if my HDPlex V3 box is compatible ?

Roon/HQP server under windows (10/11 or 2022 server)

 

Intel Core i9-11900F (2.5 GHz / 5.2 GHz)
Gigabyte Z590 AORUS PRO AX
G.Skill RipJaws 5 Series Noir 32 Go (2x 16 Go) DDR4 3200 MHz CL14

 

Or any other proposal for this CPU which seems interesting for its power and PTD.

[Zauurx]

I'm not sure I agree with the explanation (tag task) but I agree with the observation.
Other elements (monitoring, latency) lead me to believe that Roon upstream stabilises the process.
It acts like an HQP client with a different interface and now, certainly optimised buffer management.

 

And yes, have a lifetime license. 😉

[Zauurx]

7 hours ago, Patatorz said:

Any more information around a release date, price and specifications ? Seems nothing on sotm website 

For specifications, here :

https://www.sotm-audio.com/sotmwp/english/portfolio-item/smb-q370/

And manual :
https://docs.sotm-audio.com/lib/exe/fetch.php?media=manuals:smb-q370_productguide_v1.0b.pdf

 

But no price... 

[Zauurx]

Yes, but the limitations of this card are also related to the cpu :

Support for 9th and 8th Generation Intel® CoreTM processors

Not necessarily the last one but 8/9... it's not 10... and even less 12.
At least it's compatible with my i5 8400 (8th generation) but I'm not ready to do DSD256 with 7ECv2 !

The idea is to benefit from the latest generations.

For cpu, lastest is often best

An i9-10900, mentioned above, allows DSD 256 7ECv2 (i think).

It's still fanless.
So a well powered Aorus MB (or equivalent) is certainly my future choice.

 

For the DDR, I agree but at SOTM it is limited to 2666Mhz

[Zauurx]

11 minutes ago, Patatorz said:

DSD 256 7ECv2 Is making such difference with HQP that you need such power ? With which DAC ?
what do you mean by « well powered » ?

The best is the enemy of the good.
A French expression that would shut down all audiophile forums.
It's mostly the DSD256 that works very well on my system with a downward extension (vs DSD128).
But a little more (7ECv2) gets us closer to the graal !! 😉

 

At the moment, my happiness :

(I can't do better with the i5 8400)