Building a DIY Music Server

[JonD]

Many thanks to all of you for the replies, much appreciated 🙂

[StreamFidelity]

11 hours ago, seeteeyou said:

Dual Xeon 4114 will only boot with RDIMM like D51.23244S.001 as shown below, please be cautious and DO NOT get the regular UDIMM

 

I think the latencies are quite high.

 

Apacer [email protected] MHz RAM

1,000 / 2,666 * 19 = 14.25 nanoseconds

 

It can be done better, for example [email protected] MHz RAM

1,000 / 2,400 * 10 = 8.33 nanoseconds (0.00000833 ms)

The 2.400 MHz is not very high, but in combination with the CL value, the latencies are very low.

[JonD]

So I have a 4114 source already. Do i need to stick to another 4114 or can i also pick up a 4210. Looks to be almost the same chip and there are plenty of Intel chips around where the difference in the model number is more about binning than a change in CPU.

[RickyV]

28 minutes ago, StreamFidelity said:

 

I think the latencies are quite high.

 

Apacer [email protected] MHz RAM

1,000 / 2,666 * 19 = 14.25 nanoseconds

 

It can be done better, for example [email protected] MHz RAM

1,000 / 2,400 * 10 = 8.33 nanoseconds (0.00000833 ms)

The 2.400 MHz is not very high, but in combination with the CL value, the latencies are very low.

Back then when somebody, I forgot his name, found out that these wt apacer ram sounded beter we noticed the latency but still it sounded beter. Apperently there is something else besides latency that influences SQ.

[LTG2010]

2 hours ago, RickyV said:

Apperently there is something else besides latency that influences SQ.

It might be the Ram refresh rate, tRF-c, the slower the refresh / longer the cycle, the better the sound. You can adjust this in some bios, but I couldn't find a setting in Sage bios.

[MarcelNL]

It also could be different things, why does Optane sound better than a regular SSD?, is it the material used or the implementation or the latency, frequency? Same question would IMO probably apply for RAM.

[StreamFidelity]

25 minutes ago, MarcelNL said:

why does Optane sound better than a regular SSD?

 

It could also be due to the latencies. See image.

Source: PC Perspective INTEL’S OPTANE DC PERSISTENT MEMORY DIMMS PUSH LATENCY CLOSER TO DRAM

 

I consider the lowest possible latency to be one of the keys to audio PCs.

 

Latencies can also be set in the software. With ASIO drivers (USB), Taiko Audio experimented with the buffer settings to my knowledge.

 

If you are interested, I wrote something about latencies (click on translate): Audio PC latencies

[MarcelNL]

thanks, I can read German! I did read up on computer latency, my take home message is that is likely is important but not THE key driver for many processes. I'll read your article!

 

[seeteeyou]

Indeed, we could go for multivariate analysis (like what Emile did) rather than univariate analysis by focusing on latencies alone

 

https://www.whatsbestforum.com/threads/taiko-audio-sgm-extreme-the-crème-de-la-crème.27433/page-100#post-616762

https://audiophilestyle.com/forums/topic/30376-a-novel-way-to-massively-improve-the-sq-of-computer-audio-streaming/page/616/?tab=comments#comment-1014096  

On 12/16/2019 at 7:37 AM, romaz said:

I had asked Emile why he felt he needed to use 48GB of RAM in the Extreme when this seemed like overkill and would potentially be a significant source of noise:

 

"Well RAM is a topic on its own, to start with, the 2 cpu’s are split into domains (NUMA / SNC), so you really have 2 x 6 dimms, 6 for each CPU, they are not shared. Music services have their own cpu/dimms and the OS has its own cpu/dims. So its sort of a core and endpoint into a single machine going beyond just core allocations for individual processes. These Ram modules are a custom order type, similar to the Apacer types popular in the Jplay forums, but taking it just a bit further. They do create less noise and draw less current then other offerings. If more dimms reduce performance, it typically means your power supply is negatively impacted by the increased current draw. As occupying more memory channels increases bandwidth and reduces wait states, you do get better individual process performance."  "What you really want to do is reduce your hardware active processing times as much as possible. The net effect is much like a class A amplifier, you have a higher baseline power consumption, but power draw does not vary much, and this is very good for a more “natural/relaxed” sound. I hope this makes sense 🙂 But you do need a power supply which is very comfortable supplying the load. You really want the least possible variation in load, and higher cpu power / bandwidth systems are better at that with very low load music playback processes."

 

As for the Extreme being a core and endpoint in a single chassis, this was interesting for me as well.  You basically have 2 CPUs with each CPU having its own dedicated RAM bank (24GB each) and so there is a genuine distribution of tasks between 2 machines just like dual Pink Faun 2.16Xs.  While I very much like what I heard in the Aries Cerat room at Munich this past May where dual 2.16Xs were playing, this configuration costs north of $30k, has fewer cores, uses a noisy SSD, consumes more than 200 watts, and capably functions as a room heater.  I find the Extreme to be a more elegant and practical solution and at least on paper, I believe it is the most technologically advanced music server at this time.  My brief listening experience in Taiko Audio's listening room did nothing to dissuade me from this opinion.  

 

https://audiophilestyle.com/forums/topic/30376-a-novel-way-to-massively-improve-the-sq-of-computer-audio-streaming/page/697/?tab=comments#comment-1075973

On 8/25/2020 at 11:50 PM, lmitche said:

Project goal: Find a replacement for the unobtainable Apacer DDR4 2666mhz UDIMMs that matches or exceeds Apacer SQ. Best case would be another Samsung B-die memory module or whatever variant of Samsung memory chips that replaced B-die.

Latest results:

There are a plethora of Samsung B-die memory modules available from the major memory packaging and distribution companies. Take a look here for the "google" of Samsung B-die.https://benzhaomin.github.io/bdiefinder/

 

I have tested 3200, 3600, and 4133 mhz modules from Team Group and Gskill in the past few weeks. Despite all being B-die, there are differences in SQ between these modules. The more expensive "higher" binned parts do indeed sound best even when running below their maximum rated speeds. All of the above beat the Apacer 2666mhz SQ.

Next steps: Done for now

 

The Corsair DDR4-5000 Vengeance LPX Review: Super-Binned, Super Exclusive

https://www.anandtech.com/show/15089/the-corsair-ddr4-5000-vengeance-lpx-memory-review

Quote

We confirmed with Corsair that these are E-die. Interestingly Corsair is using its 8-layer PCB design which is impressive and is likely a reason why it is capable of reaching DDR4-5000 in a low profile kit.

https://www.anandtech.com/show/15089/the-corsair-ddr4-5000-vengeance-lpx-memory-review/5

Quote

With this kit, Corsair is sending a clear message. The company wants us to know that their custom 8-layer PCB internal screening process and anodised aluminium heat spreaders are capable of producing memory that hits the highest speeds on the market.

 

There's current draw, super-binned parts, Samsung B-die, Micron E-die, 8-layer PCB design, what else?

[MarcelNL]

sounds like a plan, yet hard to pull off as the effects are subjective....

My experience is that the corsair vengeance LPX (not the 5000 but an older 3200 version) I replaced with the Apacer 'unobtanium' sounded 'dirtier' It helped using them with the precooked LPX settings but the Apacer outperformed them sonically.