Pink Faun 2.16

[ASRMichael]

Hi

 Has anyone here upgraded their MB clocks with Pink Faun OXCO? Really interested in the difference between before & after the upgrade?
 

 

[ASRMichael]

8 hours ago, flkin said:

Yes I have, sometime back.
 

 

Have a look at my notes on that. 

Thanks, I’ve had a look, it seems you are upgrading from standard OXCO clocks. I’m interested in anyone who has upgraded from standard mobo clocks to Pink Faun Ultra clock. Thanks anyway. 

[ASRMichael]

There just example pictures PF sent me to show it the mods would be done. 

[ASRMichael]

6 hours ago, nbpf said:

Is there a chance Pink Faun will come out with a modular board like for example the Allo USBridge Signature but with  OCXO clocks? It would be nice to have a high-quality, low-power board which can be extended with state-of-the-art SPDIF and USB outputs according to needs. The 2.16 is great but it is far too big and too heavy for running very thin applications like mpd, upmpdcli or a Roon endpoint.  

Choice s board and ask Pink faun to install the clocks for you?

[ASRMichael]

11 minutes ago, Armsan said:

Hi all,

Just to update my voyage, (at home!), with this outstanding Pink Faun 2.16 Ultra server, and share with you that finally the Renderer was beaten by the Pink Faun USB Bridge with Ultra clock. Roon is streaming to HQPlayer as output. By suggestion of Jord, I tried HQPlayer with no upsampling at all, PCM and DSD files carried in their native form along the MSB Pro USB which converses the signal to optical and then the Select DAC does his magic. Before trying this approach it seems to make no sense but compared to various levels of upsampling, filters, etc, it gives me the best SQ. Of course this is in my system but give it a try if you haven’t yet, specially if you have a ladder DAC as Jord pointed. Resuming I hear a complete unstressed presentation of the music even more live sense than before. The load of the 16 CPU cores is 2% average.

Happy listening.

Armsan

Always great to hear when someone is happy with their purchase. 
 

Presume you are using AudioLinux? Would be interesting to know how the PF performs using Euphony-HQPlayer. 
 

I have PF USB Ultra clock, love it. Especially powered by Sean Jacobs rail. Once I upgrade my CPU/MB I plan for PF MB clock mods. 

[ASRMichael]

1 hour ago, flkin said:

 

Hi Guys

 

RAM timing and voltage adjustments for better sound

 

An update of the latest changes I've made in my Pink Faun 2.16x that I would like to share. What follows is a bit of old news so many here might have already been there and done that but it's here with added timing details for those searching for info in the future.

 

The PF can take ECC RAM version and so after @k27R pointing me to a supplier, I picked up 2x4G Apacer industrial DDR4, 2400Mhz ECC, unbuffered, unregistered UDIMM RAM to try. This RAM has Samsung industrial high temperature chips on it which is apparently the key to the performance.

 

 

The SEC on the chip stands for Samsung Electronic Corp and the "I" in the BIRC indicates Industrial specs.

 

 

I choose to use 2 pieces of 4G to take advantage of the Dual Channel Memory mode allowing double data rates on the Pink Faun's ASRock motherboard. For this to work, the RAM must be placed in odd or even slots. I used slots 1 and 3.

The original UDIMMs had cooling fins attached and it is said that lower temperatures allow better sound so I picked up some cheap but fancy cooling fins for them which are currently on the way to me.

Right off the bat, they sounded hard. Not as good as the original G.Skill 3200Mhz ram that came with the Pink Faun. Yes more details and information but presented in an in your face way. It is said that the RAM takes around 100 hours to burn in. But I found it more like 150 hours or so before they settled in. After this without any more adjustments, nice! A step up from the stock RAM. More details, soundstage larger and clearer, slightly thicker image. If it were only this alone, it would be worth the switch for what in hifi is a low price. Turns out that higher speed RAM does not necessarily sound better or have lower latency. For example of the RAM timing I researched, some DDR3/1333MHz has a lower latency when compared to DDR4/3200MHz modules. The clock cycle time might be faster in the DDR4 but the CAS Latency is lower and overall the latency, in nano seconds, is lower.

What follows is for the record of what was tested but is not necessary reading for the results. Please skip ahead to the 3rd batch of tests for the relevant test.

 

With further timing/latency adjustments in the Pink Faun BIOS, much more can be achieved with the Apacer RAM. The adjustment menu appears once you have set the RAM timing manually to 2400MHz. There are about 20 different settings to adjust. After a weekend of Googling and learning, here are some of them that might matter: tCL, tRCD, tRP, tRAS, tRRD, tRC, tWR, tWTR, tFAW, tRD, rRTP, tRFC and tREF. This last one is something that could be important but unfortunately it wasn't possible to test as it didn't appear as adjustable in the BIOS. I found a website that claim that the tREF, the time between data refreshing in the chips, can be extended up to 4 times longer - surely a huge reduction to RAM latency if this were possible. But if this were true, surely it would require far higher RAM voltages to maintain the charge in the volatile memory cells and thus generate a high heat.

 

Also from theory and computer gaming discussions, it seems that RAM latency is most affected by tCL, tRCD and tRFC. But there are some general rules to follow for the adjustment of these three settings. For tCL, tRCD and tRP should be the same value or at most, a single digit less. eg for the Apacer RAM above of CL17 (which is the tCL), the tRCD and tRP can be 17 or 16. So allowable CL-tRCD-tRP settings for CL17 would be 17-17-17 or 17-16-16. Then to test reducing the latency one could then force the RAM to operate at say 16-15-15 and so on.

 

This is where I started my first batch of tests and listened to the sound of RAM with the following settings of CL-tRCD-tRP :

 

Please note that this applies to the Pink Faun 2.16x with the ASRock 370x professional gaming board. Also note that the following should have RAM voltage set to auto (default) as the more extreme settings require higher voltages to work.

 

16-15-15, 15-14-14, 14-13-13, 12-11-11

along with other settings for:

 

tRAS from 39 to 25 (default 39)

tRC from 56 to 49 (default 56)

tRFC from 313 to 260 (default 313)

 

If one wanted to hear an extreme difference RAM adjustments make just try this:

 

CL-12 (default 17)

tRCD-11 (default 17)

tRP-11 (default 17)

tRAS-25 (default 39)

tRC-49 (default 56)

tRFC-260 (default 313)

 

You will hear every bell and whistle from the recording. Details offered in extreme but the soundstage is smaller and tighter somewhat like oversampling  gone wrong. The change in sound is remarkable and far larger than CPU overclocking and underclocking manipulations.

 

After my first batch of tests mainly done to see what settings are possible and the general effect of RAM timing adjustments, for my second batch of tests, I made smaller adjustments and reduced the variables to only between 1 to 3.

 

Out of many tests, I found the following :

 

tRFC/280 offers details and larger soundstage than the extreme setting above.

tRAS/36+tRC/55 has 3D sound.

tRAS/37+tRC/55 was pretty good too especially with HQP

tRAS/36 alone was good for human voices and relaxed in presentation

 

3rd batch of tests

Then I tried a new approach which in hindsight should have been my first. In the third batch tried simply overclocking the RAM set speed and set everything else to Auto.

 

RAM speed set to 2400MHz with manual voltage set to 1.1v - sounds large and loose

2666 @ 1.1v - small and recessed

2800 @ 1.1v - tight, large soundstage, even larger than 2400, detailed, thin

2933 @ 1.1v - couldn’t boot up

2800 @ 1.2v - thick body, louder. Heavy bass, too thick

 

At this stage it became clear that the tuning should be between 1.1 and 1.2 volts

 

2800 @ 1.15v - body ok, bass still heavy, soundstage extremely large

2800 @ 1.13v - bass just right, balanced, detailed, soundstage good, not harsh

 

Just realised that my PF graphics card was installed throughout the tests and needed to be removed for final listening before final conclusions. I have found a long time ago that the PF performs better without the card installed. My findings below are without the card.

 

2800 @ 1.13v - bright, bass not enough

2800 @ 1.15v - better but a touch fat

2800 @ 1.14v - perfect balance!

 

For all this I had been using my preferred Rizen CPU under clocking settings of 2.250GHz @ 0.9 volts when the default was 3.4GHz @ 1.35v. But for the new RAM settings I revisited this again and this time around, I preferred the original setting of 3.4GHz @ 0.9v. The final piece of the puzzle fell into place with this last adjustment allowing high energy treble the last bit of control.

 

CONCLUSION

 

The best setting for the Apacer industrial DDR4-2400 ECC UDIMM  in the Pink Faun 2.16x (2018 version) running Euphony/Stylus in Ramroot is:

1. RAM - overclocked at 2800MHz @ 1.14 volts
2. CPU - 3400GHz @ 0.9 volts

It seems that overclocking the RAM from 2400MHz to 2800MHz makes a good difference but only if the RAM voltage is fine tuned. In my particular setup, I found that 1.14v sounded a best compromise between bass definition and hardness with details. This worked in synergy with the CPU at 3400GHz under powered at 0.9 volts whereas without the Apacer RAM, 2250 GHz sounded better.

 

All said, the switch to industrial Apacer RAM made a large difference in the Pink Faun performance. The change is not subtle and immediately obvious. Soundstage far larger, instruments larger in proportion to the soundstage giving the effect of being closer, more details in the background with clarity within the soundstage positions. Sound is more natural and effortless. With tuning, bass tighter and treble extensions more neutral and accurate. Without BIOS adjustments, the sound was already a step up with only the RAM swap. But after BIOS adjustments, truly a new machine. It is clear that RAM timing/voltage changes make a far larger difference than CPU timing/voltage changes and network mods I've tried.

 

I hope that other Pink Faun users can try my findings and confirm it. Or perhaps expand on my findings?

Fantastic write up. I appreciate the info, must have taking a while to do all these tests. Worth posting this on noval thread also, sure many followers will appreciate also. Cheers 

[ASRMichael]

2 hours ago, flkin said:

 

Hi Guys

 

RAM timing and voltage adjustments for better sound

 

An update of the latest changes I've made in my Pink Faun 2.16x that I would like to share. What follows is a bit of old news so many here might have already been there and done that but it's here with added timing details for those searching for info in the future.

 

The PF can take ECC RAM version and so after @k27R pointing me to a supplier, I picked up 2x4G Apacer industrial DDR4, 2400Mhz ECC, unbuffered, unregistered UDIMM RAM to try. This RAM has Samsung industrial high temperature chips on it which is apparently the key to the performance.

 

 

The SEC on the chip stands for Samsung Electronic Corp and the "I" in the BIRC indicates Industrial specs.

 

 

I choose to use 2 pieces of 4G to take advantage of the Dual Channel Memory mode allowing double data rates on the Pink Faun's ASRock motherboard. For this to work, the RAM must be placed in odd or even slots. I used slots 1 and 3.

The original UDIMMs had cooling fins attached and it is said that lower temperatures allow better sound so I picked up some cheap but fancy cooling fins for them which are currently on the way to me.

 

What RAM/Memory heatsinks are they? Tried googling Ultra but couldn’t find them?