Pink Faun 2.16x setup, listening tests and a comparison with Antipodes Cx+Ex
Some 2 months ago, I did a review of my choice to use the Pink Faun 2.16x as my streamer and why I took this path. It can be found here in another thread:
It was a cumulation of a years worth of reading, experimenting and learning about computer audio and what matters in the creation of good sound. I don’t claim to have discovered every aspect and nuance of how to squeeze out every last ounce of audio goodness as some have done on this site, a lot of which I still have yet to understand, let alone master, but I think I understand the basics of what is required for a decent attempt.
And so I put aside my Trifecta with Clock and settled on a one box solution in the form of the Pink Faun 2.16x streamer. It's an extreme build - like a bucket list of all things audio. It’s not a minimal, simple build using low power, simple circuits in the hope that noise and jitter can be controlled this way. On the contrary, it’s a massive 20kg build packed with an 8-core CPU on a gaming board. The design and hardware details are in the above mentioned review and it’s a monster. But oh boy, does it work. As Piero Olmeda, the author of AudioLinux says, The "PinkFaun Streamer 2.16 is probably the best computer for audio on the market". The review is more of discussion of the technology used and the build but doesn't have any listening tests which was to come later.
So as promised, here are my Listening Tests:
Firstly, it takes months for the break-in process to complete. Jord, the proprietor from Triple M Audio Shop, says around 300-500 hours. In my 2.16x unit, it took around 3 months to settle down. This makes earlier comparisons before this time difficult as the sound is still developing.
At the present time, the unit playing through Roon or HQPlayer is wonderful. The character of the 2.16x is body. More body than most other streamers I have used to date. It's not a thick sound but muscular sound. The soundstage is large, full and deep. More about this later.
2.16x through a re-clocked D-Link DSG105 Switch
From my router, I passed the ethernet signal through DSG105 switch with re-clocked signal from my tx-USBultra that was fed a master clock signal from a Mutec Ref 10. I felt little or no change, if anything direct sounds better. Ethernet cables were the SOtM dCBL-Cat7 cables and iSO-Cat6 device in between.
2.16x as a NAA fed by a SonicTransporter i5 (STi5)
Sound good, initially seems better than the 2.16x alone. Jord tells me that his standard setup is to play 2 units of 2.16 together, one as the Roon server and the other as the up-sampler. Apparently the split of duties makes the unit sound better. This is echoed by other companies like the new Antipodes CxEx two box solution (will compare below). Adding the clocked switch mentioned above to the setup makes the sound drop a little and not easy to distinguish the difference between the 2.16x direct and the STi5/2.16x/switch. Overall it's not necessary to use a reclocked switch.
The STi5/2.16x dual box with added clock via the tx-USBultra sounded initially amazing. Remove the clock and the sound is slightly less dynamic, details around the same but slightly more blurry. With the clock, the sound is more forward.
With this base, I did a lot of tests with different DC cables and experiments from mu-metal conduits to POE-DC16 (16 separate wires DC), the Stammhein DC cleaning boards in all variations of 5A, 3A and 2A. Sounds changed from smoother to detailed, to more or less 3D effects, textural changes, clarity affected. But I couldn't conclude a single best setup. There were too many variables.
Returned to the 2.16x to DAC directly for a sound check. Surprisingly how good it sounded but slightly less clarity than what I heard in this multi-box system with clock and SOtM equipment.
as time went on (or perhaps as the 2.16 further broke in), I found it difficult to listen to this setup for a long time without getting a touch tired. As magical as it sounded, dynamic, amazingly detailed, it's not totally natural. Listening to live music, you don't get tired (unless it's a head banging Heavy Metal session in which case you get deaf but still not tired 😄) so there's something not quite right.
Using the Imagine Dragons, Believer track as testing for treble irritation, it a good one to see if the music imparts energy and excitement but without any harshness. With the multi box system, frequencies became a little harder with some of the filters I experimented with in HQPlayer and so I finally ended up not using the STi5 as the Roon server and returned to using the 2.16x only as a single source doing both jobs of playing Roon and upsampling.
I think the 2 box solution warrants further investigation but presently with my STi5, it doesn't match the 2.16x as a source.
SonicTransporter i5 -> SOtM sMS-200(clocked) -> tX-USBultra (clocked Ref10) -> DAC to compare
The sMS-200 was the pre-evo version (before I sent it to SOtM for the change). This chain sounds good but clearly a few steps down from the 2.16x alone.. Everything is less, less details, less focus, less body. Harsher and less delicate. Complex harmonics in the sound missing. Timbre less and emotion too. It's amazing how things can sound so clearly less when not too long ago, I though this was a pretty good setup already.
The power setup in the above chain was the usual mix of sPS-500, LPS-1 with the usual POE-DC cables in all their different guises.
A diversion about Arch Linux OS and Convolution (Room correction):
The 2.16x is made for upsampling. It's body-centric sound begs for tuning and focusing. It's matched well to my Vinnie Rossi LIO DAC2.0 which has a NOS mode, allowing the streamer to make all the changes to the data stream without complications of resampling later on in the DAC. It's a powerful streamer with the AMD Rizen 8-core CPU and 16G of RAM with an optimised OS. Jord has the following to say about the ArchLinux he uses in the 2.16x.
AudioLinux in the 2.16x
" ArchLinux is a version of linux, like Debian or Fedora, which gives the most freedom to the developer to build a custom version. The installation system only offers a minimal base, transparently exposed during system configuration and keeps patching to a minimum, thus avoiding problems that upstream are unable to review later. "
AudioLinux is not a separate Linux distribution but a customized Archlinux image designed for audio developed by Piero Olmeda. The standard version of AudioLinux is already known to be good but Jord commissioned a dedicated version for the 2.16x with better Ryzen 8-core support and further stripped the minimum installation of functions not absolutely need for audio in a bid to reduce latency. This version is not for sale.
The low latency of AudioLinux is pretty spectacular. From Piero's site http://www.tophifi.it, it's mentioned there that the latency is tested in the region of 2.5ms even with HQPlayer playing. The same machine under Windows (probably not optimised) measures around the 500ms mark. The developer of the optimising software Fidelizer Pro for Windows tells me that he is able to achieve around 100ms for "resource fetching at kernel level" which still is two orders of magnitude above what Piero claims.
I understand that AudioLinux RAM OS is already available but not in a commercial state yet. When RAM OS is fully operational and steady it will be available to Pink Faun 2.16 customers on request.
I have long wanted to try to eliminate the room from the audio equation but never got around to it. Finally decided to make the effort do get it done properly. Recommended to me by @lpost, he found convolution useful in his setup and suggested to use a service by Thierry at Home Audio Fidelity https://www.homeaudiofidelity.com/.
After all, what's the point of optimising the sound of audio hardware into a coloured state in order to compensate for room errors? My audio room was built up to Louden room dimensions (Ratio 1:1.4:1.9), properly dampened on certain surfaces and specifically designed non-symetrical corners to avoid standing waves to avoid such problems. Even then, the room had certain node and phase issues causing bass to be muddy (only found this out later at the end of the exercise). After measuring the acoustic characteristics of my audio room and sending it to Thierry, I received back a set of 4 matrix files which I entered into the HQPlayer matrix engine which allowed the software to correct the bass nodes and phase errors.
What a difference! 👍 The listening sweetspot expanded considerably, the music stopped making soundstage shifts when I moved my head slightly and the bass tighened significantly clearing some muddiness from the bass and lower mids.
Convolution matters and it's a necessary step to take for getting the audio system to perform properly be it computer audio or otherwise. The effect is far greater than optimising wires or power supplies or clocks and is one of the greatest advantages of computer audio - the ability to use easily use convolution compared to tradition non-computer music sources that cannot perform this function easily.
For those readers that have used a Audio Video projector system and had their system colour space measured tuned to display Adobe RGB and have suddenly found their picture quality improving substantially becoming more natural and cinema like, it's the same thing but for sound. It's the tuning of the room to neutral sounding so that the equipment can show their true character.
So, that done, the 2.16x was ready to be playing alone and ready for upsampling.
Upsampling with HQPlayer
The upsampling I choose to use is the HQPlayer offered embedded and licensed into my 2.16x. Prior to convolution, I found in the past that HQPlayer upsampling and filters gave strange and shifting results. I would prefer one set of filters for a while and then they wouldn't sound good later. Often I reverted back to the filterless, NOS (non-upsampled sound) that was at least consistent. But after convolution, this all changed. HQPlayer matches really well with the 2.16x and integrates like part of the total solution.
There are so many options and combinations and I have spent months (years actually on other devices) experimenting with many of them. My end result is as follows:
Specific for my 2.16x streamer and downstream equipment (Vinnie Rossi LIO DAC2, VR DHT pre-amp with 300B Takatsuki tubes, YBL Passion 1000 mono blocks, Wilson WP6/7 speakers) my favourite filters are, not in order of preference:
Poly-Sync-Ext2: Slightly laid back, smoother and natural
Closed Form-M: Dynamic, large, detailed, focused and tight
Closed Form: Dynamic, large detailed, natural. Effects are slightly less than the above CF-M
Sinc-M: Balanced and natural
Other filters that I like include:
Poly-Sync-xtr-lp: Highly detailed, pretty good match for the Antipodes CxEx machine
No filter: HQP allows no upsampling with no filter and this unexpectedly sounds better than Roon alone.
Upsampling frequencies are all available in HQPlayer to 705.6K for all the above filters under PCM with auto-rate family checked. Without, the Sync-M and Poly filters can upsample 44.1 to 768K as they are apodizing filters. I tend to like filters up to the 705.6K for the above filters apart from "No Filter" where I select 44.1 only. I tried less upsampling to 192 or double to 88.2 only but prefer 705.
I also played with SDM filters (DSD) but prefer the PCM filters as even with the power of the 8-core CPU, the 2.16x is unable to upsample to 512/48 with some of my preferred filters above. Also for dynamics, it's a little less than the PCM filters so the SDM options may be preferred for smoother sounds.
A quick note about using convolution and upsampling. Convolution uses CPU power and so when combined together, some of the heavier HQP filters cannot be played while convolution is on. Especially so with the SDM filters.
Dither and Modulators that I prefer are :
PCM - Gauss1
SDM - ASDM7
Others that sound pretty good to me in the 2.16x are the TPDF and NS5 for PCM and AMSDM7+512fs for SDM.
Sorry for the detailed and perhaps confusing HQPlayer settings discussion above. Its the nature of the beast and to use it requires a lot of time to experiment. Part of the fun for some.. In summary, HQPlayer is essential for the optimised 2.16x sound. The natural 2.16x sound is with body and after applying convolution, Closed Form M filter, Gauss dither, upsample to PCM 705, it's as optimised as I can get the 2.16x to become. At this setting, listening to the 2.16x is like watching performances on TV. Everything is so clearly placed in front of you and real.
The comparison tests - with the Antipodes Cx and Ex twin box
@Kritpoon ordered his Antipodes twin Cx and Ex box to replace his Dx unit some time back, before I ordered my Pink Faun streamer. Unfortunately due to a series of hardware issues with power and software OS issues, the unit had to be returned to NZ to be adjusted. As with many high end equipment, it's rather sensitive and even the switch on procedure is tricky. Nevertheless, he eventually received good working units and after having burnt them in, he brought two units over one weekend and we connected it up in the following way:
Cx -> dCBL-CAT7 cable -> Ex -> Habst Ultra USBIII -> DAC
Power from Topaz Isolation Transformer into Palic AC-3000 (CX) and DHLabs PowerPlus (EX) AC cables.
Tracks were all from Tidal as we wanted to use the same one on both machines and not from local SSD. Connection to Router was through a dCBL-CAT7 ethernet cable.
My 2.16x was simple with just the Habst Ultra USBIII connecting to the DAC. Power cable was the MIT Z-cord III.
From the Chie Ayado CD, Track "Only You", a very nice live recording of vocal and piano. We used this for our first test with the preferred settings of both streamers first.
For the Antipodes CxEx:
convolution on for room correction
For the 2.16x:
convolution on for room correction
Upsample to PCM 705 with Closed Form M filter, Gauss1 dither
The CxEx soundstage is smaller, less details and slightly laid back. Tried to use upsampling at 192 and also 705. Prefer NOS sound without upsampling. The 2.16x had a larger soundstage with more micro-details and texture. The difference is clear and we switched back and forwards 3-4 times.
Next we used our favourite test track Eric Clapton unplugged, Tears in Heaven to continue. Effectively the same results as with Chie Ayado. The 2.16x has more presence, body of music but still focused and detailed compared to the CxEx with or without upsampling.
We found, to Kripoon's surprise, that the HQPlayer in the Cx was working (embedded trial version) and so we tried to use this and configure the Ex as an NAA instead. Got it to work and found the details of the music increased with widened soundstage. This was with upsampling to 192, but it is still unable to top the 2.16x with the latter's sound more effortless with a less stressed delivery when played at the same sample rate.
At this time, we attempted a 3 box solution, with CX as server, 2.16x as the upsampler and EX as renderer. Couldn't get it to work. my Pink Faun LAN card hadn't arrived yet and being short of a LAN port on the 2.16x couldn't get a USB-LAN adapter to work.
Next up was the CxEx under NOS/HQP, with external Mutec Ref10 clock via a txUSB-ultra. Sounds good, the best the Antipodes had been all night. Tried turning off convolution and in some way it sounded better but there was too much bass from the room as expected. Perhaps the convolution files only worked fully with the 2.16x?
In conclusion to our comparison, the CxEx combo sounded good, highly musical and very good with the clock attached. And perhaps even better without convolution given the constraints of including the room effects.
But it still cannot top the 2.16x when played with convolution and upsampling. Here, every sound is replayed with great control and placed precisely in the soundstage, musical, almost technical especially with the Closed-Form M filter. In comparison, the CxEx in its best configuration NOS/HQP/clock/no convolution sounds musical but slightly looser not as focus. The sound is also slightly duller, less engaging.
Is it fair to bring a high end unit into a room without room filters build to match that exact unit? Perhaps the convolution files of the 2.16x cannot work perfectly in the CxEx since it was generated using the 2.16x or loads the smaller Antipodes CPU too much for it to perform Roon or HQPlayer upsampling together with convolution at it's best? I had plenty of time to optimise the 2.16x to my set of equipment and it shows. Plonking in another set of equipment and keeping the same settings might be said to be unfair. But that's what we had to do given our timeframe and what we found is as described. We didn't have time to change the filters too much to see the effect so we used the ones that sounded best with the PF. That said, the difference between the 2.16x and the CxEx is clear especially when upsampled and filtered with HQP.
Playing the last track in The Little Willies, Nora Jones and Richard Julian singing on the 2.16x is simply amazing. They are in the room with you and their emotion and enjoyment laughing about Lou Reed is fun to listen to. They try to be deadpan about it but fails. ☺️
Character of the Pink Faun 2.16x is body and space and the HQPlayer filters adjust the presentation of this body. In other streamers, HQP tends to make sound thinner. But in the PF, it maintains body but the edges of the image are clearer placed solidly in the soundscape deep, black and wide in 3D with textures more pronounced. This is achieved without an external reference clock attached. It's as good as I have ever heard in my system and far better than all my previous experiments with computer audio to date.
Thanks for reading, hope it was entertaining. 😁 🙏
With kind regards, Kin