- Part 1 - Introduction and Digital Audio Optimization Foundations (Link)
- Part 2 - Enter the Extreme (Link)
- Part 3 - First Impressions and Basic Configuration
- Part 4 - Tweaking Up the Extreme (Link)
- Part 5 - Extended Listening Impressions, Learnings, and Conclusions (Link)
In Part 1 I shared some of my experiences and learnings in digital audio optimization and how that foundation brought me to the Taiko Audio SGM Extreme, a true Summit-Fi digital music server from Taiko Guru Emile Bok.
In Part 2, I reviewed in detail the very impressive specifications, build quality, and design details for the Extreme, and how all the pieces come together for a world class digital music server.
In this Part 3 of my review of the Taiko Audio SGM Extreme, I fire the beast up for a first listen, and look at the impact of various out of the box system configurations and settings.
Out of the gate, I wanted to see what the Extreme could do with no special tweaks or advantages (generic USB cable, cheap ethernet cable going through a couple cheap switches to my router, plugged into a power strip and a random power outlet with all my other AV equipment, no vibration isolation, etc). I wanted to establish a baseline, then incrementally introduce system optimizations to see what impact they have. I was also keen on hearing how a “stock” Extreme with no OCD advantages compares with my reference NUC music server setup.
After cajoling my daughter and partner out of their rooms with chocolate, I imposed on them to help me lift the Extreme into its initial home:
From left to right:
- Voxativ 9.87 speakers with 4D drivers (104dB sensitivity) sitting on a Voxativ Pi bass units
- A stock NUC7i7DNKE in standard case with consumer RAM running Audiolinux in RAM, with Roon server on an Optane drive
- Paul Hynes SR4 power supply powering a SOtM tX-USBultra Special Edition USB regenerator at 12V
- An Uptone Audio UltraCap LPS 1.2 power supply powering an (not pictured) Uptone Audio EtherREGEN ethernet regenerator, with Cat 8 cables to the server and endpoint NUCs
- A battery powered (PowerAdd Pilot Pro2) NUC7i7DNKE in a fanless Akasa case with wide temperature industrial Apacer RAM running Euphony in RAM (sometimes running HQPlayer Embedded connecting to my NUC Roon server, and sometimes Stylus + HQPlayer in a one box solution), booted from an Optane drive
- A Monoprice USB SlimRun optical USB extender cable and Uptone Audio USPCB connecting my endpoint NUC to a SOtM tX-USBultra Special Edition, and a Sablon Audio 2020 USB cable connecting the SOtM tX-USBultra Special Edition to my Chord DAVE DAC
- A Chord DAVE DAC on a Taiko Audio Daiza panzerholz vibration isolation base (more on those later)
- A very intimidating and imposing Taiko Audio SGM Extreme (on a Pottery Barn towel) with generic power, USB, and cheap Cat 5 network connectivity to the generic switch and wall ethernet for my media console
- My audio equipment is powered by a pair of dedicated 30A circuits (one for DAC and Voxativ bass units, one for NUC digital chain), each with their own Topaz ultra low capacitance isolation transformers, with Pangea AC-9 and AC-14 power cords (the LPS 1.2 is always mains disconnected, so I have its SMPS charging supply plugged into the normal house power outlet)
- Back and forth between Voxativ Ampeggio speaker cables and Iconoclast SPTPC speaker cables directly connecting the Chord DAVE to the Voxativ 9.87’s
- In my bedroom closet, I have AT&T fiber to my home (TIDAL for streaming), and a Mac Mini (stock) as a file server with my music library
My Roon server NUC has some local content on the Optane drive, and mounts my music library from my Mac Mini file server. I configured HQPlayer with the sinc-M filter, upsampling to PCM to 16fs (705/768kHz), mimicking what I had when I had a Chord Blu2 then a Chord Hugo mScaler.
First Impressions – Straight out of the box
After powering on the unit for a couple hours, I connected it to my DAC and quickly set up Roon for TIDAL (no music library, no HQPlayer, Redbook quality lossless direct to DAVE) just to do a quick test of whether things were working.
Right off the bat, I was struck by the remarkable presence and clarity I was hearing, even without the benefit of 1M tap upscaling. This is by far the best I’ve ever heard DAVE without mScaler, and is in many ways superior to what I’m used to hearing even with 1M tap upscaling (for background on Chord’s Hugo mScaler upsampler, see Rajiv’s excellent review of the Chord TT2 and Hugo mScaler combo).
The sense of speed and precision is unbelievable, even without the timing and transient benefits I’m accustomed to with 1M taps upsampling. There is a tangible physicality to music, particularly from voices and string instruments. Edges are just faster and much more controlled than I’m accustomed to, which is making everything feel so present and in the room.
Thankfully, this isn’t some sort of false sense of detail. In the past I’ve noted that certain kinds of RF and digital noise can cause a false sense of detail at first list, but that quickly gives way to fatigue. That’s definitely not the case here. Music reproduction is so neutral and unencumbered, hyper hyper clean, and without that characteristic sense of digital haze or stress that I associate with RF or digital noise. Soundstage imaging is similarly unstressed, compelling, and comfortably present. Speed and control of the thwack on percussion is quite remarkable.
Shifting to some favorite performances on shoddy recordings, Shamas-Ud-Doha Bader-Ud-Doja (from “Shahen-Shah” by Nusrat Fateh Ali Khan) is absolutely vivid and present and (incredibly) real! Dynamics off the chart, making any noise in the recording just not matter any more. A glorious performance channeling the angels, brought to vivid tangible life...wow!
Going to some test tracks that I’ve leaned into in the past when doing network optimizations, there is a visceral physical bass sensation on Diana Krall’s The Boulevard of Broken Dreams that I’ve only heard in my NUC reference system with the Uptone Audio EtherREGEN in the chain. Hearing it with no network optimizations at all is a first for me, and makes me wonder how far the Extreme can go with some TLC on the network side.
Hearing how remarkable “Boulevard of Broken Dreams” sounded, I just had to revisit some old favorites. So What on Miles Davis’ “Kind of Blue” brought Bill Evans into the room! From what I’m hearing, I will need to spend a lot of time getting to know Bill Evans all over again. Well no time like the present...it’s Waltz for Debby time! (from the Village Vanguard sessions) Scott LaFaro’s bass is incredibly present, subtle, and confident. I’m almost teary hearing LaFaro play in this way.
Back to another favorite recording, but one where I’m always on edge waiting for playback to break – Fields of Gold by Eva Cassidy (“Nightbird”). Wonderful stuff - that sense of dynamic compression or approaching clipping is WAY diminished. Stressful passages are relaxed and natural. I keep waiting for the recording to break, but it just doesn’t.
Alas, what was going to be a quick 10 minute sound check quickly turned into 2 hours of kid-in-the-candy-store time. Alas2, it is a weeknight and school night, so time to call it a night.
With the weekend upon us, the first order of business was to get the Extreme properly configured.
After brushing off some Windows command line cobwebs that I thought I would never have to revisit, I kicked off a robocopy (robocopy \\SERVERNAME\media\music d: /MIR) of my music directory on my Mac Mini media server to the local D drive. 6 hours and 4TB and 40,000 songs later, it was time to do a restore of my Roon backup. A quick redirection of audio storage in my restored Roon backup to point at the D drive on the Extreme, and I was ready to move on to HQPlayer Desktop setup.
Upsampling with HQPlayer Desktop
HQPlayer Desktop installation was straight forward (Taiko does offer to install and configure HQPlayer for you if requested). I installed my license and brought over my favorite settings from my NUC (sinc-M with LNS15 and 16fs target for PCM, sinc-M with DSD7 256+fs with DSD256 target for DSD), and selected the Chord ASIO driver for HQPlayer. I configured an HQPlayer audio device in Roon Server on the Extreme for “localhost” and I was cooking with gas. My final HQPlayer settings on my Extreme are below (more on the HQPlayer tweaking later).
As expected, HQPlayer took an already glorious sound and brought back all that mScaled goodness I was missing. For those not familiar with HQPlayer filters, my preference is for the sinc-M filter as being closest to the million tap sinc reconstruction filter in the Hugo mScaler. The 15th order noise shaper to my ear is not quite as engaging as the proprietary WTA filter in the Hugo mScaler, but it is pretty close and very high quality.
I listen almost exclusively to PCM content (I have found that Chord DACs are able to do much more magic with PCM than DSD). The vast majority of my music is PCM, but when possible, I always try to purchase music in the format it was originally recorded to, so I do have a smattering of DSD in my library. Rather than convert DSD content to PCM in Roon or HQPlayer, I have a preference for DSD content being upsampled to DSD256 in HQPlayer, and letting my Chord DAVE do the DSD to PCM conversion internally. My preferred HQPlayer DSD upsampling settings are pictured above.
One thing to note: when HQPlayer Desktop starts on the Extreme, a page pops up showing that it doesn’t seem to see one of the Extreme CPUs (see above). In checking with Jussi Laako, developer of HQPlayer, apparently this is a Windows-specific restriction with multi-socket logical processors and HQPlayer. As a practical matter, this has been a non-issue for me. There is so much horsepower in the one Xeon CPU that I haven’t had an issue with any of the upsampling I’m doing (<1% CPU utilization for PCM upsampling, and <10% CPU utilization for DSD upsampling).
Sound quality with HQPlayer has been outstanding. Except where indicated below, all my listening tests were done with the above HQPlayer settings.
JPLAY ASIO vs Chord ASIO vs WASAPI Driver
As I mentioned earlier, the Extreme comes with the standard Windows WASAPI USB driver, as well as a pre-installed and configured JPLAY ASIO driver. In addition, Emile installed the Chord ASIO driver for my Chord DAVE. So which of these drivers sound best?
Each of these drivers have intrinsic capabilities and a characteristic sound. The WASAPI driver is limited to 24/192 output, so not the best match for PCM content being upscaled to 24/768 by HQPlayer, but perfectly good for standard 16/44 content. JPLAY has no limits on PCM output, but is limited to DSD128 for DSD. The Chord ASIO driver has no limits for PCM or DSD to my Chord DAC.
JPLAY is interesting in that it has a settings screen with a lot of configurable parameters (the default I received with my Extreme is shown above). For my listening tests, I did not adjust any of these settings (just don’t have confidence that I understand the OS-level tunings that Emile has implemented).
For my initial driver test, I exited HQPlayer to go direct Extreme to Chord DAVE (no upsampling) so I could give the WASAPI driver a fair shake vs JPLAY ASIO and Chord ASIO drivers. To eliminate network influences, I listened to local copies of music on the Extreme (FLAC or ALAC 16/44 content, with some FLAC 24/192 content).
The WASAPI driver was extremely nice, especially compared to what I’m accustomed to pre Extreme (even non-upsampled, this is the best music I’ve ever heard from my system). However the WASAPI driver was consistently slower and flatter and less dynamic than the JPLAY and Chord ASIO drivers.
For example, on the remarkable violin solo at the beginning of movement 4 of Reiner’s Scheherazade, there is a presence and physicality to the violin that is simply next level with the Chord and JPLAY drivers (JPLAY more so than Chord). Each start and stop of the bow is tangible, and the subtlety and intonation of the performance is incredibly vivid and real. When the rest of the orchestra joins in, the Chord ASIO driver gives remarkable detail and transparency to each of the sections and even individual players, but with the JPLAY driver, I am able to hear for the first time individual performances in the orchestra sections.
That remarkable sense of standing next to Maestro Reiner with the orchestra before you is absolutely intoxicating. I thought I had heard all this remarkable recording and performance could give, but there is so much more beauty and subtlety in these performances to explore and absorb (and to be absorbed by).
With recordings like Organ Prelude (“JS Bach Magnificat” by Dunedin Consort) and O thou tellest good tidings (“Handel: Messiah” by Dunedin Consort) and Fecit potentiam (“Arnesen: Magnificat” by TrondheimSolistene), the sense of the space and depth is remarkable. The cathedral is an integral part of the performance, and the sense of space and being within it is intoxicating.
Switching to my favorite depth test Four Surround Voices (“The Ultimate Headphone Demonstration Disk” by Chesky), the Chord and JPLAY drivers both have a significantly better sense of depth (including well behind the listener...this is true holographic surround sound), but with the JPLAY driver that sense of space was a touch more relaxed and real.
While transparency is extraordinary with the WASAPI driver, the Chord and JPLAY drivers just take things to a completely different level. I’ve never heard this level of transparency and physicality, especially for complex performances. The WASAPI driver is the best I’ve ever heard (pre Extreme), but it is outclassed by the Chord ASIO and JPLAY ASIO drivers in every possible way (absolutely no contest).
If I were to reflect on what I am hearing, my impression of the change is very similar to what I experienced with my NUC music servers as I introduced RAM boot, and significantly increased the size of the memory buffers for Squeezelite and the ALSA sound driver. Even the WASAPI driver on the Extreme is well beyond what I heard on my NUC, but the sound improvements have a strikingly similar character as I go from WASAPI to Chord to JPLAY. If WASAPI is a 3 (on a scale of 1-10), I would put Chord at 8 and JPLAY at 9. Both significantly better than the WASPI driver, but the JPLAY slightly outpacing the Chord to my ear (mainly in transparency). Consistently, the Chord ASIO driver opened up the dimensionality of the recording (holographic) with remarkable spatial resolution and subtlety, and the JPLAY driver taking things to simply the next level (just effortless to be in the space of the performance).
Putting away the WASAPI driver for the remainder of my testing, I then switched to Chord ASIO vs JPLAY ASIO with HQPlayer running (you can select the driver from the Device drop down in the Out Device Settings section of the HQPlayer settings above).
With HQPlayer, the overall level of detail and sense of space is taken up another order of magnitude. All that amazing sense of speed and coherence and detail that those of us that have been addicted to mScaled music on Chord DACs have come to love is now elevating what I’m hearing to an incredible place. Even what I was hearing with no upsampling was well beyond anything I’ve ever heard in my system, but now things are elevated even more still.
The JPLAY advantage for transparency, more relaxed detail, and naturalness is amplified even more with HQPlayer in the chain, at least with PCM content. Switching to some DSD test tracks, I have a slight preference for the HQPlayer with the Chord ASIO driver as output. I can’t say for sure, but I suspect this is due to the limitation of the JPLAY ASIO driver to DSD128 output.
For those that are doing similar tinkering with their Extreme, please remember that these are three drivers talking to the same device (WASAPI, JPLAY ASIO, Chord ASIO), and two applications talking to these drivers (Roon and HQPlayer). There are lots of opportunities when rapidly swapping back and forth to get into a state where you have conflicts that block being able to connect to your DAC. If you get stuck, disable audio devices in Roon, quit HQPlayer, quit Roon Server, and bring things back up.
Based on what I’m hearing, I’ve settled on JPLAY in HQPlayer as my primary listening audio device in Roon for PCM content. When listening to DSD content, I switch HQPlayer to use the Chord ASIO driver. In this configuration, I play music exclusively through Roon to HQPlayer, but change the settings in HQPlayer between JPLAY (PCM content) and the Chord ASIO driver (DSD content) as needed, although I’m delighted to listen to either JPLAY or Chord ASIO drivers in HQPlayer for either PCM or DSD content (differences are small).
Choice of Music Location
So how about the choice of where to put your music library? As I mentioned previously, my Extreme has 8TB of local PCIe M.2 storage, so I was able to copy my entire 4TB music library from my Mac Mini media server to the Extreme. This allows me to compare the exact same source files local on the Extreme, vs via an SMB share on my home network from my Mac Mini, vs (what I hope is) the same content streaming from Tidal (unoptimized network configuration in all cases).
With PCM content, HQPlayer, the JPLAY driver, and an unoptimized copper ethernet connection, I found that local content on the Extreme had consistently and notably better dynamics and speed than the exact same content over the network from my Mac Mini server.
Where I could find Tidal content that I was confident was from the same mastering as my redbook content (my content was typically ripped from CD for these tests), TIDAL was consistently a significant step down in dynamics and speed vs what I was hearing locally or over my LAN. That being said, I obviously can’t control for different masterings on TIDAL vs what I have locally on my systems, but the impact was consistent across the tracks that I auditioned.
A reminder that the above impressions were using a completely unoptimized network. I will revisit these tests in the context of a highly optimized network and internet connection in Part 4 of this review.
One Box Extreme Solution vs Extreme as Roon Endpoint
For this test, I closed HQPlayer on the Extreme and played local content using the Extreme Roon Server (one box solution), and compared Extreme as the world’s most expensive Roon endpoint, with music being played with local content from my NUC Roon Server (generic NUC running Audiolinux in memory).
The network path between the NUC and the Extreme for these tests was: NUC -> Cable Matters Cat 8 ethernet cable -> EtherREGEN (LPS 1.2 powered) -> generic Cat 5 -> generic switch (SMPS powered) -> generic Cat 5 -> Extreme.
Interestingly, the one box solution on the Extreme had notably better dynamics and speed than the same content being played from a NUC-based Roon server to the Extreme as a Roon endpoint (the Extreme advantage isn’t just an endpoint phenomenon). Compared to the one box Extreme solution playing content from my Mac Mini over the network to the Extreme, using the NUC as the Roon server with local content did have better dynamics and speed. Clearly, the Mac Mini in my test setup can not keep up with local content, whether directly to the Extreme, or indirectly through a NUC Roon server to the Extreme. This test was interesting though to give a qualitative sense how much music storage vs Roon server vs Roon endpoint contributes to the aggregate performance of the one box solution on the Extreme.
Note: If you do try this at home, a reminder to disable the audio endpoint on your non-Extreme Roon server when you are done. As I was experimenting with this configuration, I created some instabilities where I intermittently couldn’t use the JPLAY driver on the Extreme (“Unable to connect to audio device”). It turned out that on my NUC Roon server, I still had the JPLAY driver Roon endpoint on the Extreme selected (but not being used). Apparently, that was enough to occasionally lock up the JPLAY driver on my Extreme and prevent me from connecting to the audio device via JPLAY. Disabling the endpoint on the NUC Roon server cleaned everything up.
Stock Extreme vs Reference NUCs
When I first started this evaluation of the Extreme, I had anticipated spending considerable time doing a detailed comparison between the Extreme and my reference NUC system. Alas, that plan only lasted a couple minutes before I realized how meaningless it would be.
There is simply no comparison to be drawn. As much as I adore my NUC setup and as much joy and true wonder it has given me, there is simply no aspect of it (other than size and price) that is notable when compared to the Extreme. The Extreme, in all audio aspects, is orders of magnitude beyond what I had thought was a pretty damn good system.
So all this goodness and wonder is straight out of the box. How does the Extreme scale as you give attention to power and network and USB and other ins and outs? It turns out it scales remarkably well, which I will dive into in Part 4 (LINK) of this review.
Community Star Ratings and Reviews
I encourage those who have experience with the Taiko Audio SGM Extreme to leave a star rating and quick review on our new Polestar platform.