![]() |
With Raspberry Pi 'Touch' streamer showing album art running RoPieee, and the Topping DX9 VU meters playing, this looks quite nice in the audio rack! |
Let's finish up with the last installment of the review/measurements for the Topping DX9 "15th Anniversary" DAC (see Part I, and Part II previously). As we have seen up to this point, it's quite a unique looking DAC and the initial PCM measurements over USB2.0 look great as a modern hi-resolution converter. In Part II I noted that there was an issue with the 176.4/192kHz frequency response that Topping was able to correct quickly with a firmware update. I'll be using the latest firmware 1.23 for the measurements here.
Let's continue then to dive in and explore the objective performance with some of the other main features for this DAC.
![]() |
Playing "Quad-rate", DSD256 at 11.28MHz. Nice use of dual-screen set-up to show data about the digital stream. |
I. DSD
Let's enter the world of DSD for a little bit.
I assume there are still a number of audiophiles listening to DSD regularly. There are many recordings out there available (for example on sale at NativeDSD.com), but as usual, be mindful of the fact that since native 1-bit DSD is not conducive to anything more than basic editing, the vast majority of DSD at some point had to be edited in multibit-PCM (like 24/352.8kHz "DXD"). Over time, I've added to the list of SACDs that appear to be just upsampled/transcoded PCM and I've seen in recent years DSD256 and DSD512 material that probably originated from hi-res PCM like 24/96 based on the Nyquist roll-off.
Here are some 1kHz THD+N measurements in DSD for your consideration compared to 24/96 PCM at the same output voltage level:
For context, have a look at this previous article on DSD and my "standard" DSD signals.
Those are good DSD 1kHz THD+N results. Using the SoX-created DSD samples with CLANS noise-shaped modulator setting, the sweet spots for DSD performance appear to be DSD128 and DSD256. Notice that as the DSD rate increases, the ultrasonic noise can be pushed further up and thus less likely to have noise energy amplified and sent to tweeters.
Bigger isn't always better as you can see with the DSD512 performance where the harmonic distortion is a bit higher for this device. As is the case with most modern hi-res DACs, there's usually nothing to worry about. Even with the higher THD+N, at DSD512 all harmonic products are still below -115dB and DSD512 playback with the limited tracks I have sounds great.
Internally, the AK4499EQ operates as a multi-bit sigma-delta device (presumably 7-bits like the AK4499EX) which is what PCM input gets converted to inside the DAC. In comparison, DSD is a noisier 1-bit sigma-delta signal that requires quite aggressive noise-shaping to achieve low noise floor within the audible frequencies. It should be no surprise I think that PCM measures better with basically every DAC I've ever measured and compared with to DSD.
Here is the 1/10-Decade Multitone 32 in DSD64 to DSD512 (click on image to zoom):
Beautiful performance across the board. By stacking the FFTs as above, we can see the change in the ultrasonic noise as we increase the DSD rate up to 96kHz. The test signal is reproduced very well for each DSD rate, with at least 119dB noise plus distortion-free range below the peak level.
While the Dunn J-Test is not designed to detect jitter with 1-bit DSD data, we can try anyways. Here's what it looks like for the Topping DX9 - 24-bit J-Test converted to DSD using SoX:
Interesting, we can see some very low level sidebands down around -140dB or less. Over the years, I've generally found minimum issues with the J-Test in DSD form.
II. S/PDIF - Coax and TosLink
Next, let's move on to an examination of the S/PDIF interface and performance. This DAC features 2 coaxial and 2 optical inputs; more than enough I think for older transports like CD spinners and streamers like the Chromecast Audio.
To start, I confirmed again that indeed the new firmware 1.23 discussed as an addendum last time fixed the frequency responses at 176.4kHz and 192kHz which are the maximum sample rates supported for S/PDIF:
Notice that the graph above is with TosLink input. In some older and lower quality DACs, the fiber optic interface might not be able to handle 192kHz. No problems here.
Numerically, let's have a peek at the RightMark results - measured from XLR out for lowest noise level comparing USB, Coax, and TosLink digital inputs:
Most notable is the IMD+N sweep where we can make out distortion for TosLink > Coax > USB. Very tiny differences as we're talking about distortion levels below 20kHz of less than -110dB so nothing that would suggest the ability to differentiate between the digital inputs in blind listening (unblinded listeners can claim all kinds of things).
The thing with S/PDIF is that while it's a convenient unidirectional interface with billions of devices supporting it, from a strictly fidelity point of view, it transmits both the clock and data on a single line which the DAC has to lock onto and try to reject as much of the temporal variability (jitter) as possible. As a result, there is often an increase in S/PDIF jitter which we can see here:
I've been using the old Squeezebox Touch as my S/PDIF coax and TosLink streamer connected to the LMS/Lyrion music server here at home over WiFi for these kinds of tests over the years (as per the hardware picture above). For this test, I used a generic 6' RCA cable as coaxial, and inexpensive 6' glass TosLink cable. As you can see, there's a significant difference between asynchronous USB2.0 compared to the S/PDIF inputs.
While Topping advertises that the AK4118 SPDIF receiver might be the lowest jitter IC, it's clearly more jittery compared to the USB performance. Having said this, are we likely to be able to hear the difference!? No, of course not. The Dunn J-Test is intended to stimulate jitter so we can see/measure it. Real music is not like this test signal. While present, notice that the highest level sidebands are down below -120dB from the 12kHz peak. Furthermore, since jitter is worse at higher frequencies, 12kHz is an octave beyond the most sensitive audible frequencies (around 1-5kHz). Even if jitter anomalies were to show up in actual music >10kHz, it would be extremely unlikely to be noticed due to the combination of low amounts, and reduced human physiological audibility.
Between the coax and TosLink, it's quite typical to see a bit more jitter with the optical TosLink. Notice the slightly higher sidebands in the TosLink tracing.
This graph argues against old-skool audiophile reviewers who complain about the use of the USB interface because it's a computer "printer" interface. 😉 As you can see, even when just plugged into a typical Windows 11 Beelink MiniPC, jitter is better than dedicated audio S/PDIF. Furthermore, there is no evidence on the RightMark results earlier to suggest that USB is any more noisier than S/PDIF in a modern DAC with 24-bit signals.
Alas, I don't have an I2S transport so no test results for that input. Also, I don't have an AES/EBU transport easily available (I guess I could unpack my Squeezebox Transporter with AES out). And while wireless Bluetooth isn't meant for highest fidelity playback, it sounds great and I use it all the time, but as a lossy system with compression based on psychoacoustic techniques, I don't anticipate anything other than the resolution limits shown in previous testing like this.
![]() |
The six NFCA discrete headphone amplifier modules. |
III. Headphone Amplifier
With the higher voltage output, we see clipping using the 20Ω load at 7.6Vrms or around 2.9W. Note that I used a conservative 0.1% THD distortion threshold. At the usual 1% THD+N, this would be 3.1W into 20Ω, <1% THD+N. On the graph, while I only showed THD, the THD+N level is essentially identical to the THD line above 500mV. This result is consistent with Topping's advertised 3.3W x 2 into 16Ω at <1% THD+N.
IV. Preamp Output
V. The AMPT
As I've done with previous DAC reviews, I've recorded a standard test sample for your listening pleasure. See the Archimago's Musical Performance Track (AMPT) discussion a few years back with sample files from other DACs you can also download and compare.
Topping DX9 "15th Anniversary" DAC AMPT (173MB)
24/96 capture through RME ADI-2 Pro ADC
VI. Conclusions
A friend recently wrote to me after seeing the images of the Topping DX9 DAC (Amazon Canada. AliExpress links here, and here, if cheaper from China) in Part I saying that this thing looks "sexy". Indeed, I do not believe I would be in "want" for any better-looking DAC! With the well-built enclosure, tempered glass see-through top panel, dual bright high-res color OLED screens, it is quite striking. When you turn on the top amber LEDs and select the classic-looking VU meters, there's a nice retro look to a device that is clearly of a modern heritage when we peek inside.
Of course, beauty isn't just skin-deep even though let's be honest, the look of luxury hardware is important! There is great "beauty" to be experienced when it comes to the high-fidelity audio. It converts digital audio data with resolution well beyond human audibility when it comes to low noise, low distortion, impeccable channel balance, flat frequency response well beyond audible frequency limits, with a good selection of filters (including NOS and intermediate phase settings). Performance of both the XLR and RCA outputs were excellent; in fact, I was quite impressed by the resolution of the RCA out compared to a few other DACs I've tested.
USB and S/PDIF (coax and TosLink-optical) inputs performed well and essentially were equivalent. Not unexpectedly, the asynchronous USB2.0 input performed better in jitter testing than S/PDIF. There are other devices that will reject jitter better, but there's nothing in the performance here I found to be of audible concern.
Beyond the basic DAC line outputs, we see excellent performance with low distortion for the pre-amp outputs as tested up to 10Vrms/+22.2dBu XLR in high-gain mode (the device is rated up to +26.6dBu if you need very high output voltage). This is great especially for some of the low-gain, very low distortion amplifiers like the Benchmark AHB2 (+22dBu for full amplitude), Topping's own LA90 Discrete (+19.5dBu sensitivity) or B200 Monoblock (+23.8dBu sensitivity) connected with XLR in. With these kinds of amps, one could achieve better than THD+N -110dB from digital in to amplifier output! No loudspeaker (or headphone) in existence would be able to reproduce that kind of resolution. Unless you need analogue/phono inputs, the Topping DX9 can be the heart of a supremely high performance digital system.
Let's of course not forget that this is also a high performance headphone amplifier at least on par if not exceeding Topping's own A90 Discrete Headphone Amp. The DX9 has 6 discrete NFCA amplifier modules that can push up to an advertised 3.3W x 2 into 16Ω <1% THD+N unbalanced, and 10W x 2 into 16Ω <1% THD+N balanced across the 3 outputs (1/4" phono single-ended, 4.4mm TRRS balanced, 4-pin XLR balanced) simultaneously. While I did not run balanced headphone out testing, I can confirm that the single-ended output into low/medium/high loads achieved excellent performance consistent with these specifications. This should be more than enough power except for the most demanding low-sensitivity headphone designs!
Honestly, I'd be quite happy if this was the last USB DAC I ever purchased. No doubt, this is an example of a Chinese audiophile manufacturer making an entry into the world of "luxury" high-end in the DAC-space with a device that's more expensive than typical, but still clearly very affordable with high value when we look around at the prices in the high-end audio Industry and compare features.
Subjectively, the AKM converters have a reputation of being "smooth" with more authoritative bass weight. While I don't necessarily agree since what I hear is simply a very good high resolution, high-fidelity sound, I've seen it often described this way.
The Topping DX9 has the makings of a "classic" product as a limited edition (~1800 units made) device based on the now-unobtainable flagship AKM AK4499EQ converter, no longer made due to the Japanese factory fire during the pandemic lockdowns. Scarce core component, beautifully applied art of industrial design, matched with high-fidelity engineering to mark Topping's 15th anniversary.
As Steve Jobs used to say: "One last thing..."
I wrote in Part II:
"With the usual 1kHz pure sine wave signal, the XLR line outs from this DAC were capable of achieving THD+N -121dB (SINAD 121dB) on my test bench quite easily. At this stage of my audio reviewing journey, whether it's actually -121dB as I measured or -124dB (see Topping advertised measurements) is immaterial for listening pleasure. 🙂"
While true, when a product performs at THD+N of -120dB already, anything beyond that is completely academic. However, for "bragging rights", it's fun to measure the limits of a device to talk about impressive engineering numbers. So with the E1DA Cosmos ADCiso (Grade A) in mono mode, utilizing the Cosmos APU, let's have one last look at THD+N/SINAD and the Dynamic Range for this device measured with the highest precision at my disposal:
![]() |
One characteristic I've typically observed with AKM DACs is a nice descending "harmonic cascade" where higher order harmonics (beyond 2nd and 3rd) remain suppressed. It has been said that lower-order harmonics like 2nd and 3rd are likely inaudible and might even be "euphonic" given their prevalence in tube and vinyl reproduction to add subjective "fullness" to the sound. It's those higher order odd products like the 5th, 7th, 9th that we probably want to maintain as low as possible. |
THD+N -125dB average (SINAD 125dB), with dynamic rage/SNR of 130dB over USB2.0 connected to a general Windows 11 MiniPC.
As a hi-fi DAC, I think that'll do. 😉
--------------------
In closing: DAC Perceptible Perfection achieved - and as humans, desiring more.
![]() |
Parting shot of the Topping DX9 from the listening position. 2-channel XLR out upmixed with Auro-Matic on the Integra receiver. Playing some Namewee's Ghosician (2021, DR6); pretty funny but can be crude and controversial Malaysian rapper. |
Hi Archi
ReplyDeleteVery refreshing kind of having a look on a product with unbeatable facts and so it is refreshing to read this with a refreshing smile on my face. Thanks.
All the best
Juergen
Thanks for the note Juergen,
DeleteYeah, quite the machine! I always find it interesting and a bit surprising that devices like this are not more commonly discussed among audiophiles on the forums I visit.
There seems to be a split in this hobby between those who are comfortable talking about the electronics, objective performance characteristics, and those who don't want to (or maybe don't want to think that objective results apply to oneself?).
Perhaps similar to the polarizations we see in society unfortunately.
Hope you're having a great spring out in Europe!
Hi Archi,
ReplyDeleteI can see it performs well on paper, but did i miss the part where you talk about how it sounds? How does it compare sound wise to other DACs in it's class, or what you have?
Cheers,
Ruck
Hey Ruck,
DeleteI did the more traditional thing in the layout of this review and placed the subjective listening impressions in Part I talking about various albums as well.
I don't think I'm being too controversial in saying that this DAC sounds fantastic compared to other highly accurate DACs I've heard here; and I'd have a very hard time if not outright fail in telling a difference if blinded.
The most expensive DAC I've listened to in this room with this gear would be a previous-gen Linn Klimax/DSM2 Katalyst streamer (Sample C here) which had an asking price of around US$20k in 2020 for context that I quite enjoyed. I still have a Linn Klimax Organik (I think, need to double check the version) in my sound room that I need to run some measurements and publish on.
Correction, I have the Linn Klimax DS-2 Exakt here in my room... Will see about publishing some Linn measurements here in the next little while.
DeleteHi Arch,
ReplyDeleteComprehensive, detailed and thorough. I do not think more could be said about this Dac. It is on my wish list! // Cheers Mike
Thanks Mike,
DeleteIt's always a relief to finish off a multi-part review! Since I usually publish on a Saturday AM if I can, I'll take the evening off and just listen. Last night, spent hours with this DAC connected to my Integra receiver listening to the stereo output upmixed to Auro-3D with Dirac Live Bass Control DSP room correction.
Simply wonderful evening of enjoyment digging deeper into some albums I had missed over the years like Alison Krauss + Union Station's 1997 So Long So Wrong. Many times I've been impressed by just how "immersive" music sounds coming out from a very high resolution source upmixed to multichannel!
I agree with Mike - a great review, thorough and full of insight! And on my wish list too now :)
DeleteHi Archi,
ReplyDeleteThanks for the review. I read all 3 parts with great interest and pleasure. I've been using DX9 for over a year and I'm quite happy with it. I bought a silver one first, and later a black one, so now I have two of them.
Cool Jack,
DeleteStunning looking machine that performs excellently. This DAC could easily be my "end game" 2-channel DAC+headphone amp, focusing on the music from this point on.
Congrats on having both members of the family. 😉
"Digitally generated test tones are often used to measure D/A converters;
ReplyDeleteit is important to choose test frequencies that are not correlated with the
sampling frequency. Otherwise, a small sequence of codes might be
reproduced over and over, without fully exercising the converter.
Depending on the converter’s linearity at those particular codes, the output
distortion might measure better, or worse, than typical performance. For
example, when replaying a 1-second, 1-kHz, 0-dBFS sine wave sampled at
44.1 kHz, only 441 different codes would be used over the 44,100 points. A
0-dBFS sine wave at 997 Hz would use 20,542 codes, giving a much better
representation of converter performance. Standard test tones have been
selected to avoid this anomaly. For example, some standard test frequencies
are: 17, 31, 61, 127, 251, 499, 997, 1999, 4001, 7993, 10,007, 12,503,
16,001, 17,989, and 19,997 Hz."
This quote is from Ken Pohlmann's book "Principles of Digital Audio".
Archi, don't you think that in view of the quoted above testing a DAC with 1 kHz sine wave may not test it properly?
Happy D50 III + PA5 II camper here. AFAIK these two are the best value offerings in their own categories as of now.
ReplyDeleteHi Archi,
ReplyDeletePrior to version 5.40.78 beta rew had a gross error in IMD vs Level plotting, the results were underestimated by about 12 dB. It looks like the version you used has this error.
IMHO there are not enough performance measurements at high frequencies for a complete picture. If still available, it would be very interesting to see THD (without noise) vs frequency performance, accurate THD measurement at 10kHz and wideband fft (without weighting) with a weak signal such as the signal for the DR test.
Thanks!