Topping D90 Balanced USB DAC Review

[majingotan]

It is well known that no processor currently available can do DSD 512 with the EC modulators in HQP. But, no need to fear, because the AKM DAC chip does better with DSD 256 than it does with DSD 512. This is right in the data sheet for the AKM 4499. So best performance with this DAC chip is at DSD 256, not 512. This is likely because the switching elements in the chip are not fast enough for the higher rate of 512. In fact, Andreas Koch has mentioned a few times in interviews that he believes DSD rates above 256 are a bad idea, because no switching devices are fast enough to remain fully linear at this rate.
I no longer have the D-90 here (sold it), but I preferred it with DSD 256 inout from Roon (linear phase filter, precise, 7th order CLANS modulator) with the D-90 in "direct DSD" mode (aka "DAC" mode in the set up menu).

BTW, the poster above is not correct when saying D-90 does not run in a non-oversampling mode, in DSD direct there is no oversampling. But there is not a non-oversampling mode for PCM input.

Do you know how NativeDSD can upsample to DSD512 with EC modulators? What if instead of sending the stream to D90 you can output to a DSF file and play that file back. That would use 0.1% CPU utilization instead because the computer is NOT filtering realtime and just sending the stream to the D90 at 45.2 Mbps. Any low powered PC can also play DSD512 with no hiccups whatsoever if played from a file rather than realtime oversampling/filtering

Side note: Funny that those mastering studios use a terribly measuring Nagra HD DAC X as their reference DAC lol

 

[jhwalker]

Do you know how NativeDSD can upsample to DSD512 with EC modulators? What if instead of sending the stream to D90 you can output to a DSF file and play that file back. That would use 0.1% CPU utilization instead because the computer is NOT filtering realtime and just sending the stream to the D90 at 45.2 Mbps. Any low powered PC can also play DSD512 with no hiccups whatsoever if played from a file rather than realtime oversampling/filtering

View attachment 58179

Side note: Funny that those mastering studios use a terribly measuring Nagra HD DAC X as their reference DAC lol

Signalyst makes an HQPlayer Pro (I think it's called) that is more expensive but has the ability to write to files.

 

[barrows]

Do you know how NativeDSD can upsample to DSD512 with EC modulators? What if instead of sending the stream to D90 you can output to a DSF file and play that file back. That would use 0.1% CPU utilization instead because the computer is NOT filtering realtime and just sending the stream to the D90 at 45.2 Mbps. Any low powered PC can also play DSD512 with no hiccups whatsoever if played from a file rather than realtime oversampling/filtering

View attachment 58179

Side note: Funny that those mastering studios use a terribly measuring Nagra HD DAC X as their reference DAC lol

This because they use HQPlayer PRO, which is sample rate conversion software which does not do conversions “on the fly”, during playback, it converts the files over time, not in real time during playback, so the software can take take its time. This is entirely a different process than oversampling in real time during playback Using HQPlayer. of course, audiophiles could buy a copy of HQPlayer PRO and do the same thing, although it is going to take an awfully long time to convert one’s entire library to DSD 512 this way, and a huge amount of storage space if you are going to store all of your files as DSD 512.
Additionally, I have serious doubts that there are any advantages to DSD rates over DSD 256, considering the switching speed of converter hardware, as mentioned preiviously. I do not have time to search it out right now, but Andreas Koch has explained this quite clearly in interviews on the topic, and if you are interested in reading the technical explanations why, I am sure you can find this info by googling his name, and “interviews”. If you do not know who he is, and you have interest in DSD, you should look him up anyway.

 

[pacman]

Let me know if I'm understanding what's been discussed here right now.

People playing PCM audio by upsampling it via DAC/shaddy program/plugin to DSD512? What the actual f*ck is this?

 

[Veri]

Let me know if I'm understanding what's been discussed here right now.

People playing PCM audio by upsampling it via DAC/shaddy program/plugin to DSD512? What the actual f*ck is this?

There's a bit of a community around HQplayer... archimago did an article on it:
http://archimago.blogspot.com/2019/02/measurements-look-at-hqplayer-325.html

 

[Peib]

It is well known that no processor currently available can do DSD 512 with the EC modulators in HQP. But, no need to fear, because the AKM DAC chip does better with DSD 256 than it does with DSD 512. This is right in the data sheet for the AKM 4499. So best performance with this DAC chip is at DSD 256, not 512. This is likely because the switching elements in the chip are not fast enough for the higher rate of 512. In fact, Andreas Koch has mentioned a few times in interviews that he believes DSD rates above 256 are a bad idea, because no switching devices are fast enough to remain fully linear at this rate.
I no longer have the D-90 here (sold it), but I preferred it with DSD 256 inout from Roon (linear phase filter, precise, 7th order CLANS modulator) with the D-90 in "direct DSD" mode (aka "DAC" mode in the set up menu).

BTW, the poster above is not correct when saying D-90 does not run in a non-oversampling mode, in DSD direct there is no oversampling. But there is not a non-oversampling mode for PCM input.

Which DAC did u buy to replace the Topping? (if u replaced it)

 

[barrows]

Which DAC did u buy to replace the Topping? (if u replaced it)

I did not buy any DAC to replace the D-90. I purchased the D-90 because I wanted to evaluate the new AKM 4499 DAC chip, and considering the measurements of the D-90, I figured it was an easy way to evaluate the sound of the DAC chip. I was/am especially interested in this DAC chip in its "DSD direct" mode, because I am big believer in DSD conversion being better sounding than most PCM delta sigma conversion.
I mostly build DIY DACs, and so I wanted to know if it might be worth it to build a DIY DAC with the AKM 4499.

My current DAC, which I preferred to the D-90 (but not in every sonic category), is a DIY DAC based on the ESS 9038 PRO chip-but this was not a fair fight: if this DIY DAC was made into a commercial product it would likely need to retail around $10K or so, considering the attention to detail in its build and the sophistication of its power supplies, USB, and clocking circuitries.

Those here at ASR who believe all DACs sound the same are welcome to ignore my posts on all this. With either of these DACs I am oversampling all files to DSD 256 using Roon (linear phase, precise filter, 7th order CLANS modulator), as I prefer the sound of high rate DSD in conversion, and this also allows the DAC to do much less processing-instead the processing (oversampling, re-modulation) is moved to a computer where it can be done using much more complex algos with far higher precision than be accomplished by the small DAC chip.

BTW, if I had the cash to spare, I would just just purchase a Mola Mola Tambaqui, and be done with all this, but alas I cannot afford that DAC, but it sounds fantastic to me, and of course its measurements are beyond reproach as well.

 

[majingotan]

BTW, if I had the cash to spare, I would just just purchase a Mola Mola Tambaqui, and be done with all this, but alas I cannot afford that DAC, but it sounds fantastic to me, and of course its measurements are beyond reproach as well.

If cost is no object is allowed, you can compare the filtering provided by the Tambaqui since the implementation is similar: upsample everything to 32 bit 3.125 MHz PWD (I assume DSD 256 Roon upsampling is also converted to that prior to noise shaping) and despite that "DSP" happening before noise shaping, measurements provide SOTA accuracy and precision as evidenced by Amir's review. The nice thing about off-the-shelf DS DAC chips is that you bypass their own oversampling filter by using an external one from Roon or HQPlayer. I believe this "DSP" is transparent (i.e. measures beyond "normal audibility" and should not measure worse than AK 4499's own digital filters so no harm in using HQPlayer or Roon. IMHO, it's a huge chunk of analog stage handling that make or break the measured performance of the DAC and not really the digital filters. Whether the use of different digital filters is proven to be audible, we must always perform a DBT volume matched test with 95% confidence (19/20) that the correct oversampling filter is chosen.

One can perform an unofficial DBT volume matched test by downloading a DSD 256 or DSD 512 (preferably this) track off of NativeDSD and compare that to the CD quality version. In doing this, you are comparing the digital filters used by AK4499 (from CD quality) against NOS (from DSD512).

 

[Veri]

you make a lot of sense, @majingotan

 

[Matias]

This because they use HQPlayer PRO, which is sample rate conversion software which does not do conversions “on the fly”, during playback, it converts the files over time, not in real time during playback, so the software can take take its time. This is entirely a different process than oversampling in real time during playback Using HQPlayer. of course, audiophiles could buy a copy of HQPlayer PRO and do the same thing, although it is going to take an awfully long time to convert one’s entire library to DSD 512 this way, and a huge amount of storage space if you are going to store all of your files as DSD 512.
Additionally, I have serious doubts that there are any advantages to DSD rates over DSD 256, considering the switching speed of converter hardware, as mentioned preiviously. I do not have time to search it out right now, but Andreas Koch has explained this quite clearly in interviews on the topic, and if you are interested in reading the technical explanations why, I am sure you can find this info by googling his name, and “interviews”. If you do not know who he is, and you have interest in DSD, you should look him up anyway.

I always wondered why people go so far upgrading CPU and GPU and throwing processing power to crunch numbers with these upsampling playbacks instead of buying more storage and doing it offline. But to each his own.

 

[barrows]

I always wondered why people go so far upgrading CPU and GPU and throwing processing power to crunch numbers with these upsampling playbacks instead of buying more storage and doing it offline. But to each his own.

IMO, there are a few factors at work here. While my library is not huge, it is about a Tb of music, and oversampling everything to DSD 256 offline would take an awfully long time. In addition, in terms of HQPlayer, Jussi makes improvements to his filters and modulators quite often, and re-doing the oversampling over and over would get really tiresome! Also, some find it useful to use different filter/modulator settings for different systems/DACs, and again, re-doing all of your oversampling would be extremely labor intensive.
So yes, to each his own!

 

[majingotan]

IMO, there are a few factors at work here. While my library is not huge, it is about a Tb of music, and oversampling everything to DSD 256 offline would take an awfully long time. In addition, in terms of HQPlayer, Jussi makes improvements to his filters and modulators quite often, and re-doing the oversampling over and over would get really tiresome! Also, some find it useful to use different filter/modulator settings for different systems/DACs, and again, re-doing all of your oversampling would be extremely labor intensive.
So yes, to each his own!

Never thought that oversampling digital music would reach close to levels of basic protein folding simulation in terms of processing power so I'm always surprised on how that operation can be so intensive even with very fast desktop computers. One DSD512 album I dowloaded off of that NativeDSD yielded 17 GB (the Yuko Mabuchi Trio Miles Davis cover), I can just imagine how much terabyte of storage you would need to upsample your tb of music to DSD256 and especially DSD512

 

[barrows]

Well, it does not have to be so processing intensive, for example Roon can easily oversample to DSD 512 (in its most resource intensive settings) on my I5 Mac Mini. But Jussi's HQPlayer and the newer EC modulators take way, way more processing power (and the sound quality difference is notable, at least to me). I am not sure why Jussi's approach takes so much more power, but he has probably spent more time developing SD modulators than anyone else, and he really appears to be pushing boundaries in terms of the math and how he thinks about it, where others seem to be content to suggest that there is no need to go further than what can be done on a relatively low power DAC chip.

 

[AudioQuality777]

Hi.Which of the filters from 1 to 6 is the most natural sound for your taste in D90?It seems to me that filter 3 gives out synthetics in the sound and little naturalness.

 

[barrows]

Hi.Which of the filters from 1 to 6 is the most natural sound for your taste in D90?It seems to me that filter 3 gives out synthetics in the sound and little naturalness.
View attachment 58516

If this was directed at me, as mentioned, I did not use any of those filters in my listening to the D-90, as I fed it only DSD 256 in direct DSD mode. In direct DSD mode the oversampling/digital filtering is completely bypassed.

 

[Viper Necklampy]

Hi.Which of the filters from 1 to 6 is the most natural sound for your taste in D90?It seems to me that filter 3 gives out synthetics in the sound and little naturalness.
View attachment 58516

I guess the most correct is the best. I heard Linear Phase fast roll-off is the most correct in digital audio, but maybe d90 don't have it.. I'll pass the words to more competent ones!

 

[barrows]

I guess the most correct is the best. I heard Linear Phase fast roll-off is the most correct in digital audio, but maybe d90 don't have it.. I'll pass the words to more competent ones!

I would suggest that there really is no "most correct" digital filter for 44.1 kHz input, as digital filter design is a set of tradeoffs with that input rate: there is no possible digital filter which has "perfect" output, one can only trade one imperfection for another, and make a compromise:

Steep fast cut off filters produce a lot of ringing energy, and ringing energy is an artifact not present in the original music.

Shallow slow roll-off filters produce less ringing, but allow for more alias products, and alias products are artifacts not present in the original music.

So a filter designer has to decide which artifacts are/might be less objectionable in a psychoacoustic sense, and then make appropriate choices/compromises based on that.

One should also note that music signals rarely (never for the most part) have full-scale tones anywhere near 20 kHz, or even 15 kHz, so some of the scary looking things we see in filter measurements using full scale white noise are not actually going to be present (or will be at far reduced levels) during the playback of real music.

There is one exception to the above I am aware of, adaptive filters. Ed Meitner claims his DACs use digital filters which change in real time according to the program material, adapting to the music as it plays, in order to avoid artifacts in conversion as much as is possible-but this is a very unusual approach.

 

[AudioQuality777]

I guess the most correct is the best. I heard Linear Phase fast roll-off is the most correct in digital audio, but maybe d90 don't have it.. I'll pass the words to more competent ones!

Yes o Linear Phase fast roll-off .
This 1 filter in D90 it seemed that it is more pdkhodit as a universal option.In 1 filter there is meat(density ) and power in the sound.As good as possible for energetic music.

 

[BDWoody]

In 1 filter there is meat(density in the bass) and power in the sound

How would the filter choice impact 'density in the bass' etc...

 

[blodsbror]

From a usability perspective, is there any "auto sensing" on any of the inputs (or at least it remembers the last input) ?. Also, how well does the remote work, as far as needing to point it directly at it, or not ?. As I combine stereo/TV - then it's quite nice for those less technically inclined to not need to remember which input etc to change to for TV sound etc.