Review and Measurements of Schiit Yggdrasil V2 DAC

[trl]

Just for fun, I used Amir's project file to run a J-test check on a cheap DAC I had on my bench. Only change from his setup was that I addressed the DAC's ASIO driver directly instead of using ASIO4ALL. The low output will probably give away the identity, but at under $100, that should demonstrate what jitter performance could be engineered into just about any product.
View attachment 13516

Need some more clues, please. I'm sure it's an USB-powered DAC, but need more data, please.

 

[SIY]

Rather than veer this one off to other equipment, I'll do some measurements and start another thread when I have the opportunity.

 

[maxxevv]

Are you saying that everything with a dac inside should sound the same?
I cannot tell my cd players or dacs apart with 10 seconds of blind test, but certainly I know their sound signature because I used them for 10 to 20 years, none sound the same, not even close.
The Topping D30 that is measured here, sound very good on the optical input, still acoustic guitars sound wrong, a bit artificial. I cannot prove it and I cannot measure it, so feel free to ignore it..

They will sound almost the same if the circuitry implementations, particularly the endstage amplification circuitry are identical. (To the point that it will take considerable effort to tell them apart under controlled conditions.)
I'm very sure none of yours are identical in that regard.

Case in point would the idea of opamp rolling in DACs. Between a ES9028Q2M DAC /Amp swapping between NJC5534 / AD797 and LT2118 opamps and a ES9038PRO DAC and the same ES9028Q2M DAC running AD797 chips .... its far, far harder to tell them apart for the latter case.

I can assure you, as I have done the tests myself. Its way harder to tell betwen DAC chips than OpAmps.

 

[SIY]

Sigh.

 

[Theo]

So, back to "burden of proof is on those who make extraordinary claims. Especially when they're making money from them."

IMO it would actually be of interest to one of those companies to publish such a proof, so they would demonstrate a competitive advantage! The fact that there is none who does it tends to prove that it is impossible to show any relevant measurement to substantiate their claim. Except for speakers, some manufacturers (serious ones) show actual measurements.

 

[Thomas savage]

all further off topic posts will be deleted, please start new threads or seek out old ones if you want to talk about other things.

 

[amirm]

Thank you Thomas. This happened before and we created a dedicated debate thread for measurements. Please see: https://audiosciencereview.com/foru...-debate-thread-about-audio-measurements.2580/

 

[amirm]

@amirm can you upload a project file?

Sure. Here is the link to Jude's linearity test with my measurements added to it which I post earlier. https://www.dropbox.com/s/7qp0wle146yhglt/Project 2018-06-30 0301 EDT-amir.approjx?dl=0

Let me know if you can't access it.

 

[amirm]

Here is some more data I had created to demonstrate the linearity error in Schiit Yggdrasil I tested. It uses the "FFT method" of converting the time domain signal (what we hear) to frequency spectrum (what we can understand). Using a high-resolution FFT and zooming way into our test tone, we can completely ignore noise and distortion and see if the DAC itself at least makes an attempt to output what it is told prior to its output getting polluted by other factors.

First, Yggdrasil DAC spectrum as I step through a few amplitudes. The data is sent as PCM samples to it, its analog output captured, digitized and then transformed by FFT above:

Starting from top, -100 dB looks more or less correct (our eyes and chart resolution is more limited than linearity graphs I normally show). Looking OK at -105 too. But by -110 we can see that the peak is at a different level than what the DAC is told to do.

Much more interesting is what happens when we tell it to produce -115 dBFS (20 bits of resolution). It shows -117.5 dB instead. But no matter how much lower we go from there, the output refuses to change! All the graphs from there on to -130 dB keep outputting -117 dB.

It seems that the low order bits beyond 20 (which their internal DAC supports) are being thrown away.

Now let's run the same thing on Topping DX7s:

Now we are cooking with gas. Notice that we get progressively smaller domes all the way down to -140 dBFS with less than a 1 dB or so of error. This is a proper 24-bit DAC.

JA at stereophile observed the same thing in the unit he tested: https://www.stereophile.com/content/schiit-audio-yggdrasil-da-processor-measurements

Fortunately that need not be the case. 24-bit data can be dithered down to 20 bits instead of truncating and continue to produce proper signal. Yes, you will have higher noise floor but fortunately for Schiit, our linearity measurements perform a ton of noise filtering (to the extreme in the above FFT method) so the graphs would have looked correct. Perhaps you have seen this in Jude's measurements? I think you have .

This tells me that Schiit has fixed this problem in newer units but is unwilling to state it as such for the fear of everyone asking for an upgrade. I disagree. Less is wrong with stating that you have fixed a problem. RME has done that with "FS" version of my ADI-2 Pro. While it makes me somewhat unhappy, I am glad the problem is resolved.

Same thing is true I believe in the unbalanced output frequency response error. Whatever was wrong in the analog circuit could have been easily fixed and retrofit to new units in production.

Summary
There is no doubt that the linearity errors have been a design problem with Schiit Yggdrasil DAC. The three devices I have measured show it. And JA's review shows it. The cat-and-mouse game to avoid saying it is now fixed is very unfortunate. Heaven knows I would have wanted to spend my time measuring the countless other gears in my inventory than re-testing and finding what I had found before.

I guess there is a bonus here for owners of Schiit Yggdrasil. Because Schiit is claiming that there has never been a problem, you all can contact them and ask them if they will measure and fix the linearity error for you. Tell them if they refuse, you will be sending your unit to me to test instead. That ought to get their attention.

 

[Candlesticks]

Summary
There is no doubt that the linearity errors have been a design problem with Schiit Yggdrasil DAC. The three devices I have measured show it. And JA's review shows it. The cat-and-mouse game to avoid saying it is now fixed is very unfortunate. Heaven knows I would have wanted to spend my time measuring the countless other gears in my inventory than re-testing and finding what I had found before.

The people who buy esoteric designs like Multibit do not care about these objective measurements anyway. It's your time and if you find it fun as a technical exercise then that is great but in terms of utility it's a waste of time that could otherwise be spent on finding an objectively good $50 DAC that thousands would buy through recommendation.

 

[Grave]

^The worse it measures the better it sounds! /s

 

[amirm]

I hope you can get your hands on another quality oversampling multi-bit DAC, such as those from Soekris, and compare it to the Yggy, or even a vintage one from the 90s based on the 20-bit chips from that era, e.g. PCM63 or UltraAnalog. It would be more of an apples to apples comparison than putting it against a Delta-Sigma unit that's based on the (almost) top of the line ESS chip of current production. That said, there is no excuse for the mains noise.

Your post was in the back of my mind when I thought about showing more FFT linearity measurements. So I went and ripped out my Mark Levinson No 360S which I purchased back in 1999 (?) from my main system. Based on specs I can find, it says it uses four PCM1704 DACs. Internally the PCM1704 is dual 23-bit multi-bit DACs used in sign magnitude implementation to avoid glitches (unlike AD DACs used in Schiit Yggdrasil). Here is the full specs: http://www.qlshifi.com/jszl/PCM1704.pdf

Here is how she did on the same FFT at power-on:

As we see, it clearly goes down to -140 dBFS albeit, with some error.

I found the manual and was fascinated by them specifying linearity error in there! https://www.manualslib.com/manual/631824/Madrigal-Audio-360s.html?page=44#manual

Here is Linearity measurement using Jude's/AP script:

As we see, it (in brown) easily beats the Yggdrasil DAC I tested (in red). But if the Yggy is now fixed to produce the blue graph per Jude's measurements, then it runs with it to better than -125 dB but from then on it gets worse.

Still 3 dB error at -140 dB for a 20 year old DAC is not so bad .

I will run more tests on it when I have time.

 

[amirm]

The people who buy esoteric designs like Multibit do not care about these objective measurements anyway. It's your time and if you find it fun as a technical exercise then that is great but in terms of utility it's a waste of time that could otherwise be spent on finding an objectively good $50 DAC that thousands would buy through recommendation.

I hear you but I think the situation is different. The marketing of multi-bit DACs has made many people assume they are more accurate. That they are not, is very much news to many.

Why would Schiit launch this PR campaign through Jude and Atomicbob to say my measurements are not true? Clearly people care.

There is so little data out there that people run with whatever they read online. I can tell you with confidence from reading other fora that the measurements and objective engineering evaluations we do here both on performance and manufacturing is making a significant difference. I would not at all purchased my new APx555 if I thought all of this is a waste of time.

Importantly, it is not my goal to change the opinion of select few in the upper stratosphere of high-end audio. They are as you say creating their own audio science and believe whatever they like. There is a far, far larger group of audiophiles at the lower end of price scale many of whom absolutely read and care about this information. And use it to base their purchase decisions on.

All of this is based on just a few months of testing. My outlook is measured in years when we have a large library of data backing each other. We need to have the good be distinguished from the bad, with no emotion, no commercial interest and on the side of the consumer. Ultimately I can't change everyone's mind but I think many will.

 

[Blumlein 88]

I hear you but I think the situation is different. The marketing of multi-bit DACs has made many people assume they are more accurate. That they are not, is very much news to many.

Why would Schiit launch this PR campaign through Jude and Atomicbob to say my measurements are not true? Clearly people care.

There is so little data out there that people run with whatever they read online. I can tell you with confidence from reading other fora that the measurements and objective engineering evaluations we do here both on performance and manufacturing is making a significant difference. I would not at all purchased my new APx555 if I thought all of this is a waste of time.

Importantly, it is not my goal to change the opinion of select few in the upper stratosphere of high-end audio. They are as you say creating their own audio science and believe whatever they like. There is a far, far larger group of audiophiles at the lower end of price scale many of whom absolutely read and care about this information. And use it to base their purchase decisions on.

All of this is based on just a few months of testing. My outlook is measured in years when we have a large library of data backing each other. We need to have the good be distinguished from the bad, with no emotion, no commercial interest and on the side of the consumer. Ultimately I can't change everyone's mind but I think many will.

One of my favorites, which I believe the people at Schiit mention regularly, is that multi-bit DACs output the actual sample values in the PCM music stream while sigma delta DACs aren't able to do this. Yet we see from measurements the multi-bit DACs (at least from some sources) can't do it either. And while one might claim sigma delta DACs are simulating the signal, they clearly simulate that signal more accurately than multi-bit typically does. It is just another way to make something sound inherently better to those who don't know how things work.

 

[SIY]

@amirm One minor suggestion- when you're looking at single tone amplitude, you might want to switch the FFT window to Flat Top (if you're not already doing that) for most accurate magnitude. AP Equiripple is an excellent compromise window (I use it 99% of the time, except for noise measurement), but if we're picking nits, we have the option of using the sharpest tool.

 

[gvl]

Your post was in the back of my mind when I thought about showing more FFT linearity measurements. So I went and ripped out my Mark Levinson No 360S which I purchased back in 1999 (?) from my main system. Based on specs I can find, it says it uses four PCM1704 DACs. Internally the PCM1704 is dual 23-bit multi-bit DACs used in sign magnitude implementation to avoid glitches (unlike AD DACs used in Schiit Yggdrasil). Here is the full specs: http://www.qlshifi.com/jszl/PCM1704.pdf

Here is how she did on the same FFT at power-on:
View attachment 13539

As we see, it clearly goes down to -140 dBFS albeit, with some error.

I found the manual and was fascinated by them specifying linearity error in there! https://www.manualslib.com/manual/631824/Madrigal-Audio-360s.html?page=44#manual

View attachment 13540

Here is Linearity measurement using Jude's/AP script:

View attachment 13541

As we see, it (in brown) easily beats the Yggdrasil DAC I tested (in red). But if the Yggy is now fixed to produce the blue graph per Jude's measurements, then it runs with it to better than -125 dB but from then on it gets worse.

Still 3 dB error at -140 dB for a 20 year old DAC is not so bad .

I will run more tests on it when I have time.

Thank you, very interesting yet 24 bit PCM2704 beating 20 bit DACs isn't overly surprising.

 

[mi-fu]

There is a far, far larger group of audiophiles at the lower end of price scale many of whom absolutely read and care about this information. And use it to base their purchase decisions on.

I for one fall into this category. I am a novice in this audio hobby. Weeks ago, I began to spend time on online forums to look for a good DAC for my use. At first, I considered one of the Schiit products, because there are so many rave reviews. Very soon, I began to a problem: There are so many inconsistencies in those comments. There is something lacking in those flowery languages. Later, I discovered this forum and looked at many tests done here. Numbers convinced me, not pure subjective narratives with superlative adjectives.

Eventually, and perhaps not surprisingly, I bought a D50, then DX7s.

I don't want to digress this thread from the review. But I want to say that these reviews are important and helpful to average consumers. And I learn more about this hobby from these reviews. So I don't think it is just a case of "haters gonna hate." In fact, many people are unbiased consumers who want more fact-based information.

 

[Thomas savage]

I hear you but I think the situation is different. The marketing of multi-bit DACs has made many people assume they are more accurate. That they are not, is very much news to many.

Why would Schiit launch this PR campaign through Jude and Atomicbob to say my measurements are not true? Clearly people care.

There is so little data out there that people run with whatever they read online. I can tell you with confidence from reading other fora that the measurements and objective engineering evaluations we do here both on performance and manufacturing is making a significant difference. I would not at all purchased my new APx555 if I thought all of this is a waste of time.

Importantly, it is not my goal to change the opinion of select few in the upper stratosphere of high-end audio. They are as you say creating their own audio science and believe whatever they like. There is a far, far larger group of audiophiles at the lower end of price scale many of whom absolutely read and care about this information. And use it to base their purchase decisions on.

All of this is based on just a few months of testing. My outlook is measured in years when we have a large library of data backing each other. We need to have the good be distinguished from the bad, with no emotion, no commercial interest and on the side of the consumer. Ultimately I can't change everyone's mind but I think many will.

Once you unearth ‘ issues’ with products and the manufacturers take note and (silently) ‘fix’ them, err do we get a consultancy fee?

Seems despite the protestations you have in effect done some a service , effectively working for them lol

We should get paid

 

[amirm]

@amirm One minor suggestion- when you're looking at single tone amplitude, you might want to switch the FFT window to Flat Top (if you're not already doing that) for most accurate magnitude. AP Equiripple is an excellent compromise window (I use it 99% of the time, except for noise measurement), but if we're picking nits, we have the option of using the sharpest tool.

That's a good point. Alas, the selectively becomes poor with flat top and lobing gets excessive, making the graph look well, offensive if you get my drift

 

[Blumlein 88]

Once you unearth ‘ issues’ with products and the manufacturers take note and (silently) ‘fix’ them, err do we get a consultancy fee?

Seems despite the protestations you have in effect done some a service , effectively working for them lol

We should get paid

And that doesn't include the free PR such 'controversy' provides.