Schiit Yggdrasil Stereophile Review + Measurements

Discussion in 'Digital: DACs, USB converters, decrapifiers' started by purr1n, Jan 20, 2017.

  1. Cspirou

    Cspirou They call me Sparky

    Pyrate
    Joined:
    Sep 27, 2015
    Likes Received:
    8,200
    Trophy Points:
    113
    Location:
    Northwest France
    How do you get 21 bit resolution from using 20 bit chips?
     
  2. TraverS

    TraverS New

    Joined:
    Sep 28, 2015
    Likes Received:
    10
    Trophy Points:
    3
    Something like 2×2^20=2^21
     
  3. Rotijon

    Rotijon Friend

    Pyrate
    Joined:
    Sep 28, 2015
    Likes Received:
    261
    Trophy Points:
    63
    I think a lot of it was also how Schiit marketed the dac. The way it was hyped at Headfi and here (Or changstar back in the day), made it sound as if it was the second coming of Christ at 1/20 the price, which probably pissed off a ton of manufacturers.

    No doubt, it was damn good and its pricing "truncated". But performance wise, it was no where near the best DAC on the planet like it was made out to be. I mean for a long time, this DAC was spoken alongside the DCS Vivaldi, Totaldac Dual's and top end MSB's.

    Measurements wise (and to my ears, sound wise), the DCS Vivaldi still beat it quite handily. Mike calls this the best DAC he can make, but i don't believe that for a second.

    I get that Schiit wants to make stuff for the masses but please make a cost no object (internal, not bling bling casing) reference piece (or digital chip for upgrade of the Yggdrasil). Even Toyota have the Lexus LFA and Honda the Honda NSX. Right now, all Schiit has is the Honda Civic Type R or Toyota 86, amazing cars no doubt, but it aint a supercar.

    For reference to the measurement comparison, please refer below.

    http://www.head-fi.org/t/766347/schiit-yggdrasil-impressions-thread/3840#post_13193607

    And for the record, the measurement performance of DAVE is better than the Vivaldi's (assuming robs measurement is right, waiting for third party confirmation). Sound wise, i hate to say this, but the DAVE which looks pretty empty on the inside (I'm pretty sure the BOM is less than 1,500) is one of the best sounding DAC's i have ever heard.

    http://www.head-fi.org/t/766517/chord-electronics-dave/1665#post_12341546
     
    Last edited: Jan 21, 2017
  4. landroni

    landroni Friend

    Pyrate
    Joined:
    Aug 12, 2016
    Likes Received:
    1,164
    Trophy Points:
    93
    My understanding is that Chord hides behind (theoretical) simulations, instead of reporting actual measurements, for their DACs. Do they really keep a 180 dB ADC handy?
     
  5. Kattefjaes

    Kattefjaes Mostly Harmless

    Pyrate
    Joined:
    Sep 5, 2016
    Likes Received:
    4,521
    Trophy Points:
    113
    Location:
    London, UK
    OMGz, some people who shill via inky marks on bits of flattened tree that are largely ignored by modern ape descendants don't like your DAC, Jason and Mike.. whatever are you going to do?

    Seriously, Stereophile is part of the precipitate rather than the solution, at this point.
     
  6. Rotijon

    Rotijon Friend

    Pyrate
    Joined:
    Sep 28, 2015
    Likes Received:
    261
    Trophy Points:
    63
    Im waiting for the third party review and measurement. Hopefully stereophile so its easier to compare.

    @katterfjaes
    The main reason why audio prices are so high is due to a lack of emphasis and evolution on objective measurements (ie;Squiggly lines).

    Take cars, measurement is not everything, the toyota 86 is a joy to drive, but does not exactly post the best 0-60 figures etc, one could argue that your money is better spent on a WRX. But try to make a ferrari with a 7 second 0-100 time and sell it for 300k and you'll see hell.
     
  7. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,594
    Trophy Points:
    113
    Location:
    near Munich, Germany
    I wouldn't read too much into those standard measurements from stereophile. Sure, you may see some of the filter behaviour and noise/distortion and jitter performance, but I bet there's much more to digital audio. Just as you can't really tell how headphones sound from looking at coupler measurements, or amps from distortion, FR and square wave measurements.
    I think INL/DNL measurements are more useful for DACs. I also liked how MSB visualized the clock accuracy as a form of frequency inaccuracies (here).

    Overall the Yggdrasil measurements looked superb, even compared to much more expensive DACs.

    The DCS Ring DAC to me looks like a bunch of randomly chosen 5bit D-S DACs in a sort of resistor ladder architecture.

    Doesn't the Schiit MB upsampling filter basically allow you to downsample the signal back to the original, while with the other filters you will always get some added noise?
     
  8. Grahad

    Grahad Friend

    Pyrate
    Joined:
    Jan 18, 2017
    Likes Received:
    135
    Trophy Points:
    83
    Location:
    SGP
    Schiit's supermegacomboburrito filter is like taking the original data points, draw a curve between those points, and add dots on the curve to the dataset. Other filters draw a curve using the original points, then define a new set of points based off the curve (but not off the original data). Or at least that's how I understand it.
     
  9. Cspirou

    Cspirou They call me Sparky

    Pyrate
    Joined:
    Sep 27, 2015
    Likes Received:
    8,200
    Trophy Points:
    113
    Location:
    Northwest France
    I don't really understand then. I would think adding another chip adds another 20 bits which would be more like 40 bits.

    But from what I have read for a balanced DAC, the second chip is an inverted version of the original digital signal. This still means that the 1st chip takes a 20bit signal while the other chip takes an inverted version of that digital signal. Where does the 21st bit fit in this?

    Does anyone have some links to some literature about this topic?
     
  10. slowsound

    slowsound Acquaintance

    Joined:
    Apr 27, 2016
    Likes Received:
    43
    Trophy Points:
    18
    Location:
    Canada

    2^20 = 1048576
    2* (2^20) = 2* 1048576
    2097152 = 2097152
    2^21 = 2097152
     
  11. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    That's how it is sort of explained. You can do that at the expense of loosing some accuracy on the estimation of the points between the original data points.

    However, the filter Schiit is using is so large that the difference between doing it one way or the other (Schiits vs other filters than minimize an error metric), with the same size filter, is sort of academic.
     
  12. landroni

    landroni Friend

    Pyrate
    Joined:
    Aug 12, 2016
    Likes Received:
    1,164
    Trophy Points:
    93
    Schiit's ping-pong with Stereophile has been incredibly satisfying, so here it is for our own amusement:

    Original Stereophile comment:
    [​IMG]

    Apparently this got @schiit's blood boiling, so after a letter to the editor here's the amended position (HF link):
    [​IMG]

    And @schiit's response:
    [​IMG]
     
  13. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    I think they are not using one DAC to handle the 20 bit LSBs and another to handle the 20 bit MSBs. Or using a multiplier approach.

    What could be happening is that they take 20 bits from one branch of the balanced connection, and 20 bits from the other branch. Then they add them.

    While

    20-bit X 20-bit = 40 bit,

    20-bit + 20-bit = 21-bit.

    I believe they are using the later approach.
     
  14. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    I believe you can tell a lot by looking at the distortion, noise, jitter and FR measurements. But the issue JA is having with Schiit's approach seems to me over blown if you consider more expensive offerings with lower measurement performance and higher praise from Stereophile.

    INL/DNL are important measurements for DACs (precision DACs that is). But it is possible that a delta sigma DAC will destroy a precision DAC in such a measurement, unfairly. And in some ways, things may become academic again.

    Also, downsampling of the DAC's analog output signal to the original is not going to happen, cuz it's not digital anymore. If however you were to pass the output of the DACs through an ADC and compare to the original, things might get interesting. Most other vendors might use smaller FIRs which I think might be more of a cap on their performance than the coefficients used themselves.
     
  15. 3X0

    3X0 Friend

    Pyrate
    Joined:
    Jan 31, 2016
    Likes Received:
    563
    Trophy Points:
    93
    His quote frequently gets pulled out of its original context of practicality.

    @baldr's original words were: "The Yggdrasil is the best I can do short of spending 4-5 times the money."

    And like a few others I'm also suspecting a conflict of interest.
     
    Last edited: Jan 21, 2017
  16. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    There is possibly some of that. I also do not like the anti-Delta Sigma message Mike sends out sometimes among others. And it's possible that is starting cause some issues.

    However, if Mike believes precision DACs are the sex, then Yggdrasil might indeed be the best he can do with that technology given the lack of audio ASICs supporting this approach.
     
  17. Grahad

    Grahad Friend

    Pyrate
    Joined:
    Jan 18, 2017
    Likes Received:
    135
    Trophy Points:
    83
    Location:
    SGP
    BITS DON'T WORK THAT WAY.

    If you want to get 40 bits out of 2 20 bit DACs, you take the most significant 20 bits, feed it into a DAC, and you take the 20 least significant bits, feed it into the other DAC. You then either multiply the first DAC's output by 2^20 (or reduce the second DAC output by 2^20), and sum the results. Pretty much you're dependent on DNL and INL of the first DAC to see how much ENOB you end up with.

    AD5791 comes specified with 1 LSB INL and DNL: This means that adjacent steps may differ from the ideal by 100% (for the LSB). The trouble with that means that a 1 LSB error in your MSBs (Most Significant Bits) DAC contributes to a 20 bit (or 2^20 LSB) error, or your entire output of your second LSBs DAC. So what happens is 20 bits + 20 bits = 20 bits, because of the 1 LSB spec.

    Doing this only works IF a theoretical 20bit AD9999 guarantees 40 bit linearity and distortion, so your 20 bit + 20 bit = 40 bits, but you might just as well make a 40 bit DAC anyway.

    You can't add them either, it doesn't make any sense: It serves to reduce error contributed by random noise (by having more outputs, summing those output reduces the effect of noise: or why the central limit theorem exists, and the main marketing point of most 90X8 DACs). The more additions you perform, the less 'wrong' your output is, but you don't get additional accuracy (except in terms of ENOB).

    Feeding 20 bits into 2 separate (perfect) DACs and summing the output only gets you the output of 1 perfect DAC, just twice as loud.

    There are much more complicated ways to magic up resolution (or how delta-sigma works for e.g.), but I don't understand them enough to feel confident in explaining them.
     
  18. Scott Kramer

    Scott Kramer Friend

    Pyrate
    Joined:
    May 3, 2016
    Likes Received:
    1,446
    Trophy Points:
    93
    This is interesting, the designer of the AD5791, AD5781 popped in once, wonder where that went with MIke:

    http://www.head-fi.org/t/752962/who-is-thinking-of-buying-the-schiit-yggdrasil/210#post_11956866

    Also, Mike on 21-bit:

    http://www.head-fi.org/t/766347/schiit-yggdrasil-impressions-thread/1830#post_12342937

    Two twenty bit DACs (one per phase) double the resolution for a balanced signal. Double the resolution only adds 6db (one more bit) for a total of 21 bits. In this case, 20+20 equal 21.
     
    Last edited: Jan 21, 2017
  19. Merrick

    Merrick A lidless ear

    Pyrate
    Joined:
    Jan 6, 2016
    Likes Received:
    12,569
    Trophy Points:
    113
    Location:
    Portland, OR
    Again, I think it is important to note that Mike said Yggdrasil is the best DAC he could make without ballooning the price 4-5 times beyond what Yggdrasil costs. In other words, this isn't the best DAC he could possibly make, but it is the best DAC he could possibly make at the price point he wanted to hit.
     
  20. dmckean44

    dmckean44 In a Sherwood S6040CP relationship

    Pyrate
    Joined:
    Nov 17, 2016
    Likes Received:
    1,425
    Trophy Points:
    93
    Location:
    Peoria, IL

Share This Page