Review: Battle of Schiit Audio DACs

[Svperstar]

I gave my Mimby away because I was hogging all the gear but I agree totally. The smooth sound of it is fantastic. I miss Mimby much. Upgrading to the BiFrost 2 has taken away the sound I loved and replaced it with a cleaner, sterile sound.

I still have it here but I don't really like it compared to my other DACs.

 

[Shives]

Ahh do you want to play battle Schitts? Why yes!!
Mimby sounding good? Really?
And you say you’ve noticed an improvement of sound removing it? I’m confused.

 

[don'ttrustauthority]

I still have it here but I don't really like it compared to my other DACs.

I'll give ya fifty for it.

 

[frki16]

This is a review, detailed measurement and comparison of Schiit's Modi 2 (uber), Modi 2 Multibit and Modi 3. I have reviewed the Modi 2 uber and Modi 3 before so this is mostly focused on how the Modi Multibit compares to them. I have purchased all of them personally at different times. The Modi 2 Uber is now replaced by Modi 3 with the latter costing USD $99 plus shipping. The Schiit Multibit fetches quite a premium compared to that to the tune of USD $249 plus shipping. Is the extra cost worth it? This is what this review is about.

Starting from the top, Schiit seems to be playing with its design in each one of these units:

View attachment 18284

Switch type, color and finish seem to vary some. Overall though, the same recipe is used which results in a very sturdy package of heavy gauge metal for such a small product. I am not a fan of the symbols for inputs and have to keep hunting for which one does what. If you don't have to deal with as many DACs as I do, that is probably a non-issue.

Anyway, you probably came here to see how the Schiit Multibit measures compared to its alternatives. So let's get into that.

Measurements
As usual, we start with our dashboard:
View attachment 18285

Good news is the high output of 2.1 volt as some of the low-end units output less (we like to see 2 volts minimum). That is where the good news ends though. We are greeted with pretty high levels of distortion for a DAC. THD/SINAD are dominated by the high second harmonic. But even outside of that, we have a spray of lower amplitude tones which substantially raise our apparent noise floor.

Without notice or change in version or model name, Schiit seems to have improved the design of Modi Multibit. Using their specification of 0.006%, we see that our results are very similar to theirs even though mine was purchased months ago (0.006% versus 0.007%). So our setup and testing fixture is correct.

SINAD (signal over power of noise and distortion) puts the Schiit Modi Multibit in forth their of performance among DACs tested:
View attachment 18286

For reference I am showing its sister products in yellow (Modi 3) and green (Modi 2 Uber). As we see, Modi 3 wipes the floor with it, producing hugely superior distortion and noise metrics. The Schiit Multibit sadly, sits near the performance that I get from the base motherboard DAC and its much more expensive brother the Schiit Yggdrasi ($2,500, SINAD 86).

Let's look at dynamic range:

View attachment 18287

Another disappointment. Folks looking for "black backgrounds" need to look elsewhere with dynamic range of only 90 dB (6 dB shy of what CD format can produce). Schiit Modi 3 redeems the brand though with respectable Dynamic Range of 111.

Jitter and noise are elevated too compared to competition:

View attachment 18288

While probably not audible (close sidebands are masked and the others below audibility for the most part), they are still bothersome design flaws, having nothing to do with whether you want to have a multi-bit (core) DAC or not.

Intermodulation distortion is not bad though:
View attachment 18289

There is some saturation at high levels (I measured bout 1 dB or so in SINAD). Overall Modi 3 is a clear winner but the gap is smaller here.

Crosstalk is likewise not as good but not an audible concern either:
View attachment 18290

This again measures very similarly to what Schiit publishes for the new version (their graph stops at 20 kHz, mine goes to 60 kHz).

I know you all are waiting for the linearity test and here it is:
View attachment 18291

This is incredibly broken. Using 0.5 dB deviation as threshold of too much error, we only have 12 bits of good resolution (I used to use 0.1 dB deviation in my older reviews so being more generous here).

We can see some correlated error in the zigzags up and down which likely indicates incorrect conversion of 24 bit samples to internal (16 bit?) resolution of the DAC chip. Seems like the bits are truncated rather than properly dithered. This is a huge no-no and results in obvious errors and distortion for no good reason.

The inset shows measurements published by Schiit for the new version. Sadly they use huge vertical scale of +13 and -12 dB which compresses the error graphically making the line look more flat than it is when compared to my measurements which use +- 5 dB. Squinting though, we can see that their measurement shows a lot of improvement here. My suspicion is that they have fixed the signal processing error I mention above. The fix "linearizes" the distortion but raises the noise floor.

Note that the Modi 3 (in blue) nails the linearity with less than 0.5 dB error all the way to -120 dB (20 bits). So even if we take the new measurement from Schiit, we see that the multibit design cannot hold a candle to traditional delta-sigma design. There is a reason the industry has moved to delta-sigma: we get far better performance and at lower cost too!

But Music is Not One Tone
Anticipating the usual defense that sine waves are not music, I thought I do some further testing. Before that, remember, anything that distorts one tone, distorts all tones. In that sense, we are seeing actually less effect of distortion in above measurements than it exists with music. We can prove that by running our tests with multiple tones. Specifically, let's throw 32 tones at the DAC and see what happens (not all visible). First, let's test Schiit Modi 3 this way:
View attachment 18292

The vertical tall peaks are the tones in the source signal. An ideal DAC would have just those with the bottom going to zero volts. We don't have an ideal DAC here so the minimum is above zero and we have a bunch of spikes which get thicker and thicker as frequencies go up. This is because we get distortions products of previous distortions added to new distortions. Got that?

Now let's test the Schiit Modi Multibit the same way:

View attachment 18293

As our one tone tests predicts, we now have a ton of "grass" under the tones (correct technical term!). We start with much higher distortions so as frequencies increase, more and more of them pile on top of each other, masking any low level detail. So once again, if you think multi-bit gives you more accurate/detail, think again. It cannot do that because it is stepping on the signals themselves.

None of this is "musical either." You have a spray of harmonics from every tone in your music. Fortunately the main tones mask these so you don't hear this schiit but technically and objectively they are there, destroying the low level music signals.

We can skin this cat a different way. Let's throw square wave at both DACs. Square wave can be decomposed into a single tone and odd harmonics forever. The amplitude of harmonics changes though as frequencies go up (unlike above where they were all same level). This perhaps is closer to music where low frequencies have much more power than highs.

Here is Schiit modi 3:
View attachment 18294

First, we see proof that everything is just a sum of pure tones. Whether this is a square wave or your music, it can be decomposed into tones as the FFT shows.

With Modi 3 not being an ideal DAC, we see similar amount of noise and distortions at the bottom of the tones. Now let's look at Schiit Modi Multibit:

View attachment 18295

Look at how the distortion spikes exploded higher compared to Modi 3. Fortunately the levels are at -100 dB so you don't hear them much but they are there to destroy your low level music detail.

Conclusions
There is no doubt whatsoever whether you use the older version or new, the Schiit Multibit produces a much less faithful analog output of the digital samples we feed it. This shows up in every measurement and is in full agreement with what Schiit itself publishes. No way can you step on audio samples and destroy them yet have some claim to better accuracy. The Schiit Modi 3 easily and by a huge margin outperforms the Modi Multibit. So if it is accuracy and faithfulness to the recording is concerned, then the Modi 3 by far is your best bet from Schiit.

What about the positive reports of better fidelity of multibit products? There are multiple factors contributing to that:

1. People are sold on the idea that multi-bit has better fidelity. It is a myth that has no relationship with reality. But with folks thinking otherwise, they will hear such improvements when they test products sighted with knowledge of one being multi-bit and another, not.

2. We are pretty deaf when it comes to hearing non-linear impairments no matter how proud we are of our hearing abilities. This is why the ailments of multi-bit DACs are not heard and #1 dominates the perception.

3. Your hearing is elastic. When you get in the mode of testing audio products, you focus and you hear more detail. That happens even if you make no change to the system. So it is trivial to read into a multibit DAC that it has more detail, air, etc. But in reality, nothing of the sort is there as the measurements clearly show.

I performed a bunch of listening tests comparing the Schiit Yggdrasil against Topping DX7s. I can tell you with confidence that when you match levels and test blind, there is absolutely nothing euphonic or even different about Yggdrasil multibit DAC. My testing got truncated but from limited work, when testing very low amplitude signals, I could hear the higher distortion Yggdrasil to sound more bright. That is what distortion does: increases high frequency content due to harmonic distortion. I could not replicate this at any level close to normal listening though.

Also, keep in mind that much of the distortions I show are due to poor engineering, not because the multibit architecture. Why else would the new version have much better linearity than old? Clearly Schiit didn't think poor linearity and truncation of bits was a good idea or they would have left it alone.

Bottom line, I cannot recommend buying Schiit Modi Multibit at any price let alone at 2.5X higher than its excellently executed Modi 3 competitor.

-------------
As always, questions, comments, recommendations, etc. are welcome.

If you like this review, please consider donating funds to support these reviews using:
Patreon: https://www.patreon.com/audiosciencereview)
or upgrading your membership here though Paypal (https://audiosciencereview.com/foru...eview-and-measurements.2164/page-3#post-59054).

Without your support, I wouldn't be able to purchase products like this and test them.

why You don't make multitone measurements for topping devices? as one said: "life is not one bit"

 

[solderdude]

why You don't make multitone measurements for topping devices? as one said: "life is not one bit"

There are multitone measurements in this Schiit thread (see post #11) for Topping devices.

Bits and multitone have no relation at all.

 

[Veri]

as one said: "life is not one bit"

Unless we're all in a simulation! 0's and 1's all the way!

 

[Bow_Wazoo]

I would be interested in measurements from the current Modi Multibit 2.
Especially from the nos mode.

 

[Bow_Wazoo]

Today I tried to measure with REW the distortions the Schiit Modi Multibit 2.
As expected, the distortions were higher in NOS mode.
I was particularly interested in the effect of feeding the Modi with different sampling rates.
I use the Neutron Player player. My smartphone is connected to the DAC via USB.
As recommended by the programmer of the Neutron Player, I use the player intere oversampling. This should make the signal processing (in my case EQ) more precise.

Feeding the Modi with 88.2, or 176.4kHz in NOS mode, seems to have the advantage that I move the garbage up.

OS

NOS 44.1kHz

NOS 88.2kHz

NOS 176.4kHz

Actually, I'm not the "DAC sound" believer, but in A/B comparison with my E50, the Modi, especially in NOS mode, sound more pleasant, actually more musical and "analog".

 

[Veri]

Today I tried to measure with REW the distortions the Schiit Modi Multibit 2.
As expected, the distortions were higher in NOS mode.
I was particularly interested in the effect of feeding the Modi with different sampling rates.
I use the Neutron Player player. My smartphone is connected to the DAC via USB.
As recommended by the programmer of the Neutron Player, I use the player intere oversampling. This is probably the signal processing (in my case EQ, more precise).

Feeding the mode with 88.2, or 176.4kHz in NOS mode, seems to have the advantage that I move the garbage up.

OS
View attachment 268289


NOS 44.1kHz
View attachment 268287


NOS 88.2kHz
View attachment 268288


NOS 176.4kHz
View attachment 268286

Actually, I'm not the "DAC sound" believer, but in A/B comparison with my E50, the modes, especially in NOS mode, sound more pleasant, actually more musical and "analog".

Thanks for sharing. That spectrum really is nothing to write home about though. If anything it makes me very hesitant to buy one of schiits "multibit" stuff.
I mean compare to something cutting-edge-recent like SMSL SU-9 Pro. So, so much cleaner in noise and artifacts.

 

[Bow_Wazoo]

I think my E50 is clean enough.

 

[Angsty]

Actually, I'm not the "DAC sound" believer, but in A/B comparison with my E50, the Modi, especially in NOS mode, sound more pleasant, actually more musical and "analog".

Distortion and non-linearity can sometimes be euphonious, but it's not accurate. I sometimes prefer the sound of my vinyl records to CDs, but I'm under no illusion that the playback is more accurate than digital.

 

[Bow_Wazoo]

I've tried to catch it

Schiit Modi Multibit 2 vs. Topping E50. Multibit (+NOS) vs. Delta Sigma

Schiit Modi Multibit 2 vs. Topping E50. Multibit (+NOS) vs. Delta Sigma. Is there an audible difference? Two tracks, recorded three times. 1x E50, 1x MMb2 OS, 1x NOS. https://drive.google.com/drive/folders/1hTvIofq6LASUWn29llD94lctM7wdJv6K Have fun

www.audiosciencereview.com

 

[Bow_Wazoo]

Today I tried to make the MMb2 NOS mode visible.
I played a 1kHz test tone with different sampling rates, recorded it and analyzed it afterwards.
I noticed that as soon as the sampling rate is not 44.1 kHz, the result is practically identical to the OS mode.
Yesterday I read from someone who programs fir filters that it largely nullifies the effect of the NOS circuit if you oversample the signal beforehand (in my case neutron player). That would prove my measurement to some extent.
But I wonder why the signal is already at 48kHz, identical to the OS?

As you can see in the spectrum analyzer, and in the spectrogram, the fundamental, and the overtones only at 44.1kHz is somewhat blurry, or lil distorted, as you would expect.

Can you explain to me why this is so?

 

[Bow_Wazoo]

Here a few more measurements of the MMb2

NOS 1khz sine wave

OS 1khz sine wave

NOS 10khz sine wave

OS 10khz sine wave

NOS 1khz square wave

OS 10khz square wave

IR NOS

IR OS

 

[NTK]

Here a few more measurements of the MMb2

NOS 1khz sine wave
View attachment 269580

OS 1khz sine wave
View attachment 269581

NOS 10khz sine wave
View attachment 269583

OS 10khz sine wave
View attachment 269582

NOS 1khz square wave
View attachment 269585

OS 10khz square wave
View attachment 269584

IR NOS
View attachment 269586

IR OS
View attachment 269587

If you don't mind, would you please run a few square wave tests with the NOS DAC at frequencies close to Nyquist? If my understanding is correct, the widths of the square pulses can only be multiples of the sampling frequency. Below is my simulation of square waves (18 kHz with sampling frequency of 44.1 kHz) reconstructed using first order hold (FOH = sample and hold stair steps).

The top graph is the original square wave (18 kHz) and the digital samples (sampled at 44.1 kHz). The bottom graph is my simulated NOS DAC output. (Of course the original square wave waveform is NOT properly bandlimited, and the sampling theorem basically says that we will not be able to accurately reproduce this waveform from its digital samples.)

Thanks.

 

[Bow_Wazoo]

Yes good idea. I'll do it today.
From a metrological point of view, I find the nos mode really scary.
But as the sound goes, I have to admit that I'm simply thrilled. At least in my chain, the sound in NOS mode is changed in the exact direction I want. To me it sounded like some transient sharpening was taken away.
The music also seems to have more coherence overall.

The combination of LYR+ - MMb2 - HE-1000 Stealth (+ Harman target), sounds like a dream come true: like a hi-res analog medium.
It probably doesn't fit every headphone.
But an ultra high-resolution HE-1000 Stealth, like the one I have, harmonizes to my ears, excellent with a bit of dirt in the signal.
I can hardly imagine a naturally rather soft and relatively less dynamic headphone, like the DCA models, in this chain.
Then, I suspect, the sound would be too blurry.
For decades I gave the Amp and the DAC too little role, in terms of their influence on the overall sound. The Schiit products I mentioned, triggered a rethink in me.
The HE-1000 Stealth sounds better on the LYR+ than on the A90D.
It also sounds better with the MMb2 than with the E50. The Hifiman reminds me of the LCD 5 in many ways. I sold it after a few months because it sounded too aggressive.

My impression is that many TOTL headphones these days have such a high resolution that the transients are just overemphasized and this makes the sound too aggressive.
You can't get that with the EQ.
With tubes and NOS, apparently yes.
With such a constellation, you get a sound that I have never experienced before.
In my experience, the opposite way, i.e. ultra-clean amps + DACs, paired with headphones that don’t have quite as high a resolution, does not lead to the same result, by the way.

 

[SIY]

If you don't mind, would you please run a few square wave tests with the NOS DAC at frequencies close to Nyquist?

Shannon-Nyquist means that after you bandlimit the square wave, it will be a sine.