Benchmark DAC3 B Measurements

Content and analysis made possible thanks to support from, and supporters.

This unit was loaned to me by a friend for testing.


I’ve discussed my thoughts on the subjective aspect/sound of the DAC3B HERE, and in this post we’ll take a look at the internals and objective performance of the DAC.

By the way, if you want to chat to me or others about the stuff in this article (or anything else audio related) do consider joining the discord!
Or if you want to support me in doing more of this stuff consider joining my patreon (link on the right sidebar) and you’ll get access to the private telegram chat!


The Benchmark DAC3B is based on the ESS 9028PRO chip, but features a few quite interesting design aspects you wouldn’t normally see in other DACs.

DAC3B Internal

Perhaps the most unusual of these is the output pads. The DAC3 is aimed at professional markets, and because of this, the stock output voltage is just over 12 volts!
This will be far too high for many consumer applications. My Holo Serene preamplifier for example has a max input level of about 8.5V. And some amps like the THX 789 have max inputs of only 1.7V (in high gain, 7V in med/low gain). So the stock voltage of the DAC3 will be too high and clip many consumer amps.

To address this and allow you to choose the level that suits you best, there are some jumpers which allow you to select a -10dB or -20dB attenuation via a passive resistor on the board.

Benchmark has also developed a powerful PLL system they’re calling ‘ultraclock’ to attenuate jitter from poorer quality sources, and have added 3.1dB of digital headroom to the DAC chip to prevent clipping of ‘intersample overs’.

This is an issue that arises due to how PCM and oversampling works.
PCM is a sampled audio format, meaning you have a sample at regular intervals (44100 times per second for redbook CD quality audio) describing what level the waveform is at for that specific point in time, and your DAC must then interpolate or play ‘connect the dots’ to reconstruct the original waveform.
The maximum level a sample can be is -0dB or ‘full scale’.
The problem is: what if all the samples are legal, and below -0dB, but the waveform itself is not?

EDIT: It seems this next part caused a bit of confusion and I’d like to clarify, the issue I’m about to demonstrate is NOT exclusive to the X18. This is something that is present on a LOT of DACs and it is a test I will be conducting on all DACs moving forward.

We can create an example of this ourselves, by creating a sine wave at exactly a quarter of the sample-rate with an amplitude such that all the samples (shown as green squares below) are below the red -0dB line, but the actual intended waveform goes above this -0dB level.

When we feed this audio to a DAC, and it oversamples, it can only convert above 0dB if it has internal headroom to begin with.
Once all bits in a sample are 1, (So 1111111111111111 for a 16-bit sample), it cannot go higher. Meaning all those extra samples that are SUPPOSED to be above 0dB instead end up staying at a flat line on 0dB cause the DAC cannot describe a level higher than that.
As a result, anything above this -0dB line just gets chopped off, causing massive distortion.
Below is the Gustard X18 playing back a file with +3dB intersample overs. You can see that the waveform is chopped and we get significant distortion as a result.

To be clear, no oversampling software is being used. This is just the prepared 44.1khz file being fed directly to the DACs, and letting their internal processing do everything as it would with any normal music.

Gustard X18, +3dB intersample overs test

In fact, even with only +1dB overs the DAC still sees added distortion, because there is no headroom internally at all and so ANYTHING above -0dB will get clipped.

Gustard X18, +1dB intersample overs test

Whereas the DAC3 has internal headroom, and so if we try +3dB intersample overs on it there is no clipping at all and it reproduces the signal as it was intended to be reconstructed (other than the DACs own harmonic distortion below -110dB).

Benchmark DAC3B, +3dB intersample overs test
Note: the random spurs are not related to this test, this DAC just has those there in one channel all the time as shown later on in this post.

You often see intersample overs addressed in dedicated oversampling hardware/software such as the Chord Mscaler which adds 2.8dB of headroom, or HQPlayer which has adjustable headroom and will indicate if intersample overs clipping is encountered, but it’s not often we see it addressed inside DACs themselves, even though it REALLY should be.
Manufacturers may be reluctant to do so because sacrificing 3dB of potential dynamic range or SINAD means they won’t get to top ASR’s list anymore, but in my opinion this is a much more important problem to address than having a tiny bit of extra SINAD.

And whilst intersample overs SHOULDN’T be something you encounter all too often with real music, mastering isn’t always perfect and they will absolutely occur.
In fact just looking at a few of the more ‘brickwalled’ tracks I have in my library, intersample overs are present all over the place:

This is a test I will be adding to my standard testing sequence going forward on all DACs.


Test Setup:

– Audio Precision APx555 B-Series Analyzer with 200kOhm input impedance set unless otherwise specified
– USB Source: Intel PC via intona 7055-C isolator
– Measurement setup and device under test are running on an AudioQuest Niagara 5000 power supply
– Audioquest Mackenzie XLR and RCA interconnects
– DAC in UAC2.0 mode, with Benchmark ASIO driver
– Intona Reference Impedance Characterized USB Cable
– Exact analyzer/filter configurations for each measurement are detailed in the full reports

Full Measurement Reports:

Reports available here:

Dynamic Range (AES17): 119.6dB

SNR: 122.1dB

Noise Level RMS (20-20khz): 3.013uVrms

Noise Level RMS (20-90khz): 7.135uVrms

DC Offset: 752uV active, 111uV idle

1khz 0dBFS Sine, 0dB pad, XLR Out:

This is the DAC3B ‘stock’, with no output padding. However for the remainder of this post I will be using the -10dB pad for three reasons:

– The analyzer itself performs better in the 5v range than the 20v range so the analyzer will influence results less
– Most consumers will need to use the DAC3B with the -10dB pad as 12.22V is too high an output level for most amps
– It’s more similar to other DACs I’ve tested

1khz 0dBFS Sine, -10dB pad, XLR Out:

1khz 0dBFS Sine, DSD128 XLR Out:

The DAC3B performs notably poorer with DSD content than with PCM. It seems that the ultrasonic filter doesn’t start attenuating until quite far out too:

-90.31dBFS 1khz sine (96khz capture bandwidth):

Filter Ultrasonic Attenuation:

The oversampling filter seems to be one of the stock ESS sinc linear phase filters.
But here we can also see a notable level of spurious noise in one channel. This occurs even when idle and we can see it a bit clearer with 1mhz bandwidth as shown below

Idle Noise FFT:

1Mhz bandwidth

We can see that aforementioned spurious noise, and additionally, two spikes up around 800khz which is likely the switching PSU frequency.

20khz bandwidth

THD+N vs Frequency:


CH1: -105dB, CH2: -109dB

No ‘ESS Hump’


With Bandpass filter
Without Bandpass Filter



44.1khz USB
48khz USB

Jitter performance is ok but with quite a lot more in one channel than the other. I’m not entirely sure why this would be given as it is a single chip and so there should be little to no difference.
Benchmark also claims their ‘UltraClock’ system to be ‘jitter immune’. This is something that several other companies such as Chord have claimed but it has not proven to be true when tested.

48khz Coax
48khz Coax with 25ns of 1khz sine jitter added

As you can see, there are some added spurs at 10khz and 14khz when jitter is added to the signal. So the ultraclock PLL is NOT ‘immune’ to jitter.
Whether they are meaning immune with respect to the level of audibility, I’m not sure. But in any case, jitter on the incoming signal does affect the DAC.



2 thoughts on “Benchmark DAC3 B Measurements”

  1. The sound because of 18 Volts on the originally National chips is very good, considering how long ago they were introduced. The sound through my JC2 preamp notably had increased detail and transients when using balanced cables. The cables cannot be std. microphone cables for evaluation. I tried premium starquad German cables which were not audiophile sound quality. I converted from my original Atlas Marvros RCA’s to the balanced version. $800 for a used 1 M pair from the Ukraine. Still, 40% of retail for the DAC HGC. On recordings having CD and the record, they are almost identical, except for the bass content. I won a DAC2 and upgraded to the DAC3 HGC. CD player coax in and Bluesound 2i into the 2nd Coax. DGG content on Qubuz is excellent, maybe better than CD as most DGG is HR. This was about the first unit to use the later DAC chip from ESS. Obviously, it a different sound to an R to R, heard from Mr. Soekris at AXPONA 4 or 5 years ago. For this listening review, balanced or unbalanced cables? Headphones or your monitor speakers? My room listening is with the Gradient Revolution speakers. The only annoying part is that the LED’s cannot be dimmed so must be covered with LED filter plastic. The remote is nice metal remote, so easily can be switched between CD and streaming. This is a reliable unit which shuts off if there is a power interruption to prevent an on surge. There is never any harshness to cause ringing on your ears on playback. There is no hum from it due to the switching power supply. It can be used as a preamp by changing from the fixed volume control function to using the remote. These last features are not on the version under review. So, that version has limited applications. With top notch balanced cables, many would find the sound satisfactory. For information, ATLAS cables from UK are
    crimped, so do not use any solder, even XLR connectors.

  2. Intersample Overs:

    It is wonderful to see a review that includes a DSP headroom test for intersample overs. The clipping of intersample overs is probably the most audible artifact produced by most D/A converters and yet the vast majority of the industry seems to be ignoring this problem. A simple analysis of the mathematics will show that a reconstructed waveform can reach 3.01 dB above 0 dBFS, even when the signal is properly band limited. In most 44.1 kHz recordings, this condition occurs many times per second, even when the recording is not squashed with excessive studio processing. Sample rate converters (SRC) and oversampling D/A converters must have a headroom of 3 dB above 0 dBFS, or they will clip many times per second when playing 44.1 kHz recordings. This problem is less severe with higher sample rates, due to the fact that the worst case occurs at an input frequency equal to 1/4 of the sample rate. In a 44.1 kHz system, the worst case occurs at 11.025 kHz, and there can be substantial energy at these frequencies. In contrast, the worst case occurs at 22.05 kHz in an 88.2 kHz system. There is not as much energy in the music at 22 kHz, so one could argue that less headroom is required at higher sample rates. Nevertheless, adding 3 dB of headroom will solve the problem at all sample rates.

    Avoid high-sample rate recordings that have been upsampled from 44.1 kHz originals. In most cases, these upsampled versions will have SRC clipping artifacts baked in.

    Jitter Immunity:

    Jitter immunity can be measured in terms of jitter attenuation. In the review, 25 ns of jitter was applied while playing a 10 kHz test tone. If there was no jitter attenuation, 25 ns would produce a pair of sidebands that would each measure about -58 dB relative to the test tone. In the test of the DAC3, the resulting sidebands measured about -145 dB relative to the test tone. This is equivalent to 1.4 ps jitter at the sampling clock. In other words, 25 ns is attenuated to just 1.4 ps. This is a ratio of 17,857:1 or 85 dB. Notice that we get about the same answer if we measure the relative difference of the sidebands. The difference between -145 dB and – 58 dB is 87 dB. So we can see that the jitter attenuation at 1 kHz is somewhere between 85 and 87 dB. This jitter attenuation is outstanding, but more importantly, it means that the jitter-induced sidebands are well below 0 dB SPL in any playback system (unless your playback system is producing an SPL exceeding 145 dB SPL).

    So from the above we can say that jitter-induced distortion is absolutely inaudible (below 0 dB SPL) in any playback system driven by a Benchmark DAC3.

    If you run the AES jitter tolerance test, you will find that the DAC3 never shows an increase in THD+N due to jitter (see “Graph 14” on page 60 of the DAC3 HGC manual).

    If you plot an FFT while running the AES jitter tolerance test, you will find that the DAC3 never shows jitter-induced sidebands exceeding -140 dB relative to the test tone (see “Graph 15” on page 61 of the DAC3 manual).

    These two test published in the DAC3 manual agree with the -145 dB sidebands measured in this review.

    Size of soundstage:

    Errors in interchannel phase and interchannel frequency response will always expand the apparent width and depth of the soundstage. Some reconstruction filters can cause frequency-dependent phase distortions which can enlarge and blur the locations of instruments in the stereo image. The DAC3 will not add a false sense of depth or width to recordings that lack space. It will also not expand the apparent size of a point source. The DAC3 is spectacular when reproducing recordings that have captured the natural depth and width of a recording space. On the other hand, a pan-potted stereo image will sound like a flat wall of sound stretched between the two stereo speaker.

    Reconstruction Filter:

    The DAC3 does not use any of the ESS filters. Our reconstruction filter is implemented at a very high oversampling ratio and this is done outside of the ESS chip. The reconstruction filter has a linear phase response which we believe is important for accurate reproduction of the width and depth of the recording space.


Leave a Reply

%d bloggers like this: