Review and Measurements of Benchmark DAC3

[March Audio]

The THD+N vs frequency graph does not seem to match SINAD. Isn't 0.0005% -106 dB? That would not match a SINAD of 112 dB, would it?

I concur with your calculation.

 

[amirm]

Now the unbalanced performance would be interesting.

I showed linearity in my graph above. I also briefly pulled up the dashboard and it too is better numbers than the other unit I tested.

 

[March Audio]

Not here because I am testing balanced output.

but it wouldnt it be 4v rms (+6dB) if you were trying to compare to output levels of typical"domestic" DACs, which begs an interesting question about testing of these "pro" interfaces. Should we include a low level test?

 

[HifiGuy]

Wouldn't it make more sense to test at 2V?

Sorry, thats why i recommend to use the calibrated mode. Think about that.

 

[HifiGuy]

And it is important to configure the dac in a right way. Use USB mode 2 and use the right driver. Maybe amirm you should read the manual first. Their are a lot of thinks to take care.

Greetings Fu

 

[Krunok]

Sorry, thats why i recommend to use the calibrated mode. Think about that.

You are right.

 

[Thomas savage]

but it wouldnt it be 4v rms (+6dB) if you were trying to compare to output levels of typical"domestic" DACs, which begs an interesting question about testing of these "pro" interfaces. Should we include a low level test?

Yes we should (imo) as we want to be testing for potential use in a home system so all things being equal a 4v rms seems a good addition.

 

[raband]

First "point of call"/test (imo) should be against the products claimed specs or sales pitch/advertising.

Then the rest of the readings to see where they fall.

 

[amirm]

And it is important to configure the dac in a right way. Use USB mode 2 and use the right driver. Maybe amirm you should read the manual first. Their are a lot of thinks to take care.

Greetings Fu

If I read the manual, what is left for you all to do???

The latest tests were run in USB Mode 2 as it is required in the installation of Benchmark drivers. Since all of my tests are at 44.1 and 48 kHz anyway, the USB mode makes no difference as I showed earlier in the thread. And again, the driver makes no difference whatsoever. ASIO4ALL is bit-exact.

BTW, when I tell you I hate installing drivers, there is a reason. See this pop up after I installed the Benchmark drivers:

I scratched my head, checked everything twice and nothing would make it show the Benchmark. Then I see that the Benchmark's ASIO interface is alive and well and the above applet is broken in not recognizing its existence.

With so many drivers trying to install the thesycon drivers, I have a mess on my hand already. Look at this list from a few drivers I have installed:

Three separate XMOS drivers? And exactly which device is TUSBAudio ASIO Driver talking to?

There is a reason Microsoft creates class drivers and companies should follow suit: to avoid this mess.

Anyway, the manual was read with respect to matching level and that is why I used the jumpers to reduce output by 10 dB, both in the original review and this one. And based on follow on recommendation by John.

Also, these reviews are interactive. I expect you all to read them and comment for what is missing. As you see, I almost always run additional tests to address points people raise. It is not like a print magazine where reviews are published once and that is that.

 

[SIY]

Also, these reviews are interactive. I expect you all to read them and comment for what is missing. As you see, I almost always run additional tests to address points people raise. It is not like a print magazine where reviews are published once and that is that.

^This. It's what got me here and keeps me here.

 

[Thomas savage]

Not reading the manual is mandatory for all males, besides that amirs ‘ deliberate ‘ mistakes keeps us all intrested and feeling clever.

God bless our humble benevolent king

 

[maverickronin]

BTW, when I tell you I hate installing drivers, there is a reason. See this pop up after I installed the Benchmark drivers:

snip...

Totally with you on this one. It's bad enough just having a few on a normal machine for stuff you user on a regular basis but a test rig that goes through as many DACs as you do would just be a nightmare.

I'd probably get a separate test machine and just reimage it after every couple new drivers.

There is a reason Microsoft creates class drivers and companies should follow suit: to avoid this mess.

To be fair to those other companies Microsoft was nearly a decade late to the UAC2 party.

 

[HifiGuy]

If I read the manual, what is left for you all to do???

The latest tests were run in USB Mode 2 as it is required in the installation of Benchmark drivers. Since all of my tests are at 44.1 and 48 kHz anyway, the USB mode makes no difference as I showed earlier in the thread. And again, the driver makes no difference whatsoever. ASIO4ALL is bit-exact.

BTW, when I tell you I hate installing drivers, there is a reason. See this pop up after I installed the Benchmark drivers:

View attachment 14235

I scratched my head, checked everything twice and nothing would make it show the Benchmark. Then I see that the Benchmark's ASIO interface is alive and well and the above applet is broken in not recognizing its existence.

With so many drivers trying to install the thesycon drivers, I have a mess on my hand already. Look at this list from a few drivers I have installed:

View attachment 14236

Three separate XMOS drivers? And exactly which device is TUSBAudio ASIO Driver talking to?

There is a reason Microsoft creates class drivers and companies should follow suit: to avoid this mess.

Anyway, the manual was read with respect to matching level and that is why I used the jumpers to reduce output by 10 dB, both in the original review and this one. And based on follow on recommendation by John.

Also, these reviews are interactive. I expect you all to read them and comment for what is missing. As you see, I almost always run additional tests to address points people raise. It is not like a print magazine where reviews are published once and that is that.

Hi amirm,

i tought a lot about wheather i write something or not. I decided to do it and clarify some point on your measurements.

The name of forum is audio science review. For me a scientific approach must be reliable, repeatable and based on facts. Any information that needed to repeat the measurements should be documented.

When i take a look on all the measurements, starts with stereophile, yours, jude and now yours again. I can not see any repeatability . A correct way to measure would be that your system is in the same state, this mean it is needed to make an image and restore that on every measurement. After restoring your system you can install any driver without problems. Yourself writing that you have problems with your systems, this is for me an indicator that your system is not reliable.

If you have two different USB-Modes and different loudness levels this should be configured right. The USB 2.0 Mode runs on a different voltage and need on windows maschines the right driver. Not every dac manufacture offers a driver, Benchmark does and it should be used.

Now to the facts: if you measure once 2 VRMS and than 12 VRMS, it is clear that the SNR changes. A lower voltage can not offer the same SNR as a higher voltage do. Comparing two dacs without the same output is to compare apples and oranges.

Greetings Fu

 

[garbulky]

Well here we have me asking Amir to measure balanced DACs using their balanced inputs. And he thinks this is unnecessary. However even a “state of the art” DAC like the Benchmark, does differently on its balanced input. Thankfully Amir decided to measure both now. Hopefully it will be standard to measure both. As we have two documented cases of balanced DACs performing better in balanced drive.

garbulky said:

So we see not only that he presented the measurements of a balanced dac using its unbalanced outputs which isn't the most optimal way to measure a balanced DAC (becaused the balanced out measures better in this case).

As I noted elsewhere, this is an incorrect argument. Just about any DAC silicon has balanced output. Yet many manufacturers put unbalanced outputs on them and using those, I am able to get superbly better results than Schiit multibit. You are being fed and PR response line and taken advantage of the fact that you are not a design engineer and don't know what I just explained about other DACs.

Conclusions
The Benchmark DAC3 as expected is a state-of-the-art digital to analog converter. Other than one set of noise spikes in jitter test, the rest of the measurements show exceptional performance. No glaring faults are seen at all. Its higher output level can be useful in room EQ applications to boot.

So of course the DAC3 goes on my recommended list.

Here is the unbalanced tests from Benchmark DAC3.
Unlike balanced output, the unbalanced response is considerably worse with one channel worse than the other.

As a way of reference, I put the Topping D50 through the same test and as before, it has textbook output to -120 dB.

While the benchmark DAC3 generally measures well on its unbalanced output, reproduction of low level signals is not very accurate. There certainly is no excuse for a $250 DAC to beat it in such tests as linearity.

So sadly, I cannot recommend the Benchmark DAC3 for unbalanced output.

 

[amirm]

When i take a look on all the measurements, starts with stereophile, yours, jude and now yours again. I can not see any repeatability .

There is actually. See this thread where I even match results of someone using a completely different analyzer: https://www.audiosciencereview.com/...ts-of-schiit-yggdrasil-dac-inconsistent.3812/

It is true that there are parameters that cause deviations from one measurement to another. For that reason, I have been more than willing to try to replicate other people's measurements if they are different than mine.

Now to the facts: if you measure once 2 VRMS and than 12 VRMS, it is clear that the SNR changes. A lower voltage can not offer the same SNR as a higher voltage do. Comparing two dacs without the same output is to compare apples and oranges.

Where possible, I work very hard to equalize outputs. See my reviews you will notice that levels are close to each other in the pairs compared. This is from the very review here in Post #1:

As you see from the height of 12 kHz tone in the FFT, the two DACs have almost identical output.

When there is a deviation, I note in the graph.

Matching using volume is something I do with headphone amps but with DACs, implementation of volume control affects DAC performance and some manufacturers complain about that. To wit, John thought I had used volume control in my original review and protested that until I showed that I had used the jumper.

Ultimately there are restrictions in what we/I can do. I can unify the industry to output the same voltage on all DACs.

 

[mindbomb]

If you are talking about the sinad numbers, that is testing thd+n, not snr, and that is usually worse with a higher voltage. And both the rme and benchmark were at 12 vrms.

 

[Kal Rubinson]

Well here we have me asking Amir to measure balanced DACs using their balanced inputs. And he thinks this is unnecessary. However even a “state of the art” DAC like the Benchmark, does differently on its balanced input.

Can I assume that terms "balanced inputs" and balanced input" are typos here? The Benchmark DAC3 has no balanced inputs, only balanced outputs.

 

[John_Siau]

Yes, they are pretty much the same. Interesting..

At the end you are right, but it would be nice to know.
@John_Siau, maybe you can share your thoughts on the cause of this deviation from linearity?

My guess is that this is noise modulation or crosstalk. Normally the linearity curve should stay flat and then bend upward when the test tone disappears into the noise floor of the measurement. But, when you see the output level dip lower than expected, before subsequently curving upward, this is an indication that there is noise modulation or crosstalk.

To test for crosstalk: Try turning the second channel on and off, and try inverting the polarity of the second channel. If the shape of the curve changes, crosstalk is influencing the measurement. This is not a linearity issue. It may just be a very low-level crosstalk issue.

If narrowing the band pass filter preserves the dip in the curve, but moves it lower in amplitude (moves it to the left), the cause is noise modulation (usually inside the DAC IC) . Many converter chips generate slightly less noise when producing very low output levels. This change in output noise may only happen at very low output levels (near 1 LSB). It is caused by fewer 1-bit switching transitions when levels are very low. The slight change in noise raises havoc with linearity measurements. This is not a linearity issue. It is noise contaminating the measurement.

The lesson from all of this:

Once noise begins to bend the linearity curve, anything below this point is highly suspect.

 

[garbulky]

Can I assume that terms "balanced inputs" and balanced input" are typos here? The Benchmark DAC3 has no balanced inputs, only balanced outputs.

Yeah typo. Sorry

 

[John_Siau]

I don't agree with this - when the pot is cranked to the max that is exactly the point to measure how the device works.

The device should be tested at +24 dBu at 0 dBFS which is what you will get at the HT setting. Turning the volume full CW will improve the SNR test results but it will eat into the 3.0 dB of headroom that we provide for the DSP. The last 3 dB is available for boosting a weak signal, but should not be used for normal listening. THD will still be excellent, but the DSP headroom is diminished inside the ESS chip but not prior to the ESS chip. Our filters prior to the ESS chip always have 3.0 dB of headroom above 0 dBFS. Most other converters have no headroom above 0 dBFS.

Amirm should test D/A converters for intersample clipping. Most are ugly. The Benchmark DAC2 and DAC3 converters will properly reproduce intersample peaks above 0 dBFS without clipping.

I have written several application notes on the topic of intersample overs. They are available here: https://benchmarkmedia.com/blogs/application_notes/tagged/inter-sample-overs