Saturday, 5 May 2018

MEASUREMENTS: MSI X370 SLI Plus AM4 Motherboard Audio (RealTek ALC892 codec). [And the importance of measuring the "low end".]

Image on the right shows the multicolored multi-channel "speaker" audio output. The standard "front" stereo channel is the left middle-from-top light green one.
Hey guys, I managed to find some time this week to write up the measurements on the inexpensive MSI X370 SLI Plus AM4 motherboard I described in my AMD Ryzen 1700 CPU workstation build from last year (note the other details like the Antec power supply used).

GlÜÜkZ reminded me that I had not posted the results which I had already collected on the motherboard a number of months back. As you'll see, I also gathered a bit more data to show what happens with the internal DAC when the computer is strained with high processor and GPU loads.  You may recall that years ago, I showed that indeed we can detect noise from a computer running at high load with my old ASUS Xonar Essence One DAC. Let's see what that looks like these days with integrated motherboard audio output!

The testing set-up is my standard portable ADC/laptop combo:
Front stereo output --> 6' shielded stereo phono to RCA --> Focusrite Forte ADC --> 6' shielded USB --> Windows 10 Measurement Laptop
As usual, I use the same 6-feet long generic shielded phono-to-RCA cable for measurements of devices with phono outs. Playback software was foobar with WASAPI plugin using the "WASAPI (event)" driver under Windows 10. MSI states that this motherboard uses the "Audio Boost" HD processor and uses Chemi-Con "audio capacitors" "for a warmer sound". Sure, I'll take their word for it. It's listed as a RealTek ALC892 Codec in any event and based on the datasheet, the DAC has been rated as 95dB A-weighted SNR. This already gives us some idea that we're not looking at an "audiophile-level" high performance device but should be decent for day-to-day use. These RealTek parts are commodity-priced components used in computers and they do provide an array of features like digital out, multichannel, microphone input, some DSP functions; so the aim is more about features than high-fidelity output quality.

Oscilloscope (1kHz 0dBFS square wave, 24/44.1):


As you can see, the channel balance is quite good overlaying the right and left channels. A nice looking symmetrical square waveform suggesting linear phase digital filtering. Not much ringing present. I double checked and there is no clipping of the peaks.
 
Note that the phono jacks on the back are meant for powered speakers rather than headphones. I measured the output impedance on the stereo jack in any event and found ~300-ohms. There is an internal "front audio" connector for headphones which I did not plug in because I use an external DAC with its own headphone amp so I won't be able to comment on the motherboard's headphone output abilities.

Impulse Response (16/44):



Yup, a standard linear phase upsampling digital filter is used. Phase is maintained (non-inverting). It doesn't have much "ringing" on each side suggesting that the roll-off isn't particularly steep which we can see in the "Digital Filter Composite" graph...

Digital Filter Composite:


Notice a bit of imaging ("up aliasing") present with the 19 & 20kHz test tones at 24 & 25kHz. It's clearly not a "brick wall" and will let some artifacts pass beyond 22.05kHz Nyquist as confirmed with the wideband white noise at 44.1kHz sampling rate. Also you can see a bit of intersample overload with the 0dBFS white noise signal compared to the -4dBFS. Though present, the intersample overloading actually is relatively low compared to many other DACs I've tested.

RightMark Audio Analyser:

Now let's proceed with a look at the RMAA results to consider the usual measures of noise floor, frequency response, and distortion. As usual, we start with the 16/44 test signal...

16/44:

As you can see, I've pitted the humble motherboard against excellent and inexpensive USB DACs like the SMSL iDEA (previously measured), and the not-as-excellent Audioquest Dragonfly Black (previously measured). Then we've got significantly more expensive USB DACs like the Oppo Sonica DAC and my "reference" for the last few years, the TEAC UD-501. Finally to the far right, we have the recently measured 16-year-old Sony SCD-CE775 CD/SACD player.

Numerically, notice that devices are typically functioning in a tight range at 16-bits. We can see that the X370 motherboard is on the lower end in terms of noise level, but it does score better than the Dragonfly Black for THD and IMD. Considering that this is a motherboard with all kinds of non-audio complexities, the stereo crosstalk value is not bad.

Here are some graphs with more details.


Notice that below 500Hz, the motherboard's noise level is relatively higher compared to the other devices. The frequency response is nice and flat. The "stand out" here (not in a good way) is the Dragonfly Black with its unusually elevated crosstalk and distortion results noted in the past.

24/96:
At 24/96 we head into high-resolution territory. Obviously the Sony CD/SACD player is off the list.



Notice the limitation of the motherboard. Remember, the RealTek audio solution isn't a particularly "high fidelity" chipset. This shows in the fact that it achieved <100dB of dynamic range overall, or about 16.5-bits of resolution. Yeah, it seems to be able to dig just a little deeper below a 16-bit noise floor, but just barely within the 20-20kHz audio band.

Some graphs:


No surprise. Still some funky anomalies with the Dragonfly Black.

24/192:
For completeness, we see that the MSI motherboard / RealTek chipset is able to handle 192kHz. Again, we see that it really can't do more than ~16.5-bits of resolution. Audioquest's Dragonfly Black is unable to go beyond 96kHz so not on this chart.




Jitter:

So, how are the jitter measurements for a motherboard DAC these days?



Not bad at all! Note that because the absolute noise floor is high, we can't see the usual jitter modulation pattern in the 16-bit test like we see in high resolution DACs (even with the Sony SCD-CE775). In any event, the primary signal is still >100dB above the noise floor and other spuriae on the FFT. There's also no significant "skirting" of the primary signal to suggest much low frequency jitter present.

As I have said over the years, "fears" of jitter as a significant cause of sonic degradation have been very much overblown especially these days with ubiquitous asynchronous DACs. Even an inexpensive motherboard like this does not suffer from jitter issues.

What does it look like under high CPU and GPU load???

"Way back" in 2013, when I first acquired my ASUS Xonar Essence One DAC, I was able to demonstrate that running my i7 CPU & nVidia GPU under full load could affect the noise floor of the DAC.

It is now 5 years later. The CPU is now the overclocked AMD Ryzen 1700 (running 3.5GHz for this measurement) with 8 cores and 16 threads. My GPU is the low power nVidia GTX 1050. How much noise is there if I run the CPU & GPU at full tilt measured from the analogue output from an integrated motherboard DAC!?

Here's what my desktop looked like when I measured the motherboard output under load! Notice that I'm running all CPU threads at 100% with 64-bit Prime95 (top right CPU load gadget). My web browser is open showing an OpenGL spinning globe, I have FurMark hammering the GPU creating heat. And there are a few other windows open including foobar playing the test tones...



Survey says...


Remarkably little difference. Notice that at "Max Load", the noise floor is higher but only by an average of ~2dB from 20-20kHz, achieving close to 100dB dynamic range still. No real difference with THD and IMD, a commensurately tiny 2-3dB loss in the crosstalk. I've included the TEAC UD-501 results for reference to a significantly higher priced device.

You can see the "Noise Level" change in the graph with higher computer processing load:


And what about jitter differences under 100% load?


Essentially no difference! What we see above are the 16 and 24-bit J-Test signals recorded under heavy load in red and under minimum load in green. Where the two plots overlapped, you see yellow - as you can see, there is a lot of yellow.

The bottom line is that there is no major change in the noise floor from 5-18kHz. No new sidebands. In the 24-bit test, there might be a little more noise at the base of the primary 12kHz signal under high CPU/GPU load but this is all very minimal and I cannot imagine this to be audible!

Conclusions:

I plugged the motherboard output into my powered AudioEngine A2 speakers for about an hour to listen to a few tracks from James McMurtry & The Heartless Bastards' Live in Aught-Three and Yiruma's Atmosfera.  It sounded fine although I'll stick with my usual USB DAC and headphone amp (the ASUS Xonar Essence One) for daily use on my Workstation. I can tell that the the noise floor isn't as good with the motherboard output demonstrated by slightly more hiss when I turned up the powered speaker volume knob to 100% (not a problem practically because usually I only leave the speaker volume knob at 50% and use Windows' internal volume control). With the "ear of faith", maybe the sound also isn't as "smooth" with the more polished and dynamic jazzy Atmosfera using the motherboard output. The piano notes don't decay out as naturally and the details of the plucked bass also not as "satisfying". As usual, given that switching cables take some time and this is all sighted listening, I wouldn't bet much money on these subjective impressions. :-)

Overall, obviously the RealTek ALC892 in this motherboard isn't "high fidelity". There's also only so much that an "EMI-shielded" audio processor, "Chemi-con audio capacitors", and "separate audio layers for clean left & right channels" can do. I suspect these design elements have optimized the output in any event to create a still respectable audio quality which limits sonic degradation even under heavy processing load! Although it can play audio up to 192kHz sample rate, the noise floor essentially limits the output to about 16-bits. Remember, for what it is, this is pretty good and unless one were to plug the outputs into an expensive powered speaker system, one likely would never notice the difference. However, if you send the output to a high quality headphone amp and use good "cans", in that situation you'd likely want to buy a higher quality USB DAC or sound card (like the ASUS Essence STX II). Of course high quality DACs can cost a heck of a lot more than the whole motherboard...

As you can see, even with the CPU and GPU running at 100% load, overclocked, using stock CPU fan, in a modern machine like this with integrated audio, while I can measure a little bit more noise from the analogue output, it's far from terrible. This is important because it tells us that while computer processing activity can add noise, as far as I can tell, there's no need to get neurotic about it! Furthermore, we can see that even with an integrated DAC, the jitter is unaffected by whatever software is being run. Yet more evidence that operating system tweaks (like Fidelizer, AudiophileOptimizer, even various player software) really can do nothing to substantially change the sound unless they actually changed the data (which of course means "breaking" the bitperfect nature of the stream)! I've written about some of this before. As usual, with tweaks like these where there is no plausible evidence for significant sonic differences, caveat emptor is in order. And better yet, put the onus on the companies to prove their claims - what a concept!

------------------------

To close off, I have noticed over the years that I often get posts from commenters on the blog, private messages on forums, and E-mails requesting measurements from "low end" devices like what I showed today... People seem to want to know: "What is the performance of the motherboard?", "How does an inexpensive device like the iPod or iPhone measure?", "Have you tried measuring such and such cell phone?"

This is why over the years I have posted results from items like the Chromecast Audio, laptops, phones like my Samsung Note5, tablets like the iPad, and "RETRO-MEASURE" posts with devices from back in the day. While "high end" magazines review the "latest and greatest" including devices that cost arms, legs, and kidneys, I think it is unfortunate that few actually bother to look at stuff people buy (like this inexpensive motherboard) and do direct comparisons with objective results between these different devices.Why beat around the bush with subjective reviewers saying things like "I think this new $7500 Giga-Femto DAC sounds better than my last $5000 Ultra-Pico DAC... But I haven't had the Ultra-Pico in my system since Spring last year..."?

Good grief. If the device has "deeper bass", then show me the frequency responses overlaid on one with "not-as-deep bass". If the new DAC has an "even darker" noise floor, demonstrate the lower noise level compared to others. If there's "even better microdynamics", at least show me that the resolution goes deep. I'm not necessarily saying that objective measurements will cover every nuance there is, but at least demonstrate these basic characteristics of the devices before going all lyrical about the subjective sound quality as if there is some good chance that these experiences and judgments somehow are "true".

Beyond the enjoyment of posting my findings on the blog and hopefully imparting some knowledge over the last 5 years while chatting with you guys, I must say that there is a personal satisfaction in finding out for myself the audio output quality of stuff that I have around here. I believe that it is only after we can come to terms with what we already have that we can then wisely determine what we might want to "aspire" to owning as audiophiles. The truth is, one could easily spend thousands just to come to the realization that what costs hundreds was already actually the superior performing product.

That last sentence my friends, I believe is why audiophile magazines do not make direct objective comparisons especially of rather straightforward products like DACs (eg. very rarely do we see the results of multiple devices overlaid on a graph to show which is better). Advertisers I'm sure would not be impressed to have their expensive product pitted against something much cheaper just to show that their device might be coming up short. Or how about if cable measurements show no difference between a US$5000 "directional" hose using all kinds of silver/gold content, funky dielectric materials and time-consuming weaving vs. a thin US$25 "entry level" wire.

In sum: I don't think it's lost on anyone that in the eyes of the Industry, impressions must be managed and maintained for the audiophile consumer. One of the premises is that cost and sound quality correlate - especially if something costs a lot of money because it's an "aspirational" product. The best way to discredit "cheap" devices is to not talk about it, and certainly not measure it or compare it directly to the "high end". In some circles, the "logos" of the Subjective Golden Eared Priesthood must be maintained to keep the faith since these testimonies hold advertising potency.

No doubt I lost the supposedly mainstream audiophile "faith" long ago (I do believe in time, the actual "mainstream" audiophile is shifting in his/her philosophical leanings). And I think in searching to be a "rational audiophile", I believe I have found a path more fulfilling... Wishing you the best in whatever path the audiophile hobby takes you!



Have a great month of May... Hope you're all enjoying the music!

BTW: I just watched Avengers: Infinity War today. I've watched most of the Marvel Cinematic Universe flicks but by no means a huge fan. Nonetheless, this is an impressive movie. A good demonstration of how important having a complex "bad guy" is!

11 comments:

  1. Nice review Archimago! I did some similar CPU/GPU tests before your Lynx L22 review:
    https://hydrogenaud.io/index.php/topic,100481.0.html
    So the observation is GPU type and motherboard power management settings can somehow affect the measurement.

    ALC892 is commonly found on budget motherboards, I also have an AsRock board at $62 with ALC892
    https://hydrogenaud.io/index.php/topic,114125.msg940070.html#msg940070

    From your ALC892 test I can see a good sign that your motherboard has some good design and improvement over the older generation motherboards.

    I see you mentioned about the CS8420 SRC chip in LynxL22, my old Roland interface also uses this chip but the reclocking quality is far inferior to my humble soundblaster X-Fi, also you can see X-Fi's reclock SRC is minimum-phase for a good reason since the main DSP chip itself is a synthesizer and effect processor.
    https://audiosciencereview.com/forum/index.php?threads/digital-distortion.2059/#post-55543

    As you can see in some replies of the post above, someone said J-Test should be performed with analog loopback so I did a follow-up analog test to prove my point. The reclock of CS8420 on my Roland is so bad that it can even affect analog measurement, I really don't know what is going on:
    https://audiosciencereview.com/forum/index.php?threads/review-and-measurements-of-ifi-ipurifier-s-pdif-digital-audio-filter.2189/page-2#post-59502

    Knowing Lynx is a very reputable company I guess that CS8420 should not perform as bad as my Roland.

    ReplyDelete
    Replies
    1. Thanks for the note Dtmer.

      Nice measurements of the difference between load and idle as well. And good test with that old 2001 Roland DAC. Yup, parameters like jitter even out of a modest motherboard like this has certainly evolved nicely over the years!

      Delete
    2. ALC1220 even supports DSD with some insane specs.
      https://www.gigabyte.com/mb/am4/audio

      Would like to see a real measurement with Audio Precision.

      Looks like professional gamers need 114dB SNR ADC for voice chat, CDs must sound bad for them :D

      Delete
  2. Hi Archimago, could you confirm if the Realtek ALC892 is clocked at 48kHz ?

    ReplyDelete
    Replies
    1. My ALC892 can show some measurable improvement when resample from 44.1k to 48 and 96k using SoX.
      https://hydrogenaud.io/index.php/topic,114871.0.html

      But honestly if people really care about such a tiny improvement they should just spend a bit and buy an inexpensive DAC like the SMSL iDEA which Archimago also reviewed.

      My motherboard also has an optical output and my ALC892 is bit perfect at 44.1 and 48k based rates, verified by recording the output with another interface and null test so if you use the optical output there is zero impact.

      Delete
    2. Thanks for the further results on the ALC892.

      Yeah, the chip can handle 44/48/88/96/176/192 without resampling as per the results above and verified with Dtmer's results. The main limitation is the noise floor rather than the ability to handle sampling and playback of the ultrasonic material.

      Delete
  3. Most PC interface are natively clocked at 48kHz, you answered my question, the quality of upsampling is adequate.

    ReplyDelete
  4. I was a little sad to see you reviewed the ALC892. As was said, it's good to see it's better than older implementations, but really... most high-end motherboards feature ALC1220. This one supposedly performs much better! (would have loved for you to review that one but oh well).

    ReplyDelete
    Replies
    1. Another great post. This board has been a blessing to reasonable folks interested in this great hobby. As most of us are on a budget, you've saved us from going down some of these audiophile rabbit holes:)
      Here's a topic I hope you will consider: I just purchased a Gustard X22 dac (a Chinese product o absolutely love). But this thing sends a seriously hot signal (as many modern dacs do). I feed this to my Parasound P5 preamp then to a Parasound A23 power amp.
      The problem is waaaaayyy too much gain and I can hardly get the volume past 9:30 on the volume of the preamp. I'm reading that this excessive gain issue is pretty common.
      I've resorted to some Shure inline attenuators to my xlr cables and they've done the trick..
      .the question is at what cost to the sound.
      I'm sure many fight this situation so I would love to hear your suggestions/reports on attenuation (digital vs analog volume controls, attenuators, etc...)
      Thanks, again, for this service you provide!

      Delete
    2. Thanks for the note guys.

      At some point when I have an ALC1220 motherboard, I'll be sure to test it out :-).

      Delete
  5. Dear
    I am interested in how you perform your tests.
    I see that you use a Focusrite as AD converter, but I suppose, being Focusrite a capture card, you need additional software... it is an out of the box solution.
    Can you tell me how to do these kind of tests?
    Thank you

    ReplyDelete