Computer Noise - Affect on Audio Quality

Hi Danny, the dev team has mentioned this concept of a noisy PC. Is this concept true in today’s modern computers?

My opinion is that it’s 2016, most computers are engineered to avoid this “noise” that may have been present in older and poorly shielded sound cards. It’s conceptually “scary” to anyone attempting to acheive some level of sonic bliss. Further, I can’t find any evidence of noise in my playback system that stems from Roon, my OS, or my hardware.

Can you provide some evidence that there is audible noise in your playback system?

Also here is good article to think about:

Thanks. :smile:

@brian and I have definitely heard noise coming from modern computers. I was audibly affecting a DAC multiple feet away, completely disconnected, just by scrolling a webpage in a browser (both Chrome and Safari). Scrolling in a web browser that has already loaded the page is purely GPU/CPU/RAM, so unlike the “BlackBerry next to the speaker”, no wireless technology was involved. The noise being heard was directly correlated to how fast and often I scrolled.

The last time I remember experiencing this was on a 1st generation MacBook Pro Retina, which is only a couple of years ago. @brian never had that machine, so his experience was on a different machine.

That link is interesting, but he didn’t push the system. Drive that GPU/CPU/Ram/WiFI/Ethernet/USB3 really hard, and then let’s check out that noise level.

Not that this is a worthy contender for audio quality, but the headphone jack on my current Macbook Pro Retina is noisy as hell when my CPU usage cranks up. It’s so bad (or maybe I’ve become so spoiled) that I won’t plug headphones into the machine unless it is via a decent USB connected DAC/amp. I’m usually using a Bel Canto 2.5 on my desk, or a Meridian Explorer on the go.

1 Like

I currently use ARM based Wandboard Quads running Arch Linux with Squeezelite as transports/endpoints. The Wandboard’s USB was measured to perform better than the Squeezebox Touch (which is itself pretty good). I’ve since powered the Wandboards using linear PSUs. When Roon Bridge and Roon headless server for Linux ships my intention is to switch from Squeezelite to Roon Bridge and to install Roon server on the same Arch Linux box that holds my music collection. That’s pretty much all that will change in my audio chain. My tablets will remain the only means of interacting with Roon’s UI.

I have a headphone dac+amp tucked under a large LCD monitor (that is also a thunderbolt, gige + usb hub) on my desk. It picks up audible analog-domain junk from the monitor. Play back silence, plug in some sensitive IEMs, use the PC, and I can hear myself computing.

I see one interesting thing in Archimago’s measurements: the noise floor on the Windows machine is ~6-8db higher than on the Mac. If the computer had no influence at all, these should be identical, since the DAC and the digital silence signals were identical…but they aren’t.

(I’m only paying attention to the silence measurements, not the ones with frequency content).

Obviously the difference in Archimago’s test was all below a -140dB level–which we shouldn’t be able to hear, but that’s not the point. If there’s truly no impact, there should be no impact at all.

His test was deliberately clean–it looks like he was even running his completely solid-state mac from the battery to avoid the impact of the switching PSU, which is good when you’re comparing sample rate converters, but not as good when you’re looking for sources of analog interference. The real world is messier.

I can appreciate your experiences but mine do not correlate with yours at all. Are you using off the shelf PCs? I’ve never used a computer I did not build. My machine is dead quiet, I detect no “computing noise” at all no matter what other task I’m running while playing music back. Running an Asus Xonar Essence STX straight out to studio monitors.

1 Like

@Rugby, in my examples, it’s always been a MacBook Pro or a Dell laptop, both obviously off-the-shelf.

@brian’s issue above was with a monitor!

o0OBillO0o: you should visit the Computer Audiophile forums - many expert people there report hearing and measuring major effects due to insufficient electrical insulation. First it got me depressed and I was close to going down the way of changing all my setting & spending lots of $$$, but financial reality brought me back and I did my best with what I have. Surely there is some electrical “noise” but as long as I am not bothered by it I decided to just leave it as is…

1 Like

We do have to keep this in context of real world effects though. The difference between a -140dB noise floor and a -132dB noise floor is utterly irrelevant to you, I or anyone else. For example most DACs have a theoretical SNR of around 120dB. Unless you’re worried about hearing air molecules vibrating there is no possible audible effect.
Moving your head from side to side at the listening position would be more significant.


The difference between a -140dB noise floor and a -132dB noise floor is utterly irrelevant to you,




or anyone else


There are many people out there for whom noise at that level might be subjectively relevant, even if the numbers say that it shouldn’t be detectable.

For example, there are seem to be many people who have a “favorite” noise-shaped dither, but use 24bit (or higher) output. This is a difference right in that same ~138-144dB ballpark.

For example most DACs have a theoretical SNR of around 120dB.

I agree that we can’t hear sounds at -132 or -140dBFS (and I think I made that clear in my post).

That said,

Human ears have a theoretical SNR of about 95dB, but by now we’ve figured out that the 96dB dynamic range of Redbook isn’t enough.

We’ve also figured out that while humans can only hear frequencies up to 22kHz, 44.1kHz isn’t enough. Both for direct reasons (psychoacoustic significance of impulse response, binning of ultrasonic content), and for indirect reasons (mastering at 44.1kHz requires steep filters that do damage).

Ladder DACs run into the limits of resistor manufacturering somewhere in the 18-20 bit range, but there are 24-32bit Ladder DACs on the market, and people still buy them and enjoy them more than the one they were using before.

DSP processes are routinely optimized to be much better than they need to be, too. Look at the stop-band attenuation of most low pass filters used for sample rate conversion. It’s usually way better than the science suggests that it needs to be.

Over-engineering is a way of life in this space. If we want to make a product that satisfies the people who are hunting down computer noise as much as the people who are staring deep into the eye of the spectrogram, we have to address many ways of thinking about and experiencing audio.


Nope I have to disagree. It’s a purely numbers game for credulous audiophiles to wring their hands over.

1 Like

I share this ^^ viewpoint now, could change though.

I have a Macbook Air and a iMac, both connected to USB DACs, and airplay devices. I believe have no noise from the computer, that I can tell. It could be there? How would I test for noise?

It could be there? How would I test for noise?

Subjectively: ABX test between Computer->DAC and Computer->Network->Bridge->DAC.

Objectively: Plug Computer->DAC and Computer->Network->Bridge->DAC into a scope with plenty of dynamic range, play the usual test signals, and analyze/understand the differences.

The first one tells you if the difference matters to you personally. If you iterate the test over a population, you can determine whether it makes a difference to anyone.

The second one tells you what the differences are, and the nature/degree of differences present. There will always be some if you zoom in far enough. That’s just the nature of analyzing analog systems.

These are not easy/cheap tests to run, so in the real world, audiophiles don’t bother. Responsible HiFi manufacturers routinely do both during product development.

When we first brought our PC-in-a-box Sooloos products to Meridian in 2008, one of the first things they did was to put each product inside a metal room, measure what kind of EM/RF interference it was emitting, and measure how it impacted performance measured in the analog domain–and they found the results pretty amusingly bad compared to their usual standards.

I’m not an EE, but I know when to respect another engineer’s measurements and do my part fixing the problem. The end of that road was marked by the release of lightweight networked endpoints that fully met Meridian’s performance standards.


I use the optical out from the motherboard of my Windows 10 PC to my receiver. Would there be a recommended soundcard that I could purchase that would be better in terms of noise ? Or is the fact that it’s in the PC case itself the issue here ?


Are you actually suffering from noise ? Really ?
And if you’re using an optical connection, you can’t be suffering from EMI anyway, so you should be fine already.

Not that I am aware of but now i’m worried !! :scream:

Only kidding, but I am curious if a dedicated soundcard could improve my system ? I’m guessing not…

Slightly tongue in cheek @Jaap74, but I suspect that removing the soundcard (that’s the humorous part) and sending audio by RAAT over Ethernet to a network renderer and then to your receiver by Optical or USB (as you prefer) would give better results than the most expensive soundcard/SPDIF solution that money can buy. You can buy an expensive network renderer (Aries) or an inexpensive one (SonicOrbiter SE), and will soon (after Roon 1.2) be able to DIY, as you prefer.

Edit: Of course, if your receiver (brand not stated) becomes RoonReady, then it will be it’s own network endpoint and you won’t need a renderer.

@Jaap74 A dedicated sound card can definitely be better than the internal audio chip set, especially if you use it as a DAC and send analogue out. To that end, I can recommend the Asus Xonar Essence STX, because I’ve used it to great success. It has EMI shielding, Burr Brown DAC and ASIO drivers.

@Jaap74: you might have jitter on the optical output, but if your DAC reclocks the audio on input you won’t have an issue.

great post!!!

Beyond ABX, the personally “method”, the resources, time&/money, have to be proportional to the desire to know the truth. …and then in the results, the truth could be not a desireble outcome!

I suggest all be selfish in this quest, but please share your results.

Absolutely agree @brian.

The noise level mentioned are indeed extremely low. However, whether or one can hear it is an opinion ultimately. Whether or not it truly exists as a problem is very personal.

I remember in the 70’s chasing down those numbers in THD, channel separation, etc. I bought equipment that chased numbers and I bought so I tended toward solid state. Friends of mine thought I was a shallow audiophile and rolled their tubes. They didn’t worry, and still didn’t worry a whole lot about noise (of course they do to some degree, but I’m make a point). They wanted what sound good to them.

There is nothing wrong with pushing the limits. For some it’s well worth it, for some a fools errand. The more the envelope is pushed the better the state of the art and for some of us, our audiophile pleasure.

Science changes, and so does art.

1 Like