An Applied Investigation Into Graphics Card Coil Whine

jones_supa writes We all are aware of various chirping and whining sounds that electronics can produce. Modern graphics cards often suffer from these kind of problems in form of coil whine. But how widespread is it really? Hardware Canucks put 50 new graphics cards side-by-side to compare them solely from the perspective of subjective acoustic disturbance. NVIDIA's reference platforms tended to be quite well behaved, just like their board partners' custom designs. The same can't be said about AMD since their reference R9 290X and R9 290 should be avoided if you're at all concerned about squealing or any other odd noise a GPU can make. However the custom Radeon-branded SKUs should usually be a safe choice. While the amount and intensity of coil whine largely seems to boil down to luck of the draw, at least most board partners are quite friendly regarding their return policies concerning it.

Comments

My nvidia card started whining after lightning

[Anonymous Coward]

Motherboard got fried by lightning, replaced said motherboard, video card now whines.

I've gotten used to it.

Re:My nvidia card started whining after lightning

[adolf]

I used to have a Sound Blaster Pro which had some lightning damage. Something on the board had turned microphonic, and you could shout at the card and hear it through the line output.

Fun stuff.

Re:My nvidia card started whining after lightning

[Mashiki]

Chances are it's the PSU causing it. That seems to be in at least my experience over the last 20 years, that if you're experiencing some type of coil whine it's related to the PSU--usually a weak rail, or a rail that's failing. This is especially true under heavy load, there are rare cases as mentioned in a lower post of other issues causing it but the majority of the time it's something you can fix on your own. Especially if it happened after a strike.

The Cause

[labnet]

I've designed lots of these little switch mode supplies. (SMPSs)
The noise comes from the inductors. Inductors are coils of wire around a ferrite. When the current changes through the wire, the wire physically expands and contacts from every other wire. This is the source of the noise. (SMPSs normally switch from 200kHz to 2MHz, so well outside our audio range)
There are a few things a designer can do.
1. Encapsulate the coil. This holds the wire tighter together and can minimise noise, but is only usually used in large inductors like those in invertors for UPSs or solar.
2. Eliminate subsonic oscillation with good multi-pole compensation. Switch mode power supplies have, have first second and third order responses which require filters to damp them. If you don't design these filters well, you can get subsonic oscillation which falls into the audio band. The power supply still regulates OK, but you can get that annoying whine.
3. Occasionally the noise can also come from a periodic load with that falls into an audio range. More capacitors on the output can help that.

Also, very very occasionally, it can come from ceramic capacitors that use a high k dielectric that are microphonic, but in my experience it is usually the capacitor acting as a microphone that upsets the circuit.

Re:The Cause

[Anonymous Coward]

Actually, it is slope compensation that you want to use to avoid subharmonic oscillation, and you can avoid that need entirely by designing your SMPS such that it never exceeds 50% duty cycle for the design load.

Re:The Cause

[Charliemopps]

Have you ever noticed if this effect gets better depending on the shape? I use transformers in audio amps I build, and often there's a marked difference if I use R-core over EI or Toroid. But this is signal noise and not physical noise. If the transformer is making physical noise, I'd never be able to hear it over my guitar solo :-p

Re: The Cause

[CODiNE]

Slashdot no longer the place you used to go?
While it's a bit more mainstream audience than back in the day... you still get great comments like this one from time to time.

So when you think Slashdot is dead, remember this post.

Re:The Cause

[labnet]

AC is correct. Most forms of subharmonic oscillation are caused by slope compensation issues, but pole filtering can also be an issue . For those who want the nitty gritty details, see this. http://www.ti.com/lit/ml/slup2...

Re: The Cause

[justthinkit]

Well said.

I would also like to thank Hardware Canucks for doing this test in the first place. Like most nerd/geek/freaks, I'm very sensitive to noise, and computer case noise is the worst because you will probably have it for the life of the box...that you use for 10 or 15 hours a day, every day.

So, thanks. And thanks.

BTW, it would be kind of awesome if the computer hardware testing sites incorporated sound tests into their general testing of stuff.

Re:The Cause

[PPH]

When the current changes through the wire, the wire physically expands and contacts from every other wire.

Not just the wires. But the core physically changes shape due to magnetostriction. The only was to reduce this is by careful selection of the inductor magnetic material and/or reducing the flux density in the core.

Re:The Cause

[plcurechax]

Naïve question maybe, but couldn't some sort of lacquer be applied on wires to prevent them from physically moving?

For small gauge magnet wire, it often is used on better, but it is not perfect.

Better components tend to cost more, which for commodity priced* products like video cards, saving a few cents can be considered worthwhile.

*) A lower bill of materials cost, can be used to past on some or all of the savings to the consumer, where for price sensitive consumers, the company with the cheapest product can end up selling potentially 30-400% more if you have the cheapest of a seemingly similar product (video card with chip Y9000), thereby increasing net profits even with lower profit margins per unit.

Re: The Cause

[plcurechax]

BTW, it would be kind of awesome if the computer hardware testing sites incorporated sound tests into their general testing of stuff.

You mean like this:
  Tom's Hardware: Sapphire's Vapor-X R9 290X 8GB - Temperature, Noise And Power.

Actually I continuously get frustrated by "enthusiast" computer sites reviews who seems to being entirely lacking in technical knowledge when it comes to anything beyond quoting the manufacturers press material. Half of them might as well have a companion site reviewing shoes and fashion tends given their display of technical ignorance.

Re: The Cause

[gweihir]

Indeed. There are people here that know their stuff.

Re:The Cause

[Anonymous Coward]

Full disclosure: I worked on this *exact* issue at NVIDIA for a brief period, although several generations ago. Usual disclosure applies - the below is my opinion, not theirs.

It doesn't surprise me that the NVIDIA reference cards do pretty well; they took the issue as seriously as some of their customers do. We made some fancy measurements to evaluate different methods of reducing the noise; indeed, some of the "common sense" things a designer can do are actually wrong. For example, #1 from the parent (fill the inductor package with something to keep the coil from moving) isn't necessarily a good idea. In some cases, that actually makes the vibration *worse*; rather than prevent the coil from moving, it helps transmit the motion of the coil to the PCB, which can then act like a sound board, making the tiny coil's vibration into something audible. (The sound board is the part of a musical instrument that is forced by the string to vibrate, making sound. It's what makes an acoustic guitar make noise when a string is plucked, while an electric guitar is relatively silent with no amp.)

That's not to say that encapsulation is a bad idea necessarily - just that this is a much more difficult problem to solve than you might imagine at first glance. And something that works well for a particular application/GPU/inductor/card combination might not work as well for a different combination.

A long time ago, I had a PowerBook G4 that had a buzzy whining that I later (while working on this problem) learned was from the inductors. I discovered that by using the CHUD Tools part of Apple's developer tools to disable the "Nap" option (this is the CPU Nap power-saving mode, not the much more recent "Power Nap" marketing-branded feature), the whining would go away, because the current in the inductors was much closer to constant.

Re:The Cause

[Twinbee]

What would typically be the source of the whine/buzz behind a convector heater (non-fan), and an LCD monitor, and an LED bulb?

Re:The Cause

[drinkypoo]

There are a few things a designer can do.
1. Encapsulate the coil.

Is this a thing I can do, say, with epoxy? Do I have to worry about heat?

Re:The Cause

[vandamme]

Or cover it with goop.

Re:The Cause

[tlhIngan]

What would typically be the source of the whine/buzz behind a convector heater (non-fan), and an LCD monitor, and an LED bulb?

Header - surprisingly, AC current on a regular wire heater. The coils of heating element do act like an inductor and cause it to vibrate some. Depending on the mount, that vibration can translate into actual vibration of the chassis causing the buzzes.

LCD monitor - the switch mode power supply is generally the cause of it. Practically everything with a SMPS is vulnerable to buzzes/whines because of the way they operate.

LED bulb - a lot of them use a SMPS, so ditto.

Anybody familiar with the manufacturing side?

[fuzzyfuzzyfungus]

I understand that high-frequency magnetics are at risk of physical oscillation(the detailed math is right over my head; but all it takes is one part of the part attracting or repelling another part of the part, at least under some input waveforms, and you'll potentially see movement, which easily enough turns to sound); but the seemingly obvious solution is just to pot the magnetics in an adequately thermally conductive epoxy or other encapsulant.

Does anybody know if that just adds too much cost, without performance benefit, and so gets cut during the BOM penny pinching? Do potting compounds have properties that degrade the performance or efficiency of common magnetics? Why is it that, if coil whine is an issue, they aren't just dipping the things in epoxy and calling it a day?

Re:Anybody familiar with the manufacturing side?

[dgatwood]

Because 99.999% of users don't care enough to complain. When you get enough whine that a sizable number of users scream bloody murder, something gets done. But for a more typical amount of supply whine, why spend the extra buck or two? And chances are, it never makes it to the BOM penny pinching stage, because unless the design is producing serious noise, corrective actions probably won't be taken in the first place.

Besides, assuming you use a standard-shape power supply, the users who really care about noise will buy ultra-low-noise power supplies to replace whatever $15 junk PSU you put in the machine from the factory anyway. :-)

Re:Anybody familiar with the manufacturing side?

[Blaskowicz]

The issue is the "PSU" on-board a big graphics card is not replaceable. It's power cirtcuitry that feeds on the order of 200 amps to the GPU and handles huge transients / power and voltage transitions.

Re:Anybody familiar with the manufacturing side?

[dgatwood]

Ah. I missed the fact that you were talking about inductors on a DC-DC converter on the graphics card itself. In that case, yeah, that's just shoddy. On the other hand, it is easy to fix:

1. Buy epoxy.
2. Remove the card.
3. ...

:-)

Re:Anybody familiar with the manufacturing side?

[slew]

I understand that high-frequency magnetics are at risk of physical oscillation(the detailed math is right over my head; but all it takes is one part of the part attracting or repelling another part of the part, at least under some input waveforms, and you'll potentially see movement, which easily enough turns to sound); but the seemingly obvious solution is just to pot the magnetics in an adequately thermally conductive epoxy or other encapsulant.

Does anybody know if that just adds too much cost, without performance benefit, and so gets cut during the BOM penny pinching? Do potting compounds have properties that degrade the performance or efficiency of common magnetics? Why is it that, if coil whine is an issue, they aren't just dipping the things in epoxy and calling it a day?

Unfortunately mechanical damping of the inductor vibration isn't as effective as simply reducing the amplitude of driving frequency in the audio bands. Remember this is a sub-harmonic that is being excited by a non-linear coupling to the audio frequency. Basically the energy in a higher frequency is being converted into a lower audible mechanical frequency.

Theoretically, simply changing the mass of the physical oscillation (e.g. cementing it to something heavier) only slightly modifies the frequency of the oscillation (potentially creating more audible noise) and it still doesn't change the energy much. Viscous damping of the mechanical frequency might help a little bit more. Unfortunately, in practice, surrounding things like solder joints in potting compounds is risky as they have a different thermal expansion coefficients and it can cause additional mechanical stress (resulting in reduced mechanical reliability).

In the end, mechanical means are still not going to be as effective as changing the circuit to reduce the amount of switching energy frequencies which are coupled to the audio frequency bands. Probably even from a total system cost point of view...

Re:Anybody familiar with the manufacturing side?

[unitron]

The issue is the "PSU" on-board a big graphics card is not replaceable. It's power cirtcuitry that feeds on the order of 200 amps to the GPU and handles huge transients / power and voltage transitions.

Perhaps you mean 200 Watts?

Re: Anybody familiar with the manufacturing side?

[GigaplexNZ]

At 1 volt, 200W translates to 200A. The voltage that the chip runs at is normally not much more than 1V. Some (maybe most?) of the coils will be closer to 12V but at some point there will be really high currents.

Re:Anybody familiar with the manufacturing side?

[fuzzyfuzzyfungus]

Modern core voltages. In the suitably...enthusiastic... segments, a GPU that pulls 180-200 watts isn't hard to find (even without any overclocking, just off the shelf values) and the core voltage is around 1.1v, so there will be a pretty significant current involved. That's why the DC-DC converter tends to be so close to the chip it feeds, resistive losses would be brutal.

Re:Anybody familiar with the manufacturing side?

[jones_supa]

Because 99.999% of users don't care enough to complain.

That would mean 1 / 100,000 users complain. I think more users care. 99.99% or even 99.9% might be a closer value to the amount of users that don't care.

Of course I'm just nitpicking and past your point, but there's a surprisingly big difference into how many 9s you slap there. :)

Re:Anybody familiar with the manufacturing side?

[fuzzyfuzzyfungus]

I'm deeply underqualified to tell you how DC-DC converters do work; but only the simplest and nastiest ('linear regulators') step down voltages resistively. Those ones are nice and simple, so they do still show up in low power applications; but their inefficiency rules them out in cases like this. I don't know exactly how it is achieved; but the more sophisticated designs are capable of 90% or greater efficiency, which keeps heat related issues at bay.

As for quoted amperages, high end graphics cards get most of their power from the 12v rail (via one or two 6 or 8 pin auxiliary power connectors, plus what the PCIe slot provides). 12v is the highest voltage easily available inside a standard PC, so is the best choice for feeding a high powered component without too much loss in the cabling between the PSU and the card. However, the only things on a graphics card that actually uses 12v are the fans. The actual GPU chip, the memory, and all the various support components use substantially lower voltages( core voltage for recent GPUs is ~1.1v, GDDR5 is ~1.5v).

Because it would be brutally impractical to transfer substantial power at such low voltages, the conversion from 12v happens on the card, as close to the actual GPU chip as possible. The amperage for the card as a whole is expressed without paying attention to the amperage after this step-down step, since only the 12v draw is externally visible; but with such low vCores, the GPU chip, rather than the card, necessarily draws almost an amp for every watt it dissipates. 200A would definitely be on the high side; but not outrageously so.

Re: Anybody familiar with the manufacturing side?

[Megol]

None. But there are many conductors instead.

Re: Anybody familiar with the manufacturing side?

[Blaskowicz]

200 amps was on the provocative side. but about par for a Radeon R9 290X in the "uber' profile or overclocked.
100 amps is routine on a CPU : 125 watt and 130 or 140 watt models or even more for a few ones. It has hundred pins, around a thousand to connect it to the motherboards, most of those are ground and power.

Re:Anybody familiar with the manufacturing side?

[dgatwood]

I'm deeply underqualified to tell you how DC-DC converters do work...

Me too, but I'll try anyway. After all, this is Slashdot. :-D

If I understand correctly, a buck (downstep) converter starts with an oscillator that drives a transistor. The transistor turns the power source on and off very rapidly. An inductor between the high source voltage (after the switch) and the low-voltage output effectively turns the resulting current cycling into a voltage drop, and a capacitor smooths the resulting power supply back into DC. A diode bypasses the source and switch, ensuring that current continues to flow when the switch is off.

The duty cycle of the oscillator determines the resulting output voltage, so if you're starting at 12V and need 3V, you would use a 25% duty cycle (an oscillator whose output is on 25% of the time, and off 75% of the time). The frequency of the oscillator must be well above the human hearing range for obvious acoustic reasons.

Re:Anybody familiar with the manufacturing side?

[dgatwood]

I guess it depends on how bad the noise is. I could believe one in 10K or one in 100K. I'd have a hard time believing that a manufacturer would ship something with enough noise to bother one in 1K people. Usually I'd expect the noise to be in a frequency range where most adults either can't hear it or can barely hear it.

Re:Anybody familiar with the manufacturing side?

[stoatwblr]

"but the seemingly obvious solution is just to pot the magnetics in an adequately thermally conductive epoxy or other encapsulant."

As another poster noted, this can make things worse, especially if the epoxy is rigid.

Some of the best methods involve potting _part_ of the coil (wax drops) or using rubber o-rings to absorb the motion/sound. The issue then becomes that those parts have to pass thermal energy in order to avoid melting.

Acoustic noise control in switchmode circuitry is an engineering discipline all of its own, however I'm willing to bet than in the next 5-6 years it will be as normal to see acoustic issues processed as part of board production engineering as RFI issues are now.

Re:Anybody familiar with the manufacturing side?

[stoatwblr]

"I guess it depends on how bad the noise is"

Noise is relative.

We're putting systems on desktops with near-silent PSU, chassis and CPU fans along with SSDs for booting, but can't quite justify the spend required to put 4Tb (local scratch space for scientific computing) on SSD (you can only do so much inside a $2k/system budget.)

As a result we've had a couple of people complain about "electrical arcing noise" from their computers, which turned out to be headseek noise when they're grinding on large datasets. In previous generations of equipment that noise was drowned out by other sources (including the fans on older equipment in the same room, or even the AC blowers), but because we've managed to eliminate case drumming the noise manifests as a faint ticking.

Re:Mac Pro 2013?

[fuzzyfuzzyfungus]

No idea if it's the GPU; but high-frequency magnetics are all potential culprits (as are low frequencies, though 50/60Hz is usually 'hum' rather than 'whine'), and a modern laptop is just stuffed with DC/DC converters keeping the various ICs fed, so if it isn't the GPU's support system, it's another fairly similar one.

To me this is good news

[fustakrakich]

I always though the noise from coming from a cap that was ready to explode.

Re:To me this is good news

[Zordak]

My anecdotal experience is that sometimes it's a ceramic diode getting hammered by out-of-spec back-voltage and ready to explode, and sometimes it does explode spraying ceramic shards all over the electronics lab. Root causes may include a dodgy transformer (pulled out of an old Hammond organ) with a highly questionable output waveform because you're a broke undergrad and it was cheaper than buying a new one.

Re:Mac Pro 2013?

[Sable Drakon]

The new 'trashcan' design ones are however built with laptop-modeled components. Designed to conserve as much space within it's overly minimal enclosure as physically possible.

Re:Mac Pro 2013?

[Lunix Nutcase]

It's a desktop.

Re:Mac Pro 2013?

[fuzzyfuzzyfungus]

My mistake, misread as 'macbook' pro. Still likely to be stuffed full of DC/DC converters. It's been ages since the PSU actually directly powered much of the more demanding silicon.

Re:Mac Pro 2013?

[Sable Drakon]

The PCBs. Take a moment to think about it. That is naturally where all the inductors are mounted.

Like bearing noise in a full-throttle jet engine

[idontgno]

Are people actually hearing inductor acoustic oscillation over FAN NOISE? If you can hear yourself think over your graphics card, YOU'RE NOT A REAL GAMER!.

Re:Mac Pro 2013?

[Noah Haders]

Silly rabbit. Even Sarah Palin knows that the leader of Canada is a Premier, not an emperor.

Re:Mac Pro 2013?

[Blaskowicz]

No idea why you were modded troll, I wanted to somewhat disagree with you - the GPU aren't really "desktop", they are on custom boards a bit bigger than on laptops but smaller than on desktops. Funnily you will have a harder time replacing them, compared to an MXM laptop GPU board which at least follows some standard. You won't be able to get one from somebody else than Apple, and that would be on their terms such as handing your Mac over to an Apple shop and have them upgrade it for you (if they even agree to it).

It's not a laptop but there is a philosophy much like a laptop. Some other desktop hardware like "nettops", Intel NUC aren't much different.

Re:Mac Pro 2013?

[Lunix Nutcase]

the GPU aren't really "desktop", they are on custom boards a bit bigger than on laptops but smaller than on desktops.

Yes they are. FirePro GPUs are workstation graphics cards.

Re:Mac Pro 2013?

[Lunix Nutcase]

The point is that the laptop uses laptop parts whereas the MacPro uses server and workstation parts.

Re:Mac Pro 2013?

[Lunix Nutcase]

What good are server components when they aren't in a server chassis?

They are good for having many cores for the multi-threaded applications the MacPro is built for. I'm pretty sure server components perform exactly the same regardless of the chassis.

The Mac Pro is marketed as a "workstation", but it's basically a mid-range desktop.

LOL. Oh please do link me some of those "mid-range desktops" using Xeons, ECC memory and FirePro GPUs. I won't hold my breath, though.

Re:Mac Pro 2013?

[Lunix Nutcase]

So?

So to what? The entire point is that the MacPro is not a laptop because it doesn't use a single piece of laptop components. That is the point in and of itself.

Re:Mac Pro 2013?

[Lunix Nutcase]

That should be amended to dual FirePro GPUs.

Return Policies = enjoy that refurbished card!

[n3r0.m4dski11z]

at least most board partners are quite friendly regarding their return policies concerning it.

ug.. Coil whine happened to me a few years ago on a brand new card so i RMA'ed the card. At the time it took some convincing to issue the rma too iirc. They shipped me some refurb card that never worked right. And the next one didn't work right (unstable or just plain DOA cant recall). By now Im up to 40 bucks just in shipping these crap cards back to the company. Never did get a working card out of it. The next card they sent me was awesome (2gb video ram at the time), but it was never stable. Ended up just buying an AMD card after that.

Should have just stuck with the damn whine, but it was driving my wife crazy (i can wear ear phones). And it was a brand new card under warranty, so I wasnt going to go desolder coils right off the bat!

damn right!

[slashmydots]

I cannot believe how loud our GTX660's from EVGA are. Wherever you're sitting in whatever country you're in right now, you can probably hear it. Some of our 550ti's and GTX650's are the same. The 650's aren't even high wattage! It's just completely unacceptable. I can hear up to about 40KHz so I've had my fair share of loud electronics that only I can hear. At least this frequency is within normal human range so more people can complain about it.

Re:damn right!

[rwa2]

I have an EVGA GTX 560Ti . My wife finally made me replace my fans.

They used really cheap fan mounts that vibrate like heck whenever they spin up.

Just replaced the stock fans with this rig today: http://www.amazon.com/Titan-Ad...

Much quieter now.

Re:damn right!

[weilawei]

You're talking about more than double the generally accepted maximum frequency for human hearing. You're solidly into the ultrasound range. Other animals can hear at those frequencies (such as cats), but I don't see the evidence of any human ever having that hearing range.

TL;DR: Citation needed.

Re:damn right!

[Megol]

I think the AC is talking about beat frequencies...

Re:damn right!

[weilawei]

You need two interfering frequencies for a beat frequency and what you're hearing is the difference between the two (their physical interference)--not the 42 kHz tone itself. The audible frequency is still going to need to be within that 20 Hz - 20 kHz (roughly) window in order for a human to hear it.

TL;DR: Citation still needed.

Security?

[manu0601]

How much internal state information is leaked in the noises?

Re:Security?

[jones_supa]

You can probably infer whether the chip is working hard or not, and possibly hear it stepping clock frequencies. Maybe if the frame rate is locked to v-sync, you could also hear a beat frequency synchronous to fps. I don't think you can find out anything more specific than those kind of things. There's a lot of power filtering going on and the chip architecture is extremely complex.

Re:Like bearing noise in a full-throttle jet engin

[ProzacPatient]

Without doing something really exotic you'll at least one to three fans to cool your typical radiator depending on it's size, and double that if you're doing a push-pull configuration.

Re:Mac Pro 2013?

[ddtmm]

I've looked into the Macs and it's similar in that it's coil whine but from the coils in the power supply and not the graphics. Most switching power supplies work at higher frequencies (that is, much higher than 50/60 hz) and produce that sound when the right amount of current passes through them. Many devices using switching supplies (which is most electronics devices) make that sound. All the NEC lcd monitors in our office have a faint whine to them as well, and also power supply related.

Re:Mac Pro 2013?

[Blaskowicz]

A difference is Mac Pro uses a Tahiti GPU (on D500 and D700) which you won't find in any laptop.
We would have to go back to the "sewing machine" or "oscilloscope" form factors of the 80s.

Re:Mac Pro 2013?

[Blaskowicz]

FirePro is a driver. You even have an APU variant http://www.sapphirepgs.com/pro...

In fact if we want to go all pedantic and nerdly, the Fire Pro variants on these Mac are said to be not such much Fire Pro as the drivers are specific to OS X and less featured. CAD, engineering etc. mostly happens on Windows.

Re:Mac Pro 2013?

[EmperorOfCanada]

That is what apple thought before the replaced the power supply and nothing changed. Plus it seems to warble at bit when new colourful things appear on the screen. I can't tell if the data coming in might be the source or the GPU making it happen.