specifically the following issues:
> are they really just simple Buck converters? (what are the actual circuits?)
> how do multi "phase" VRM setups actually work
> where exactly does coil whine come from?
> why is it not feasible to shift it into the ultrasonic range, etc?
I don't understand why they can still be such an annoyance in the Year of Our Lord 2015+1 and would like to know if it's really an intractable issue or just chinks being cheap.
It has multiple buck converters in parallel, which reduces the load on each and allows faster responses to change in load.
They're buck converters and multiphase means there's several of them in parallel. The more of them there are the more stable the output voltage is.
Coil whine is caused by current passing through the inductors, there's always some of it but it's usually inaudible. Rarely it can be heard and it becomes annoying as fuck. You can usually return the hardware in those situations
>>52384512
what are the general operating frequencies of the BCs, and why can't they just be cranked up to the 30+ kHz region?
is there some low frequency PWM bullshit going on?
>>52384587
is it load transience in the inductors that causes the problem?
I don't get why steady current would make an inductor oscillate.
it is the supply switching frequency that is the underlying issue?
Oh my god!
Finally!
A topic with level!
>>52384588
My motherboard I think is 4x100KHz, so 400KHz total.
The frequency isn't necessarily the issue, it's the load on each one. If a CPU is using 140w, you wouldn't want a single converter to have to deal with that.
>>52384630
Ha
A current one as well
>>52384626
My understanding is:
The current induces a magnetic field around the wire which then changes the properties of the wire slightly causing it to move or vibrate back and forth. The sound from the vibration is what we hear as "coil whine."
>>52384639
does anybody ever have issues with whine on motherboards?
I only ever hear about it in the context of higher-end GPUs.
>>52384588
the VRMs are definitely PWM-operated, but the switching frequencies are generally in the 100 kHz to 2 MHz range already.
I too would like to know what the deal with the whine really is, and suspect it does have to do with changing loads during and between frames.
>>52384833
>and suspect it does have to do with changing loads during and between frames.
That's it.
>>52384850
so the solution is for GPUs to do intra-frame pacing (gate off shader ALUs to make frames take longer if CPU bound, etc.), or pic related on GPU side of VRMs?
>>52384470
found a decent diagram for multi-phase VRMs...
>>52384850
Smoothing the load is probably the best option.
Some portion of the noise may come from periodic magnetostriction of the ferrite choke cores, which would probably be hard to get rid of with multi-phase designs with 1 choke per MOSFET pair.
>>52385396
Fuck meant to reply to
>>52384930
>>52384930
For a ~250W card with frame rates going down to 30 fps, you'd need 7-8 Farads of capacitance for a 1-1.2V power draw to completely eliminate the possibility of coil whine.
Would not be as bad as pic, but it would be a several pound cap with no prayer of taking less space than around an entire PCIe slot.
Hopefully AMD and Nvidia get frame pacing working intelligently to smooth their power draws.
>>52385126
Dumb question, but why no RC pairs between MOSFETs and chokes?
If the transistors are PWM driven, how much current can the inductors maintain at low duty cycles sucking off a dry hose?
>>52387186
You talking about RC snubbers?
>>52384470
Coil whine happens because the switching happens in an audible switching range over the inductors. Yes it's avoidable.
>>52387254
no, nothing actively switched.
more like pic related.
>>52387314
>>52384850
is it really the PWM switching at low frequencies or just the fact that the GPU draw fluctuates in the audible frequency range of like 30-200Hz?
and how do you fix it at the board level without adding monster caps?
>>52384588
>>52384639
>>52384833
>>52387710
When people say PWM at like 100kHz do they mean that the pulses are 10us apart or only that the width of each pulse is granular to the 10us level (and overall per-VRM switching frequencies being closer to the audible range)?
>>52387710
Think about what sound is and what happens with metal when it heats up.
If you do it fast enough it literally turns in some kind of speaker.
>>52388059
It's sawtooth pulses usually.
>>52388087
sawtooth pulses coming from where going where?
driver -> MOSFETs?
in any case, what exactly does the driver circuit look like?
>>52388177
>>in any case, what exactly does the driver circuit look like?
How the fuck should I know I don't design motherboards. I just know electronics.
Why do you care? you will not ever need this and if you do you'll already know.
>>52388069
thermal expansion is pretty subtle and fairly slow with any amount of e.g. ferrite mass in the choke.
I'd be more inclined to believe that coil whine was from magnetostriction of the core or probably from coil vibration under changing currents.
>>52388230
We're talking high frequency with low amplitude here. It doesn't take much to generate that and apparently some coils do it more then others. I bet if they were encased in epoxy or something they wouldn't do that.
>>52388228
> somebody asks a technical question on /g/ for first time in a blue moon
> "why would you ever need to know"
get back to one of the other 100 trash threads, faggot
>>52388289
That's a question that literally requires me to have seen the specific motherboard components you retard
Nobody is going to answer that with any kind of accuracy. Go take a motherboard apart if you and with pics and ask again.
Guyse, this article is what your are looking for:
http://www.geeks3d.com/20100504/tutorial-graphics-cards-voltage-regulator-modules-vrm-explained/
>>52388327
if you were
>>52388087
>It's sawtooth pulses usually.
and made a general statement and weren't able to make a general follow-up:
>>52388228
you shouldn't have said anything.
>>52388177
the controllers for integrated VRMs have gotten pretty complex in the last 15-20 years, and it really does vary quite a bit.
check out pic related for a typical design.
check out this for a ~10 year old but still valid discussion of the issues and evolution of the industry:
> http://scholar.lib.vt.edu/theses/available/etd-11142005-000332/unrestricted/XinZhang_PhD_Dissertation_ver3.pdf
>>52388948
So the VRM controller integrates PWM input and pushes a sawtooth to the MOSFETs?
>>52384470
>> where exactly does coil whine come from?
Magnetostriction. Some magnet materials (most, actually) deform slightly in a changing magnetic field. It's this movement that is responsible for the whine.
> why is it not feasible to shift it into the ultrasonic range, etc?
You want lots of ulreasonic noise around?
We used to reduce the effects of magnetostriction by "potting" coils in wax or epoxy. Eventually, the potting fails due to mechanical fatigue.
>>52390015
why haven't GPU vendors found inductors with atypically low magnetostrictive coefficients and branded them as "silent coils" or whatever the same way they tout Jap-made solid caps?
>>52384470
nonlinearities cause ultrasonic stuff to create harmonics in the lower frequencies. search up intermodulation distortion.
the only way to fix it is to use better components and dampen the coils buzzing
>>52390785
>>52390015
it seems like most of the noise is caused by load variance within a given frame, which would probably be hard to consistently chop up into 30+kHz activity bursts anyway.
you would think that at some point the GPU shader schedulers could do a better job of pacing frames, since minimized clocks and dropped vcores would help power consumption anyway.
>>52390900
? are we talking about coil whine?
>>52390984
yeah, but coil whine frequency is pretty much always directly proportionate to fps, so beyond getting non-shit inductors, it would stand to reason that flattening intra-frame power draw would be the obvious thing to do.
>>52391056
oh yes, if current spikes happen at an audible frequency it will be heard. a way to fix it would be to add some bulk capacitance on the output side.
>>52391161
as others have stated above, buffering 200+W at ~1V for 1/30s is infeasible in the form factor.
the output capacitors are really only to help smooth sub-ms scape ripple from the MOSFET arrays to tolerable levels.
>>52391639
yeah, it will decouple some of the current spike from reaching the inductor and making it whine. should work in principle, but chinks are too retarded to put 10 cent caps in.
> mechanical energy causes sound
> please shift this into ultrasound anyway
ok kid : /
>>52391806
slicing up the power draw pattern of GPU frame rendering into the ultrasonic range is doable in principle, avoiding IM may not be though.
