EEVblog Electronics Community Forum
Electronics => Beginners => Topic started by: paulca on September 27, 2020, 03:56:46 pm
-
I spend some time today trying to understand what the PC builder aficionados are talking about and came to the conclusion that most of them... well don't know either.
I raised an eye brow when they refer to DC phases. I might have pointed out that DC has no phase and I might have got "You don't know how VRMs work man, go watch this YouTube video" replies. Out of interest, that YouTube video referred to different DC phases and a buck convertor being "a choke, 2 transistors and a capacitor". Right.
All I've worked out so far is...
"VRM" is a marketing bullshit term.
The "VRM" is just a set of parallel buck converters with variable outputs.
The "phases" thing seems to be something to do with them staggering the PWM signal across the parallel bucks. Nothing actually to do with the output.
At that point trying to work out why this is done is completely lost on me because their diagrams show the outputs of all the buck converters joining together. Then show a graph of each "phase" having a different voltage... while connected together. I heard Kirchhoff spinning.
I would gather they are paralleled due to the extremely high amperage being drawn by a modern CPU. Over 100Amps.... on PCB traces. I would expect however that the CPU has multiple VCore input pins and they are parallel all the way into the CPU itself, not joined together.
Can anyone shed light on this? Particularly why do they stagger the PWM? Is it just damping the noise that pulsing all the bucks at the same time would create?
-
Those high current buck converters used typical as Vcore sources for the CPU/GPU usually are multiple converters operating in parallel to
a) split the current load across them and b) to get less output ripple, if they are operated in interleave mode.
All N of buck converters in parallel operate at the same switching frequency, but each of them has its switching frequency reference phase shifted by 2Pi/N.
So imagine for example two buck converters operating at 25% duty cycle. If you parallel them and make each other switch 180° degree apart, the sum of output current will look like a 50% duty cycle at twice frequency, because you get a pulse from each 2 times per period.
If you repeat the same with 3 buck converters 120° apart, you get triple the output frequency - that makes easier to filter to obtain a smooth ripple free output.
Hence why they are called multi-phase or interleaved converters.
-
Some reading material:
https://www.ti.com/lit//slyt449 (https://www.ti.com/lit//slyt449)
https://www.ti.com/lit/slva882 (https://www.ti.com/lit/slva882)
Oh, and VRM is not a marketing term, it's just a little obsolete. Certain classes of PC used to have discrete regulator modules. It's a bit less of a mouthful than "multiphase core voltage regulator", though..
-
Are they using the low side mosfet to dump the inductor pulse when the high side cuts out? Is that more efficient than just using a diode?
-
Of course it is more efficient. Multiply your 100 amps by the Vfwd of an schottky diode (~ about a 0.8V or so at such currents), or calculate the same with a couple low Rds(on) mosfets in parallel. Of course the mosfet will win. Not by that much, but it is worth using the mosfet, instead of plain diodes, sure.
On some older moboards, you can even find diodes instead of the synchronous rectifier. So, either got used. As always: "it depends".
-
Yeah, remember that modern CPU cores typically run on supplies of less than 1.5V DC, so they are very low voltage, high current supplies and things like a diode drop in many places in the circuit adds up losses fast.
-
Those high current buck converters used typical as Vcore sources for the CPU/GPU usually are multiple converters operating in parallel to
a) split the current load across them and b) to get less output ripple, if they are operated in interleave mode.
Also in the case of CPU/GPU regulators to provide the needed load transient response. A cpu can come out of a low power sleep state and ramp to 100+ amps in a very short period of time. A single phase regulator has to wait until the next clock cycle to respond. A multiphase converter only has to wait for the next phase to switch.
-
I quite doubt the regulator's bandwidth can be anywhere close to the switching frequency, but sure, you get N-times the output frequency and short transients must be supplied by the cap bank.