What Phase Shedding Is — and Why I Turned It Off on the Dominus
Last post I showed you how much headroom the Dominus Extreme VRM has — 32 teamed power stages, a 570A+ sustained figure from ASUS’s own product page, and a CPU that uses about a third of it at stock. That was the “why there’s so much headroom” story.
This is the flip side of the same coin. Because most of the time, on most boards, most of those stages aren’t even switched on.
That’s phase shedding. And on my board, I turned it off — I run all 32 stages, all the time. Let me explain what that even means, why the feature exists, and why I made the call I did.
What phase shedding actually is
A multi-phase VRM is a bunch of little power converters wired in parallel, all feeding the same CPU rail. The Dominus has a lot of them. Your cheap B-series board has a few. Doesn’t matter — the principle is the same.
Here’s the thing people don’t realize: running every phase all the time is not free. Each phase has costs that don’t care how much current it’s actually delivering. It’s switching on and off hundreds of thousands of times a second, charging and dumping gate capacitance, burning a little in the driver and the core losses every single cycle. Call it overhead. The gate-drive part of that overhead is roughly fixed per active stage at a given switching frequency and drive voltage — it’s there whether the phase is pushing 30 amps or half an amp. (The MOSFET turn-on and turn-off losses scale with current too, so it’s not all fixed.)
Now picture your PC sitting at the desktop doing nothing. The CPU might be pulling a couple of amps total. Spread that whisper of current across a big array of phases and each one is delivering almost nothing — but still paying its gate-drive overhead. You’re running a dozen little converters flat-out just to keep the lights on, and the useful work each one does is a rounding error next to the overhead it burns.
That’s dumb, and VRM controllers know it’s dumb. So they shed phases: at light load they shut some phases off entirely and pile the (tiny) load onto the few that stay awake. Fewer phases burning fixed overhead = better efficiency exactly when the machine is idle, which — let’s be honest — is most of the time.
Then you launch something heavy, the load ramps, and the controller wakes the parked phases back up so the current spreads out again and no single phase cooks. Shed when idle, wake under load. That’s the whole dance.
Why the efficiency actually improves
Quick bit of the physics, because it’s the part that makes it click.
A power stage loses energy two ways. Conduction losses — the current fighting through the resistance of the parts — scale with the square of the current (I²R — the same relationship I walked through in the headroom post). And switching losses — the overhead of turning on and off. The gate-drive part is roughly fixed per active stage at a fixed switching frequency and drive voltage, but the MOSFET turn-on and turn-off losses scale with current too. So it’s a mix, and the fixed part is what hurts at idle.
At high current, conduction losses dominate, and spreading the load across many phases is a huge win — each phase carries less current, and since it’s squared, it adds up fast. In the simplified model, at the same total current, doubling the number of identical, equally loaded stages quarters the conduction loss in each stage and halves it across the whole array. That’s the case for lots of phases.
At low current, it flips. Conduction losses are tiny (small number, squared, is a very small number). Now the roughly fixed per-stage overhead is most of the bill. And the only way to shrink a fixed per-phase cost is to run fewer phases. So you shed.
That’s the elegant part: the exact same array wants many phases under load and few phases at idle. Phase shedding is just the controller picking the right answer for the current conditions instead of committing to one.
How the Dominus does it specifically
Here’s where it gets fun, because the Dominus doesn’t shed the way a normal board does.
Those 32 stages aren’t 32 independent phases. The board is run by a single ASP1405I controller — which is ASUS’s badge on Infineon’s IR35201, a digital multiphase controller that tops out at eight phases. So ASUS took its eight control phases and fanned each one out to four IR3555 power stages wired in direct parallel. Eight phases × four teamed stages = the 32 you see on the board. No doublers in the middle — I got into why ASUS skipped those last time — just four real stages taking the same PWM signal and sharing the load blind.
Which means the controller doesn’t shed stages. It sheds phases — and on this board, every phase is a team of four. So when the Dominus drops a phase at idle, it isn’t parking one power stage. It’s parking four at once. Shedding happens in chunks of four. Drop down to a couple of active phases at the desktop and you’ve got maybe 8 of your 32 stages actually switching, the other 24 sitting dark.
This isn’t a fancy-board feature — cheap boards do it too
Don’t let the 32-stage monster fool you into thinking phase shedding is some halo-tier trick. It’s the opposite. Plenty of boards shed phases, budget ones included. Many modern VRM controllers support it because idle efficiency is something manufacturers get judged on, and it costs them very little to enable.
Your little four- or six-phase B-series board might be quietly shutting some of its phases off while you read this. The idea is the same as what the Dominus does — there are just fewer phases to play with, so the shedding is coarser and matters less. If anything, a $3,000 halo board with 32 stages has more room to shed than a budget board does, not less.
So when someone tells you their fancy motherboard “has phase shedding” like it’s a selling point — it’s not much of one. A lot of the cheaper boards do it too.
So why did I turn it off?
On ASUS boards this lives in BIOS under CPU Power Phase Control. You get roughly three flavors:
- Standard — sheds based on load, stock behavior.
- Optimized — ASUS’s tuned profile for daily use where load bounces around.
- Extreme — never sheds. Every phase stays lit, all the time, full stop.
I run mine on Extreme. All 32 stages, awake, permanently. And I’ll be honest with you about why, including the part where it doesn’t really matter.
The reason you’d disable shedding is transient response and stability margin. When the controller has parked most of the array and a sudden load slams in, it has to wake those phases back up — and there’s a tiny lag before they’re all pulling their weight. For a stable overclock chasing every last bit of margin, some people don’t want the VRM ever having to “catch up.” Keep all phases enabled and there’s no phase-add step. That doesn’t make the response instant, though — inductor current still takes time to rise, and the controller (the IR35201, which has its own Adaptive Transient Algorithm and Dynamic Phase Control) has its own fast response. So I picked Extreme / full-phase mode, but how much it actually helps with transients on my board is still an open question.
Now the honest part: I haven’t measured the idle power difference on my board. On a normal board, shedding saves power because the fixed overhead is a real fraction of a small idle total. What that works out to with all 32 stages lit on the Dominus, I don’t know yet.
So until that’s measured: on a normal board, shedding’s efficiency win is real and well-documented. On the Dominus, I haven’t measured the idle power difference — but I’d rather keep the whole array hot and never think about wake-up lag again than chase savings I haven’t actually confirmed. Different board, different call.
There’s also just something I like about it. This board was built to run 32 teamed stages. So I run 32 teamed stages. All of them. All the time. Feels right.
The short version
Phase shedding parks VRM phases at idle so the fixed switching overhead doesn’t eat you alive when the CPU’s doing nothing, then wakes them under load so the current spreads and nothing overheats. Many modern VRM controllers support it, cheap boards included. The Dominus does it in chunks of four, because each of its eight control phases is a team of four stages.
And I turn it off — not because shedding is wrong, but because I’d rather run all 32 stages the way the board was clearly meant to, and I haven’t measured what that costs at idle.
ROG Dominus Extreme — 32× IR3555 stages, 8 phases of 4 teamed each, run off a single ASP1405I (IR35201) controller, phase shedding disabled, all 32 stages lit permanently against a W-3175X that uses roughly a third of the board’s rating at stock.