TDP isn't watts drawn: reading a PC parts list into real power
The CPU and GPU numbers you copy off a spec sheet are thermal ratings, not measured draw — the gap runs both ways. Plus what the calculator's fixed per-component figures assume, and the second safety margin it adds on top of yours.
Adding up a PC's power draw looks like clerical work: copy the CPU's TDP, copy the GPU's TDP, throw in a few watts for drives and fans, done. The trap is in the first two numbers. TDP is a thermal spec — it describes the heat a cooler must dissipate — and it is not the same as the watts the part actually pulls from the wall. Sometimes it undershoots real draw badly; occasionally it overshoots. If you don't know which way your specific parts lean, the tidy sum you built your PSU choice on can be off by a wide margin before you've even added headroom.
How much headroom to add on top of your total, why GPU transient spikes drive that headroom, and where PSU efficiency actually peaks are covered in how to size a PC power supply. This post is about the layer underneath: whether the numbers you're summing mean what you think, and what the calculator quietly does with them once you've entered them.
CPU: TDP is a floor, not a ceiling
Modern CPUs boost well past their rated TDP. On Intel, TDP describes the processor's sustained base-power state (PL1); under load it's allowed to draw considerably more for its boost state (PL2) — often for extended windows, and on many motherboards effectively indefinitely because the board vendor has quietly lifted the time limit. A chip labeled 105 W can pull 140 W or more while it's actually working. AMD tells a similar story through a different number: the socket's real power ceiling (PPT) sits around 1.35x the rated TDP, so a "105 W" Ryzen is provisioned to draw closer to 142 W at the socket.
The practical upshot is that for the CPU, its rated TDP is a starting point that leans low under sustained heavy load. That's fine as an input as long as you know it's a floor — the headroom you add afterward is partly there to absorb exactly this gap, and if you run a chip whose board unlocks its full boost, nudging the entered figure up toward its real sustained package power is more honest than trusting the label.
GPU: the label is closer, but the name keeps changing
Graphics cards are generally better behaved: the board power figure a card advertises is much closer to its real sustained draw than a CPU's TDP is. The catch is vocabulary. Vendors variously call it TDP, TGP (total graphics power), or TBP (total board power), and they don't all measure the same boundary — some figures cover just the GPU chip, others the whole card including memory and fans. The number that matters for a PSU is total board power, the whole-card figure. Grab a chip-only number by mistake and you understate the card, which is the single biggest draw in most builds.
And even an accurate board-power figure describes sustained draw, not the millisecond transient spikes a modern GPU throws — those can momentarily hit one and a half to two times the rated figure, which is the headroom question the sizing post covers, not something you fix by editing the input.
The small numbers are averages baked into the tool
The calculator fills in the rest of the system with fixed per-part figures: roughly 7 W per storage drive, 2 W per case fan, 3 W per RAM stick, plus a motherboard-and-everything-else baseline you set. These are reasonable under-load averages, but worth knowing as assumptions rather than facts. The 7 W-per-drive figure fits a spinning hard disk under load; an idle SATA SSD sits well under that, while an NVMe drive hammered with sequential writes can briefly spike above it. Fans at 2 W assumes ordinary case fans, not high-static-pressure servers screamers or a wall of RGB. None of these individually moves the PSU tier, but a NAS-style build with a dozen drives is a case where the drive figure alone becomes real wattage worth checking against the actual disks.
The margin you didn't ask for
Here's the part that surprises people who add up their own wattage by hand and expect the tool to match. After it applies your headroom percentage to the component total, the calculator adds a second cushion when it picks a PSU: it recommends the first standard size that sits at or above your padded draw times about 1.15. So the headroom you entered isn't the whole safety margin — there's roughly another 15% folded into the size selection itself.
That's deliberate, and it's the mechanism behind the "aim for around 85% load" advice. Take a 391 W component total with 20% headroom: that's 469 W of padded draw, times 1.15 is about 539 W, and the first standard size at or above that is 550 W — which leaves the 469 W draw sitting at roughly 85% of the 550 W unit. The load percentage the tool reports isn't a coincidence; it's the direct result of that hidden 1.15 factor steering you toward a unit whose efficiency-friendly operating band your real draw lands in. Knowing the margin is there stops you from double-counting it — you don't need to inflate your entered headroom to "be safe," because the size step already did.
Enter honest numbers, let the tool do the rest
The reliable approach: enter your GPU's total board power (not a chip-only figure), treat your CPU's TDP as a floor and nudge it up if the chip boosts hard on your board, and leave the small per-component defaults alone unless you're running an unusual number of drives. Feed those into the PC build wattage & PSU calculator and it returns your total system draw, a recommended PSU tier, the resulting load percentage, and running cost — with the extra size-selection margin already applied so the recommended unit lands in its efficient band. The math is only ever as good as the two big numbers you copy in, so spend your attention there.
Related articles
What it actually costs to charge an EV vs. fill up with gas
Why a "75 kWh battery" costs more than 75 kWh to charge, the real price gap between home and DC fast charging, and a worked road-trip cost comparison against a gas car.
How to size a PC power supply without over- or under-buying
Why headroom above your calculated draw matters for GPU transient spikes, where PSU efficiency actually peaks, and a worked component-by-component wattage example.
What your 3D printer actually costs in electricity (including standby)
Why idle/standby draw quietly adds up over a month, the power difference between FDM and resin printing plus UV cure stations, and worked monthly and annual cost examples.