What PSU Does an RTX 5090 Need? (2026)

1000W is the floor. 1200W is the buy. NVIDIA lists 1000W of required system power for the RTX 5090 and recommends a PCIe CEM 5.1 compliant unit — and unusually, they tell you what that number assumes: a system built around a Ryzen 9 9950X. Run our own itemized arithmetic on exactly that pairing and it lands nearer 1200W. This is the one card in the range where our math comes out above the vendor minimum rather than below it, and that is worth understanding before you spend.

The card’s actual power specification

Two things here appear nowhere else in the 50-series lineup. NVIDIA names the processor behind its figure — everywhere else the assumption is invisible — and it recommends a specific PSU standard rather than just a wattage. Both matter more than the headline number.

Why our number is higher than NVIDIA’s

On the 5070 and 5080 our arithmetic came out below NVIDIA’s minimum, and the honest advice was to buy the vendor number anyway. The 5090 inverts that.

Summing component draw the way our PSU calculator does — 45W for an ATX board, 5W per RAM stick, 8W for a Gen4 M.2, 2W per fan, 12W for an AIO, plus a 10W miscellaneous allowance, so 93W for a reference build of two sticks, one drive, four fans and an AIO:

Sizes are peak draw plus a 20% margin, rounded up to a wattage actually sold — the smallest unit that works, not the most comfortable.

Read the third row. On the exact processor NVIDIA cites, our rule says 1200W where they say 1000W. A 1000W unit is not wrong there — 868W peak on 1000W is 87% load, which will run — but it is above the 80% line where our own efficiency table starts warning, and it leaves nothing for an overclock or a future card. That is why the recommendation here is a tier above the vendor floor rather than equal to it.

The connector is not optional in the same way

The 5070 takes two 8-pins, the 5080 three, the 5090 four — or a single 600W PCIe Gen 5 cable. Stacking four 8-pin adapters behind a card is where cable management stops being cosmetic, and it is the practical reason most 5090 owners end up on a native 12V-2x6 cable rather than the bundled adapter.

NVIDIA’s recommendation of a PCIe CEM 5.1 compliant supply is the part to take seriously. It is not a wattage rating; it is the connector and excursion standard the card was validated against. A 1200W unit that predates it is a worse pairing than a 1000W unit that meets it.

Seat the connector fully until it clicks, and do not bend it sharply at the plug. The current design shortens the sense pins precisely so a partially seated plug fails to deliver full power instead of making high-resistance contact.

Transient behaviour at 575W

Intel’s ATX 3.0 design guide requires a compliant unit to survive excursions to 200% of its rated wattage for 100 microseconds, with lower brackets for longer events — but only where the unit is above 450W and ships a native 12VHPWR / 12V-2x6 connector. Any supply you would sensibly pair with a 5090 clears the wattage half of that test easily; the connector half is the one to check.

Where this card differs from the rest of the range is margin. On a 650W build a spike has proportionally more headroom above it; at 868W peak on a 1000W unit, there is far less room before protection trips. That is the second argument for 1200W, independent of the efficiency one.

What undervolting changes

A 575W card is where the absolute savings are largest — our results database records around 100W off flagship-class cards, roughly double what a mid-range card gives back. On the 200W-CPU row that moves peak draw from 868W to about 768W, which brings a 1000W unit back to a comfortable 77% rather than 87%.

That is a real argument for tuning the card, but not for buying a smaller supply in advance. Silicon varies, the saving is not guaranteed, and a power supply has to cover the case where your card undervolts badly. Buy for the stock figure and treat the saving as headroom you gained. The complete GPU undervolting guide covers the method, and the undervolt results database lists what specific cards achieved.

If you are stepping up from a 5080

The jump is 360W to 575W — 215W more card, and a move from three 8-pins to four. An 850W supply that suited a 5080 does not carry a 5090, so budget the power supply alongside the card. What PSU an RTX 5080 needs covers that tier, and the GPU hierarchy shows what the extra power actually buys in frames.

Sources

• NVIDIA GeForce RTX 5090 official specifications — total graphics power, required system power and its stated Ryzen 9 9950X basis, the supplementary power connectors, and the PCIe CEM 5.1 recommendation — RTX 5090, verified live 2026-07-25.
• Intel ATX 3.0 desktop power supply design guide — the excursion brackets and the wattage/connector conditions that qualify a unit (Intel ATX 3.0 PSU power excursion).
• The ~660W figure for a factory-overclocked 5090 is this site’s own sizing estimate from the GPU list in the PSU calculator — a conservative peak for sizing, not a measured draw.
• Component draw coefficients and sizing rules are this site’s own published figures, listed in full in the PSU calculator methodology.