575 Watts for One Graphics Card: Has NVIDIA Gone Too Far?

Not on efficiency, but on absolute power, it has pushed the limit for a consumer card. The RTX 5090 is rated at 575 W, NVIDIA asks for a 1,000 W system power supply, and reviewers measured gaming averages of about 460–550 W with sub-millisecond spikes near 900 W. Yet its performance per watt is roughly the same as the RTX 4090's: it is faster mainly because it uses about 25–30% more power. In Singapore that adds a few hundred dollars a year in electricity for a heavy gamer, much more for an always-on AI box, and a lot of heat for a small room. With the right power supply, cabling and airflow, it's manageable. Without them, it's the most demanding card you can put in a PC.

How much power does it really draw?

Source: as listed. Board power only; the CPU and the rest of the system add more. Idle figures vary with monitor setup.

Two things stand out. The card rarely sits at 575 W in games, and a frame cap cuts its draw dramatically: 86 W at 60 fps in 1080p. But it idles higher than the 4090, and its short spikes are large enough to trip older power supplies' protection.

What does that cost in Singapore?

Source: SP Group regulated tariff for 1 Oct–31 Dec 2026, 28.59¢/kWh before GST and 31.16¢ with GST. Calculations are ours and use full rated power, so real gaming costs are lower. The tariff resets every quarter, and retail electricity plans differ.

For a gamer, the extra over a 4090 is a few tens of dollars a year. For a card running inference around the clock, electricity becomes a real line item, and a power limit (below) is worth considering.

Where does all that heat go?

Every watt becomes heat in your room: 575 W is about 1,962 BTU per hour, and an 800 W system about 2,730. For scale, a 9,000 BTU/h bedroom air-conditioner removes about 2.6 kW, so a 5090 gaming PC takes a noticeable share of a small room's cooling (our arithmetic).

Inside the case, the Founders Edition's double flow-through cooler blows hot air up toward the CPU. ComputerBase found a Ryzen 7 9800X3D reached 95 °C within 20 minutes beside a 5090 Founders Edition at low fan speeds, where it held about 73 °C next to a 4090 Founders Edition. Faster intake fans fixed it. Partner cards and liquid-cooled versions move heat differently; we compare them in RTX 5090 air vs liquid cooling and look at case fit in why your RTX 5090 doesn't fit.

What power supply do you need?

• NVIDIA's minimum: 1,000 W for a system built around a Ryzen 9 9950X, and it recommends a PCIe CEM 5.1-compliant unit.
• Headroom: Igor's Lab recommends a 1,000 W ATX 3.1 unit, ideally 80 PLUS Platinum or Titanium. ATX 3.1 supplies are designed to tolerate 200% excursions for up to 1 ms. Seasonic, a PSU maker, suggests 1,200 W for comfort and 1,300–1,600 W for sustained AI or rendering.
• Cabling: use the PSU's native 12V-2x6 cable or all four 8-pin inputs on the bundled adapter, each from a separate cable. ComputerBase found three of four inputs limit the card to 450 W, about 5% slower, and two inputs wouldn't boot.

Our 1000 W power supply myth post explains why a bigger PSU doesn't make a PC faster, and the power connector controversy covers safe connection.

NVIDIA didn't make the RTX 5090 less efficient than the 4090. It made it bigger and fed it more power, and the cost of that lands on your PSU, your room and your electricity bill.

Has NVIDIA gone too far?

The efficiency numbers say it hasn't gone backwards. KitGuru measured 4K performance per watt about 5% below the 4090, Gamers Nexus found it "about the same", and 3DCenter's meta-review rated 4K efficiency roughly level. What changed is that NVIDIA chose to spend the power. That shows in how the card responds to a lower limit: ComputerBase found it about 5% slower at 450 W and about 10% slower at 400 W, where a 4090 lost under 2% going from 450 to 350 W. Even at 400 W, the 5090 stayed about 12% faster than a stock 4090.

So for many owners the sensible setting is below the maximum. Community undervolts report 13–18% less power for 0–2% fewer frames, though results vary by card and game. We cover the method and the trade-offs in undervolting an RTX 5090.

Do you need a UPS?

A UPS won't stop the card's spikes, but it protects a long render or training run from a power interruption and lets the PC shut down cleanly. Size it to the whole system's real draw with headroom, not the card's rating. Our post on whether a S$200 UPS can protect a S$10,000 workstation explains the types and runtimes.

Buying in Singapore

RTX 5090 cards are not in stock at SourceIT at the time of writing; ask us about availability and a quotation. SourceIT doesn't list desktop power supplies or 12V-2x6 cables, so ask us when planning a build. For power protection, the APC Back-UPS Pro Gaming 2200VA (1,320 W) is S$645.00 and the APC Back-UPS Pro 1500VA is S$886.25 (both listed 6 Nov 2026). Prices change; ask for a current quotation. Use Request for Quotation on the product page, call (65) 6978 3502, email Sales@sourceit.com.sg or use our contact page.

Frequently asked questions

Does the RTX 5090 always use 575 W? No. Reviewers measured gaming averages of about 460–550 W depending on resolution and ray tracing, much less with a frame cap, and 27–46 W at idle. Very short spikes reach about 900 W.

What size power supply does an RTX 5090 need? NVIDIA specifies 1,000 W of system power and recommends a PCIe CEM 5.1-compliant unit. Many builders choose 1,200 W or more for headroom, especially for sustained AI or rendering.

How much does it cost to run an RTX 5090 in Singapore? At 31.16¢/kWh (Q4 2026, with GST), four hours a day at a full 575 W costs about S$262 a year. A card running 24/7 at 575 W costs about S$1,570 a year.

Is the RTX 5090 less efficient than the RTX 4090? Roughly the same. Reviews found performance per watt within about 5% of the 4090; the 5090 is faster mainly because it draws more power.

Will lowering the power limit hurt performance? A little. ComputerBase found about 5% less performance at 450 W and about 10% less at 400 W in games. For single-user AI chat, one community test saw about 0–5% loss at 400 W.