Skip to content
WattFit

GPU power draw: rated TBP vs measured watts

Vendor-rated board power next to what independent reviews actually measure at the card — gaming load and desktop idle — for the current three GPU generations.

As of July 2026, measured gaming power draw runs from roughly 100 W (RTX 4060 class) to 550–590 W (RTX 5090), and cards typically idle at 7–35 W; the rated TBP is a power limit, so real gaming draw often sits 5–20% below it on non-overclocked models.

The table lists NVIDIA's rated total graphics power (TGP), AMD's typical board power (TBP), and Intel's rated power alongside a measured band: the spread of card-only gaming measurements published by independent labs. A band is shown instead of a single number because measured draw genuinely varies — between games, resolutions, and board-partner variants of the same GPU.

Sources & reference date

Rated power and connector: NVIDIA, AMD, and Intel official specification pages. Measured bands: synthesized from card-only measurements published by TechPowerUp, Gamers Nexus, Igor's Lab, and Tom's Hardware, among others. Figures checked 2026-07-13. Vendor PSU column is the power supply size the GPU vendor states as the system minimum — it assumes a typical CPU, not a specific one.

Graphics cardRated TBP (W)Measured gaming (W)Measured idle (W)Power connectorVendor min. PSU (W)
GeForce RTX 5090575550–59030–4612V-2x6 (600 W)1000
GeForce RTX 5080360300–33015–2512V-2x6 (450 W)850
GeForce RTX 5070 Ti300275–30512–2512V-2x6 (300 W)750
GeForce RTX 5070250225–25010–2012V-2x6 (300 W)650
GeForce RTX 5060 Ti 16 GB180145–17010–1712V-2x6 or 1x 8-pin (AIB)600
GeForce RTX 5060145120–1458–151x 8-pin (AIB)550
GeForce RTX 5050130105–1308–151x 8-pin (AIB)550
GeForce RTX 4090450350–43019–2916-pin 12VHPWR (450 W)850
GeForce RTX 4080 Super320270–32012–2216-pin 12VHPWR (320 W)750
GeForce RTX 4070 Ti Super285250–28510–2016-pin 12VHPWR (300 W)700
GeForce RTX 4070 Super220190–2259–1816-pin 12VHPWR (300 W)650
GeForce RTX 4070200175–2059–1616-pin 12VHPWR or 1x 8-pin650
GeForce RTX 4060 Ti160130–1657–131x 8-pin (AIB)550
GeForce RTX 4060115100–1307–121x 8-pin (AIB)550
Radeon RX 9070 XT304270–34010–222–3x 8-pin (AIB)750
Radeon RX 9070220195–23010–202x 8-pin (AIB)650
Radeon RX 9060 XT 16 GB160145–18012–251x 8-pin (AIB)450
Radeon RX 7900 XTX355330–36515–302x 8-pin (reference)800
Radeon RX 7900 XT315280–33015–302x 8-pin (reference)750
Radeon RX 7800 XT263235–26512–352x 8-pin700
Radeon RX 7700 XT245215–24512–302x 8-pin700
Radeon RX 7600165140–1658–201x 8-pin550
Intel Arc B580190140–1907–351x 8-pin600
Intel Arc B570150110–1507–301x 8-pin600

How to read the measured band

The low end of each gaming band is typical of raster gaming at 1440p or a lighter engine; the high end reflects 4K, ray tracing, or a factory-overclocked board-partner card. Reviewers measure at the card's power connectors and PCIe slot with lab equipment, so these are card-only figures — whole-system wall draw adds the CPU, motherboard, drives, fans, and PSU conversion losses, which the methodology page covers. If you cap your frame rate or play esports titles, real draw can sit well below even the low end of the band.

Rated power is a limit, not a bill

NVIDIA's TGP and AMD's TBP define how much power the card is allowed to draw before the boost algorithm backs off. Well-behaved cards like the RTX 5080 or RTX 5060 Ti spend most of their gaming time 10–20% under the limit, while an RTX 5090 pushes at its 575 W ceiling in heavy 4K workloads. That is why cost estimates on this site start from the measured band rather than the sticker figure — multiplying TBP by your gaming hours systematically overstates the bill for most cards.

Idle draw is where setups differ most

The idle column has the widest relative spread because it depends on display configuration more than on the silicon. A single 1080p or 1440p monitor at 60–144 Hz produces the low figures; adding a second display or running 4K/240 Hz can hold the memory clock high and double or triple idle draw. Intel's Arc B-series is the extreme case: roughly 30–35 W idle by default, dropping to about 7 W once ASPM is enabled in the BIOS and OS power settings. Early RTX 50-series drivers also idled high (the RTX 5090 was measured at 30–46 W on the desktop at launch), which later driver updates have been reducing.

Related data

To turn these watts into dollars, use the monthly running-cost table (per card, at the US average rate) together with the state electricity rate table. For the physical side of an upgrade — whether the card fits at all — see the GPU case fit checker and the CPU cooler clearance checker.

FAQ

How many watts does a GPU actually use while gaming?

Between roughly 100 W for entry cards (RTX 4060, Arc B570) and 550–590 W for the RTX 5090, measured at the card. Most mid-range cards (RTX 5070, RX 9070) draw 200–250 W in games. The exact figure depends on the game, resolution, frame cap, and the specific board-partner model.

Is TDP the same as actual power consumption?

No. TDP/TBP is a design power limit set by the vendor, not a measurement. Reviews regularly measure gaming draw below the limit (RTX 5060 Ti: ~145–170 W vs a 180 W limit) and factory-overclocked variants can run at or slightly above it. Idle draw is far lower than TDP in every case.

Why is idle power a range instead of one number?

Because idle draw depends on monitor count, resolution, and refresh rate. A single 1080p/60 display idles lowest; multi-monitor and high-refresh setups can double or triple idle draw, and Intel Arc cards need ASPM enabled in BIOS to reach their low figures.

How much does GPU power draw cost per month?

At the April 2026 US average residential rate of 18.83¢/kWh, a 300 W card gaming 4 hours a day costs about $6.90 per month. The running-cost table lists this per card, and the state rate matters — Hawaii is about 2.5x the national average.

Which GPU family is most power efficient in 2026?

In published reviews, NVIDIA's RTX 50 series holds the top of the performance-per-watt charts, with the RTX 5080 measured around 300–330 W in gaming. AMD's RDNA 4 cards narrowed the gap versus RDNA 3, while Intel's Arc B-series is competitive at the budget end but has higher idle draw unless ASPM is configured.