PSU Wattage Guide 2026: How Many Watts Does Your GPU Need?

If you're speccing a new build or dropping a new graphics card into an old one, PSU wattage is the spec people get wrong most often. Not because the maths is hard, but because "meets the minimum" and "runs comfortably for years" are two different numbers, and most buyers only ever see the first one printed on the box.

This is a straightforward one to get right, and getting it wrong is expensive in a way that's easy to avoid. Undersize your power supply and you get random shutdowns under load, or worse, a dead PSU that takes other components with it. Oversize it wildly and you've spent money on headroom you'll never use. Here's how to land on the right number for whichever GPU you're pairing it with.

How much power your GPU actually draws

Graphics card power draw has crept up steadily across the last few generations, and it's now the single biggest line item in most gaming PCs' power budget. Manufacturer-recommended PSU wattages give you a starting point, though they already build in some headroom for a mid-range CPU and a few drives:

GPU	Typical board power	Nvidia/AMD-recommended PSU
RTX 5060 / 5060 Ti	~150-180W	550-650W
RTX 5070	~250W	650W
RTX 5070 Ti	~300W	700-750W
RTX 5080	~360W	850W
RTX 5090	~575W	1000W
RX 9060 XT	~150-180W	550-650W
RX 9070 / 9070 XT	~220-305W	700-750W
RX 7900 XTX (still widely sold)	~355W	800-850W

These aren't fixed figures. Factory-overclocked cards from board partners routinely draw more than the reference design, and the gap between "typical gaming load" and "worst-case power spike" is bigger than most people expect. That's the bit the recommended figure doesn't tell you.

Why the minimum recommendation isn't the number to buy

Modern high-end GPUs, particularly anything using Nvidia's 12V-2x6 connector on the RTX 40 and 50 series, don't draw a flat, steady wattage. They spike. Transient power spikes on an RTX 5090 can hit close to 900W for under a millisecond, well above its 575W typical board power. A PSU that's sized exactly to the "typical" figure can trip its own overcurrent protection on one of these spikes and shut the whole system down, even though average draw across the session is well within spec. This is a real, well-documented failure mode, not a theoretical one, and it's the reason enthusiast PSU reviewers now test transient response specifically. This guide is built on GPU makers' own published PSU wattage recommendations, the ATX 3.0/3.1 spec for connector and transient handling, and the transient-spike and connector-safety problems that PC hardware reviewers and builders have documented repeatedly since the 12V-2x6/12VHPWR connector arrived, rather than any single incident report.

There's a second reason to size up: every PSU has an efficiency sweet spot, usually somewhere between 40% and 60% of its rated capacity, where it runs coolest, quietest and most efficiently. A 650W unit powering a system that peaks at 600W is working flat out near its ceiling constantly. The same system on a 750-850W unit is cruising in its efficient zone, the fan spins slower, and the PSU itself lives longer.

So the practical rule: take your GPU's recommended PSU wattage, and if you're not already accounting for a power-hungry CPU (a Core i9 or Ryzen 9 X3D chip under full load adds meaningfully to the total), don't go below it. If anything, treat the manufacturer figure as a floor rather than a target.

Undervolting and power limits: the free upgrade nobody buys

One of the more useful things you can do with any modern GPU costs nothing and doesn't involve buying a bigger PSU at all: undervolting. Most current-generation cards ship with more voltage than they need for their factory clock speeds, and dropping the voltage curve a notch typically costs a small, often unnoticeable amount of performance while cutting power draw and heat meaningfully. Pairing that with a power limit in software (MSI Afterburner, or the GPU vendor's own tuning tool) caps the card's ceiling rather than just nudging its average, which is exactly what stops those transient spikes tripping a marginal PSU.

This matters most for two groups: people upgrading a GPU in an older system without replacing the PSU, and anyone running a demanding card on a unit that's a size class below the "ideal" recommendation. It's not a substitute for an adequately rated PSU, but it's a genuinely useful buffer, and it's the first thing worth trying if a system is shutting down under load before you assume the PSU itself has failed.

Efficiency rating and why the connector matters as much as the label

Once you know your target wattage, the two other specs worth paying attention to are the 80+ efficiency rating and how the unit handles high-power connectors.

80+ Bronze, Gold and Platinum ratings describe how much power is lost as heat rather than delivered to your components at various load levels. Gold is the sensible default for most builds today: a reputable 650-850W 80+ Gold unit sits in the mid-range for PSU pricing, and the jump to Platinum rarely pays for itself in electricity savings on a typical gaming rig. For current PSU pricing, browse the full product range rather than relying on a fixed figure here, since unit prices shift with the market. Where it's worth spending more is on build quality and warranty length from an established maker, since a failing PSU is one of the few component failures that can take the rest of your build down with it.

Connector safety deserves its own mention. For any RTX 4080-class card or above, use the native 12V-2x6 (or 12VHPWR) cable that comes with a genuinely ATX 3.0/3.1-rated PSU, seated fully, rather than a daisy-chained adapter splitting off two older 8-pin connectors from a single cable run. Partially-seated or adapter-heavy connections at this power level are the leading cause of the melted-connector failures that circulated widely with early 40-series cards, and it's a completely avoidable one. If you're buying a new PSU specifically for an RTX 50-series or high-end RX 9000-series card, an ATX 3.0/3.1 unit with a native 12V-2x6 cable is worth the small premium over an older ATX 2.x unit with an adapter in the box.

Matching PSU to GPU tier: practical numbers

As a working baseline for a full system (mid-range CPU, a couple of drives, standard cooling):

• Budget/1080p builds (RTX 5060, RX 9060 XT): 550-650W 80+ Bronze or Gold is genuinely enough. Don't let anyone talk you into 850W here.
• 1440p mainstream (RTX 5070, RTX 5070 Ti, RX 9070): 650-750W Gold, leaning toward 750W if you're pairing with a higher-end CPU or plan to keep the system for several GPU generations.
• High-end/4K (RTX 5080, RX 7900 XTX): 850W Gold as a sensible floor, especially if overclocking either the GPU or a power-hungry CPU.
• Enthusiast (RTX 5090): 1000-1200W, and don't skimp on the connector spec here of all places.

These bands assume a single GPU and a conventional air-cooled build. If you're running multiple storage drives, an all-in-one liquid cooler with several fans, or a heavily overclocked CPU alongside the GPU, add another 50-100W of headroom on top.

Upgrading an existing PC rather than building fresh

If you're dropping a new card into an existing system, the question isn't just "is my PSU rated high enough on paper", it's "is my PSU still delivering what its label says". PSUs do degrade, particularly cheaper units run hard for years, and a five-year-old 650W unit isn't necessarily still a healthy 650W unit. Two sensible checks before you commit to a new GPU purchase: confirm the age and brand of your current unit (anything unbranded or from a case bundle is worth replacing regardless of headroom), and if you want real numbers rather than guesswork, a plug-in mains power meter (around £15-£20) will show you exactly what your system draws at the wall under load, which is the most honest way to know whether you've actually got room.

If your current PSU is a known-quality brand, rated comfortably above the new card's recommendation, and less than four or five years old, there's usually no need to replace it just because you're upgrading the GPU. If it's the generic 500W unit that came in a budget case a decade ago, treat the PSU as part of the upgrade cost, not an afterthought, before you even look at GPU prices.

The bottom line

Size to the manufacturer's recommended wattage for your GPU as a floor, add headroom for a demanding CPU, buy 80+ Gold or better from an established brand, and if you're pairing with an RTX 40 or 50-series card, make the native 12V-2x6 cable non-negotiable rather than reaching for an adapter. It's one of the cheaper decisions in a build, and by far the least forgiving one to get wrong. Check current pricing and stock across the full GPU range before you commit, and if you're weighing up the whole build rather than just the graphics card, our PC build cost guide breaks down where the rest of the budget typically goes.