PSU Calculator: Power Supply Wattage Calculator

Short answer: most gaming PCs need a 550W to 850W power supply, and the graphics card decides which end you land on. An RTX 5060 build is comfortable on 550W, an RTX 5060 Ti on 600W, an RTX 5070 on 650W, an RTX 5070 Ti on 750W, an RTX 5080 on 850W, and an RTX 5090 on 1000W — NVIDIA’s own published minimums.

AMD publishes a 450W reference-system minimum for the RX 9060 XT, 650W for the RX 9070, and 750W for the RX 9070 XT. Pick your exact parts in the power supply calculator above: it calculates component demand separately and uses the higher of that result and the vendor floor.

How to check your PSU’s wattage

Look at the power supply itself. Its label names the model; Corsair, for one, puts that label on the side or top of the unit, with a part number that starts CP-. Look the model or part number up on the maker’s site and its product page usually gives the wattage.

That figure is the unit’s capacity. Your PC only draws what its parts need. Add them up in the calculator above and compare its minimum with the unit’s rated wattage.

Worked examples

These are this calculator’s own outputs, computed with the same formula the tool runs in your browser — so you can see a real answer without touching a single input.

Mainstream 1440p gaming build — Ryzen 7 7800X3D + RTX 5060 Ti (OC model), 2x DDR5, one Gen3 NVMe, 3 case fans, one AIO, ATX board. The calculator opens with the CPU allowance blank and withholds a result. Enter the stated 120W allowance to reproduce this worked example.

• Inputs: CPU = Ryzen 7 7800X3D, explicit user CPU allowance = 120W; GPU = RTX 5060 Ti OC (200W peak); DDR5 sticks = 2; Gen5 NVMe = 0; Gen4 NVMe = 0; Gen3 NVMe = 1; SATA SSDs = 0; HDDs = 0; case fans = 3; AIO = 1; USB/misc = 10W; motherboard = ATX (45W)
• Result: Peak draw 409W | Calculated minimum 550W | NVIDIA floor 600W (650W at the retail sizes compared) | Displayed minimum 650W | Sweet spot 850W. The vendor floor, rather than the component calculation, sets the displayed minimum.

High-end 4K gaming build — Ryzen 7 9800X3D + RTX 5080 Founders Edition, 2x DDR5, one Gen5 and one Gen4 NVMe, 6 case fans, one AIO, ATX board

• Inputs: CPU = Ryzen 7 9800X3D, explicit user CPU allowance = 120W; GPU = RTX 5080 Founders Edition (360W peak); DDR5 sticks = 2; Gen5 NVMe = 1; Gen4 NVMe = 1; Gen3 NVMe = 0; SATA SSDs = 0; HDDs = 0; case fans = 6; AIO = 1; USB/misc = 10W; motherboard = ATX (45W)
• Result: Peak draw 587W | Calculated minimum 750W | NVIDIA floor 850W | Displayed minimum 850W | Sweet spot 1200W. The vendor floor, rather than the component calculation, sets the displayed minimum.

Homelab NAS / Proxmox box — Ryzen 5 5600X on a Micro-ATX board, no dedicated GPU, one NVMe boot drive and six 3.5-inch hard drives, four case fans, air cooler

• Inputs: CPU = Ryzen 5 5600X, explicit user CPU allowance = 65W; GPU = No dedicated GPU / integrated only (0W); DDR5 sticks = 2; Gen5 NVMe = 0; Gen4 NVMe = 0; Gen3 NVMe = 1; SATA SSDs = 0; HDDs = 6; case fans = 4; AIO = 0; USB/misc = 10W; motherboard = mATX (35W)
• Result: Peak draw 266W | Calculated and displayed minimum 550W | Sweet spot 550W. No GPU vendor floor applies. The six drives account for 132W of that peak, at 22W each while they spin up.

PSU wattage by graphics card

In a balanced, gaming-first build the graphics card is the single largest draw — roughly 45% of system peak on an entry card and up to about 75% on a flagship — so it is the fastest way to a starting number. Every row below carries its own source, because two different kinds of number appear in this table and the difference matters.

RTX 3060, RTX 3050 variants and RTX 4070 Ti

Read the capacity label on an RTX 3050. NVIDIA gives the 8GB reference card 130W graphics-card power and 550W required system power; the 6GB version is 70W / 300W. The RTX 3060 reference figure is 170W / 550W. These values come from the separate columns in NVIDIA’s comparison table. Choose the matching option here, then account for the rest of your build and your card maker’s instructions.

The RTX 4070 Ti without “Super” has a 285W TGP and 700W reference-system requirement in that table. It shares those two numbers with the Ti Super, but that does not make the cards interchangeable. NVIDIA’s RTX 40-series reference system uses a Ryzen 9 5900X; its RTX 30-series comparison uses a Core i9-10900K. Neither figure is a measurement of your own PC’s wall power.

The calculator’s retail choices start at 550W. That is why the 3050 6GB’s published 300W system figure can still produce a 550W tool minimum, even with a modest CPU. It does not revise NVIDIA’s specification. The itemized demand calculation can also require more than a vendor floor.

RX 6600, RTX 3090 and a 650W PSU for RTX 3070

AMD lists 132W board power and a 450W minimum PSU for the RX 6600. NVIDIA lists 350W / 750W for RTX 3090 and 220W / 650W for RTX 3070. The graphics-card number is not the entire PC. NVIDIA bases these RTX 30-series system figures on a Core i9-10900K configuration and says requirements vary with the system and board manufacturer. The source links in their rows provide the exact specifications.

A suitable 650W unit meets NVIDIA’s RTX 3070 reference-system figure; that does not certify every 650W supply or every overclocked partner card. Check the exact card’s instructions, PSU condition, connectors and the rest of the build. Choose the new reference option in the calculator to apply its published floor alongside the component calculation. RX 6600 and RTX 3090 follow the same method; no extra legacy-card multiplier is added.

Why a vendor floor can differ from the calculator. A vendor minimum is a whole reference-system figure: NVIDIA’s own note on the RTX 5090 states its 1000W figure is based on a system built around a Ryzen 9 9950X, while AMD documents the processors used for its RX 7000-series figures. The number therefore carries assumptions about a processor and other parts you may not own.

The calculator also produces an itemized minimum from your selected CPU, GPU estimate, board, memory, storage, cooling, and miscellaneous load, adds 20%, and steps up to a retail size. It uses whichever is higher: that calculated demand or the published vendor floor. It does not apply a vendor floor where the primary specification omits one, as AMD’s current RX 7900 GRE specification table does.

These are minimums, not the calculator’s answer. The tool below targets roughly half load rather than a floor, so its headline recommendation will read higher than this table — on a flagship build, considerably higher. That gap is efficiency and upgrade headroom, not a transient-spike requirement. If you want the smallest defensible unit, use this table; if you want the quietest and most efficient one, use the tool’s number.

Note also that 600W is a size few brands actually sell — where a vendor figure lands between retail steps, buy the next size up. For a worked example of how one card’s number changes with the processor beside it, see the RTX 50-series PSU guide — per-card tables for the 5070 through the 5090, including the one case where the itemized math lands above NVIDIA’s own floor.

How this power supply calculator works

No vendor’s calculator shows you its math. Here is ours, in full.

1. It sums component power allowances. The CPU term is a configured socket/package limit or an explicit allowance. Intel desktop presets use Maximum Turbo Power (MTP); AMD, N100 and custom selections require you to enter a limit. GPU figures come from the dropdowns; everything else is itemized and added: motherboard (about 45W for ATX, less for Micro-ATX and Mini-ITX, more for EATX), 5W per stick of RAM, 10W per Gen5 M.2 drive, 8W per Gen4, 6W per Gen3, 3W per SATA SSD, 22W per hard drive (the spin-up peak, when drives draw most), 2W per case fan, 12W for an AIO pump and fans, plus whatever you enter in the miscellaneous field (10W by default, for fan controllers, RGB, capture cards, and similar).

That total is your estimated peak allowance. It is a sizing budget, not a measured maximum at the wall. Board overhead and the separate 20% margin remain estimates.

2. The minimum is peak plus 20%, or the GPU vendor’s published floor if that is higher. The calculator multiplies your peak by 1.2 and rounds up to the next size commonly sold — 550, 650, 750, 850, 1000, 1200, or 1500W; above 1500W it steps in 100W increments, because retail sizes get irregular there. It separately checks a published NVIDIA or AMD reference-system minimum where one is available and uses the higher result, rounded up the same way. That is why 600W becomes 650W in the comparison table and AMD’s 450W RX 9060 XT floor becomes 550W.

That is the smallest unit it will suggest. At exactly 1.2x, your peak sits at about 83% of the unit’s rating, so treat the minimum as “this will work,” not “this is the one to buy.”

3. The recommended size targets about half load. Power supplies are most efficient, coolest, and quietest loaded near the middle of their range, so the headline recommendation is the size that puts your peak near half the unit’s rating, and the tool flags 40-50% as the sweet spot when your build lands there. It is deliberately larger than the minimum: it buys efficiency, quiet, and room for a future graphics card. It is not a spike requirement — see below.

4. The efficiency table shows every size. The table rates each retail wattage against your build. A size rated below your peak is too small. A size under the minimum from step 2 would carry the load with less margin than the calculator asks for, or may sit under the GPU vendor’s published figure. A size that meets the minimum but runs above 65% works with little upgrade room, and 35-65% is the band worth buying into.

The transient-spike question, answered

Modern graphics cards pull microsecond spikes well above their rated power, and this is where PSU advice usually turns into “buy double.” It should not.

Intel’s ATX 3.0 design guide sets explicit excursion brackets a compliant unit must survive, and they are tiered by duration rather than being one blanket number:

The qualifier matters as much as the number. That 200% bracket applies to units above 450W that ship a native 12VHPWR / 12V-2x6 connector. A smaller unit, or one without that connector, is held to a lower bar — 150% at 100 µs. So “my PSU is ATX 3.x, it handles 200% spikes” is only true for the larger, natively-connectored units.

Where it applies, though, it settles the sizing question: a 750W unit in that class is built to ride out a 1500W spike lasting 100 microseconds. That headroom is already in the standard, so on a qualifying unit you do not buy a bigger PSU to cover transient spikes — you buy a compliant one. On an older ATX 2.x unit paired with a high-wattage card, leave more room, because it was never designed to these brackets.

What ATX 3.1 revised is the connector: the 12V-2x6 design shortens the sense pins so a plug that is not fully seated fails to deliver full power rather than making partial high-resistance contact. We are still not going to tell you whether 3.1 raises or lowers the transient bar relative to 3.0, because we have not read a source that compares the two brackets directly, and we do not publish spec claims we have not read.

Buy 3.1 for the connector. (We have since read Intel’s ATX v3 design guide for a different figure — the low-load efficiency point it sets at 10W — which is covered in the PSU tier list.)

One more that vendor calculators skip: if your unit is multi-rail, total wattage is not a single pool. A 1000W multi-rail supply may cap one 12V rail well below the card’s draw, and a high-wattage GPU fed entirely from that rail can trip over-current protection on a supply that is nominally huge. Follow the rail distribution in your PSU’s manual and use separate cables rather than daisy-chaining one pigtail.

Reading the three numbers

Continuous vs. peak vs. transient. A PSU’s headline number is its continuous rating — what it can deliver all day. Your parts list a peak draw, which the calculator sums. Transient spikes are the third number, and the standard above covers them. A unit sized only to the raw continuous total, with no margin at all, can trip its over-current protection under load, which shows up as a random shutdown rather than an error message.

Why the sweet spot is roughly half load. Efficiency curves peak in the middle of a unit’s range, and a supply loafing at half load runs cooler and spins its fan slower — often not at all, on units with a zero-RPM mode.

Worth keeping the two measures apart: the calculator’s load percentages are measured against your peak, while the 50-60% figure people quote describes typical sustained draw, which sits below peak. Size so peak stays under about 80% and typical lands near half, and both are satisfied at once.

80 Plus tiers in plain terms. The 80 PLUS badge rates efficiency at fixed load points. At these wattages the bill difference between Gold and Platinum is a dollar or two a month; treat the rating mostly as a signal of build quality, heat, and fan noise.

Gold is the floor worth buying for a gaming or always-on homelab box. Our EVGA SuperNOVA 1000 GT review walks through what that looks like in a real unit, and the power & cost calculator turns watts into an annual dollar figure.

A lower-power build is a cheaper build. Undervolting cuts real wattage — around 100W on a high-end card and roughly 50W on a mid-range one, going by the measured results in our own database — which can drop you a PSU size and lower your bill at the same time. The complete GPU undervolting guide covers the method, and our undervolt results database lists measured power savings by card.

If your PC is already shutting down

An undersized or failing supply looks the same from the outside: reboots under load, shutdowns in games, a PC that will not post. PSU failure symptoms covers how to tell a dying unit from an undersized one, and the FAQ below covers testing a supply and finding out which one you already own.

These are conservative component estimates. Real-world idle draw is far lower, and your actual numbers depend on overclocks and the specific models you choose — for a full parts list with compatibility checking, use the PC Builder, and for where each card lands on speed, see the GPU hierarchy.

CPU power connectors: what does “12 pin” mean?

Count the sockets separately. For example, ASUS lists one 8-pin +12V and one 4-pin +12V connector in the TUF Gaming B650-Plus WiFi manual. That is an 8+4 motherboard arrangement, not a single 12-pin GPU socket. Your own board may use a different arrangement.

Use the CPU/EPS cables specified for that exact PSU and follow the motherboard manual for which sockets must be populated. Seasonic’s cable guide distinguishes CPU 4+4 from PCIe GPU connectors. Do not substitute a GPU cable or force a matching-looking modular plug. A wattage calculation cannot verify connector compatibility.

Can software measure the PC’s actual power usage?

Software can expose supported component sensors. It does not turn a GPU reading into whole-PC wall power. On a machine with NVIDIA’s driver utility available, this read-only command requests its power report:

nvidia-smi -q -d POWER

NVIDIA’s documentation defines GPU power readings for the board and says unsupported data appears as N/A. Read the field’s scope and units. Do not count a reported power limit as actual consumption, or add a component reading to a larger total that already includes it.

For electricity use at the outlet, measure the PC’s AC input with a suitably rated plug-in power meter, following its instructions. Record which devices are connected. Component DC demand and wall AC draw differ because the PSU loses some energy during conversion; Corsair explains that distinction. This calculator estimates a build’s demand for sizing. It is not a live power meter.

Corrections

2026-10-23 - Component edits now withhold stale estimates. A blank, negative, fractional or out-of-range component count pauses the result while you edit. A negative count becomes a visible zero when you leave the field; other invalid counts need correction. Sharing requires a valid visible scenario. Zero components are allowed, but the selected AMD CPU still needs its separate power allowance. The 409W worked example above requires entering 120W; it is not a result shown on page load.

2026-08-15 — Core i7-14700K peak draw corrected from 192W to 253W. The CPU dropdown is documented above as carrying peak figures, and its other Intel entries do exactly that: the i9-14900K and i9-13900K rows both use 253W, which is Intel’s published Maximum Turbo Power for those parts.

The i7-14700K row carried 192W, which matches neither of Intel’s two published power figures for that processor — Intel ARK lists it at 125W Processor Base Power / 253W Maximum Turbo Power (read from Intel’s own specification page on 2026-08-15). The 61W understatement flowed straight into every calculation a 14700K owner ran, pushing peak draw, the minimum size and the recommended size all low. Corrected to 253W, which restores the table’s own Intel convention.

Scope note: this correction covers the one row re-derived from the vendor page. Several other rows in the same dropdown appear to carry base TDP where the page promises peak — AMD entries in particular, whose peak package power (PPT) AMD does not publish on its product pages. I have recorded those as open findings in my audit log and left them unchanged until each is checked against its source; see the caveat below.

2026-08-29 — three Intel rows corrected from base power to Maximum Turbo Power. The same defect class as the 14700K entry above, in three more rows: Core i5-14600K and Core i5-13600K both carried 125W, and Core Ultra 5 245K carried 125W. In every case that is Intel’s Processor Base Power, not the peak figure this dropdown is documented as carrying.

Corrected to Intel’s published Maximum Turbo Power — 181W for the i5-14600K and i5-13600K, 159W for the Core Ultra 5 245K. The understatement was worth one retail PSU size on mainstream pairings: an i5-13600K with an RTX 5070 Ti computed 514W and recommended a 650W unit, where the true 570W peak recommends 750W — and 570W of 650W is 88% load, which this page’s own efficiency table flagged at the time as “Over — risk of shutdowns under peak load”.

Change-safety note: correcting these three removed the last value="125" option from the CPU dropdown. A shared link carrying cpu=125 would previously have landed on no option at all and silently computed with a 0W CPU. The URL restore path was hardened first, in the same change, to ignore any value the dropdown does not offer and keep its default selection instead.

2026-10-12 - CPU sizing now uses a declared power-limit convention. The old AMD presets mixed TDP, PPT and unexplained figures: for example 200W for the 9950X, 230W for the 7950X, and 120W for X3D models. Those values no longer feed the calculation. AMD, N100 and custom selections require a CPU power allowance; an empty or invalid value withholds peak and PSU sizes. Intel desktop presets keep their published MTP figures and accept an explicit override for a changed board limit.

Read the configured socket/package ceiling from BIOS or Ryzen Master and enter watts. Check the setting for the profile you plan to use, including PBO or Eco changes. TDP describes a thermal specification and does not supply this allowance. A complete stock-PPT table for these AMD SKUs was not established by this check, so the calculator supplies no replacement guessed constant. The field is included in shared links.

2026-09-29 — two AMD rows raised to AMD’s board power, and the RTX 5060 Ti floor corrected. The RX 7700 XT row carried about 200W and the RX 7600 XT row about 150W, both labeled as peak draw. AMD’s own specification pages list 245W and 190W typical board power, so a peak figure below them understated both builds. Both rows now carry AMD’s figures.

The two RTX 5060 Ti options floored the headline at 550W, while the table above and NVIDIA’s page give 600W. They now floor at 600W, rounded up to 650W, and the tool’s own default build changed with them: its minimum reads 650W where it read 550W.

The same pass fixed two things in the tool itself. The efficiency table flagged every size above 80% load as a shutdown risk, including the minimum the headline recommends, which can load a unit to 83%; it now judges each size against that minimum and against your peak. And two worked examples above had drifted from the tool: the 4K build did not apply NVIDIA’s 850W floor for the RTX 5080, and the NAS build still counted hard drives at 12W each after the tool moved to 22W. Both now print what the tool prints.

2026-10-02 — GPU share links and vendor floors corrected. GPU selections in new share links now use a stable option ID. Older links saved only wattage, which cannot identify a card when values collide: 360W can mean an RTX 5080 Founders Edition or an RTX 4080 Super OC, and those families have different vendor floors. When an old wattage is ambiguous, the tool now restores the wattage as “model unknown,” applies no vendor floor, and asks the reader to select the exact GPU. It no longer silently chooses the first match.

The same pass added primary-source reference-system floors for the RTX 40 series and AMD cards where NVIDIA or AMD publishes one. AMD’s live pages now provide the RX 9070 XT (304W TBP / 750W minimum), RX 9070 (220W / 650W), RX 9060 XT 16GB (160W / 450W), and RX 9060 XT 8GB (150W / 450W). The RX 9070 XT and RX 9060 XT links below replace the dead URL patterns previously checked. The generic RTX 5060 Ti option retains NVIDIA’s 600W reference floor; ASUS independently lists 550W for its specific Dual RTX 5060 Ti 8GB OC board, which is recorded as a board-specific exception in the PC Builder dataset.

Sources

• NVIDIA GeForce RTX 50-series official specifications (total graphics power and minimum system power requirement) — RTX 5090, RTX 5080, RTX 5070 family, RTX 5060 family — verified live 2026-07-24; the 5070 and 5060 family pages re-verified 2026-07-25, and the 5060 family page again on 2026-09-29 (RTX 5060 Ti: 180W, 600W required system power; RTX 5060: 145W, 550W).

• NVIDIA GeForce RTX 40-series official specifications — RTX 4090 (450W TGP / 850W required system power), RTX 4080 family (320W / 750W), RTX 4070 family (285W / 700W for 4070 Ti Super; 220W / 650W for 4070 Super; 200W / 650W for 4070), and RTX 4060 family (165W / 550W for 4060 Ti; 115W / 550W for 4060) — verified live 2026-10-02. NVIDIA says these figures represent Founders Edition or reference designs and add-in-card specifications can vary.

• Intel ATX 3.0 / ATX 3.1 desktop power supply design guides — the 200%-of-rated power-excursion requirement for microsecond transients (Intel ATX 3.0 PSU power excursion).

• AMD Radeon RX 9000-series official specifications — RX 9070 XT (304W TBP / 750W minimum), RX 9070 (220W / 650W), RX 9060 XT 16GB (160W / 450W), and RX 9060 XT 8GB (150W / 450W) — verified live 2026-10-02.

• AMD Radeon RX 7000-series official specifications — RX 7900 XTX (355W / 800W), RX 7800 XT (263W / 700W), RX 7700 XT (245W / 700W), and RX 7600 XT (190W / 600W) — verified live 2026-10-02. AMD’s current specification table lists 260W board power for the RX 7900 GRE but no minimum PSU, so the calculator does not assign that option a vendor floor.

• ASUS, Dual GeForce RTX 5060 Ti 8GB OC specifications and recommended PSU table — ASUS lists 550W for this specific board with a Core i5 / Ryzen 5 class CPU; NVIDIA’s generic RTX 5060 Ti reference-system figure is 600W. Verified live 2026-10-02.

• Intel desktop processor power: Processor Base Power and Maximum Turbo Power from Intel ARK product specification pages — Core i7-14700K (125W base / 253W turbo), read 2026-08-15. The CPU dropdown carries the turbo figure, because the calculator sums peak draw.

• Component draw coefficients are this calculator’s own published figures, listed in full in the methodology section above.

• Corsair, Find Part & Serial Number Labels — a Corsair PSU’s label is on the side or top of the unit; part numbers begin with CP-. Read 2026-09-28.

• AMD Ryzen Master reference guide, April 2022, page 24, defines PPT as socket power capacity in watts; the current Ryzen Master utility page links the current user guide and documents profiles and BIOS controls. Checked 2026-10-12. These documents support the input convention, not a stock PPT value for every CPU in this selector.