Here’s the number you came for: NVIDIA recommends a 1000W power supply for the RTX 5090, 850W for the RTX 5080, 650W for the RTX 5070, and as low as 550W for the RTX 5060. Those figures come straight from NVIDIA’s own system requirements for each card.
That’s not the whole story, though. The recommended wattage is a baseline built around a specific CPU, and it doesn’t fully capture how these cards actually draw power under load — or the connector situation, which has turned out to be more complicated than a simple spec number suggests. Both of those gaps matter more than the headline wattage figure, so this covers both.
NVIDIA’s official numbers, card by card
| GPU | Total Graphics Power (TGP) | NVIDIA-recommended system PSU | Connector |
|---|---|---|---|
| RTX 5090 | 575W | 1000W | 12V-2×6 |
| RTX 5080 | 360W | 850W | 12V-2×6 |
| RTX 5070 Ti | 300W | 750W | 12V-2×6 |
| RTX 5070 | 250W | 650W | 12V-2×6 |
| RTX 5060 Ti | — | 600W | 12V-2×6 |
| RTX 5060 | — | 550W | 12V-2×6 |
These recommendations are NVIDIA’s own published figures, and every major PSU manufacturer’s own sizing guidance — Corsair, Seasonic, and others — builds directly from these same numbers rather than independently re-deriving them. Worth noting: NVIDIA’s recommended-PSU figure for the RTX 5090 assumes a high-core-count CPU (testing guidance has referenced something in the Ryzen 9 9950X class) — if you’re running a more modest CPU, you have a bit more real-world margin than the headline number implies, though NVIDIA doesn’t publish a lower figure for that case, so treat 1000W as the number to buy toward regardless.
The gap between TGP and what the card actually pulls
TGP, the number in that table, is a steady-state power rating — essentially an average the card is designed to sustain. It is not the peak, and the gap between the two is bigger than most buyers expect.
Tom’s Hardware’s own power-draw testing on the RTX 5090 found real-time spikes reaching roughly 900W to 1000W during gameplay, against the card’s official 575W TGP rating — nearly double, for brief moments, even though the card is rated to average well under that. That’s exactly the kind of transient spike modern PSU standards are built to absorb without the system browning out or shutting down, but it also means a PSU sitting right at NVIDIA’s 1000W recommendation is already running close to 60% load from the GPU alone before you’ve accounted for the CPU, drives, or anything else in the system. Tom’s Hardware’s practical recommendation for anyone wanting real headroom, particularly with a factory-overclocked card, was an 1500W 80 Plus Platinum unit rather than stopping at NVIDIA’s baseline number.
If 1500W sounds like overkill for your budget, there’s a more modest middle path worth knowing about: the same testing found that capping the RTX 5090’s power limit to around 80% in NVIDIA’s own control software brought peak draw down to roughly 460W, low enough that an 800–850W PSU could run the card safely, just with a small, usually hard-to-notice performance trade-off. That’s a genuinely useful option if you already own a capable PSU and don’t want to replace it purely to accommodate a new GPU.
The connector is the part that actually needs your attention
This is the part a wattage number alone won’t tell you, and it’s worth taking seriously rather than skimming past.
RTX 50-series cards use a connector called 12V-2×6 — a revised version of the 12VHPWR connector that debuted with the RTX 4090 and quickly became notorious for melting at the GPU end in a number of documented cases. The replacement, 12V-2×6, was introduced specifically to address that problem, with shortened sensing pins designed to prevent a partially-seated connector from drawing full power, which was identified as a major contributor to the original failures.
Here’s the uncomfortable part: the fix didn’t fully solve the problem. Multiple documented incidents of 12V-2×6 connectors melting on RTX 5090 cards have continued to surface well after the revised connector launched, including at least one case involving a native single-cable setup on a professional review outlet’s own system — not a third-party adapter, not a cheap cable, and not obvious user error. The RTX 5090 draws enough current through this connector, at 575W TGP with spikes well beyond that, that even small amounts of pin-to-pin current imbalance or a connector that isn’t fully seated can create a real thermal risk. This is specific to the 5090’s power draw — it’s a far smaller concern on the 5070 or 5060, which simply don’t push anywhere near as much current through the same connector design.
What actually reduces the risk, based on guidance from PSU reviewers and manufacturers alike:
- Use the native 12V-2×6 cable that shipped with your power supply, rather than an adapter bundling multiple 8-pin connectors into one 12V-2×6 plug, wherever your PSU supports it. A PSU engineer interviewed by Yahoo Tech was specific that third-party extenders and adapters, not the adapters NVIDIA and its board partners ship with the cards themselves, are where the real risk concentrates — mixing and matching cables between brands, or using older pre-2025 cables not rated for Blackwell-generation power draw, is a common failure point.
- Seat the connector fully until it clicks, and avoid bending the cable sharply within roughly 35mm of either end — a stressed, not-quite-seated connector is the scenario the shortened sensing pins were specifically designed to catch, but a cable that looks seated while actually carrying uneven current across its pins can still run hot.
- If your current PSU doesn’t have a native 12V-2×6 output, the bundled adapter from NVIDIA or your card’s board partner is considered reliable by PSU engineers — it’s specifically third-party replacement cables and adapters, especially older stock, that carry the most risk.
None of this means the RTX 5090 is unsafe to run. It means the margin for error on this specific card’s power connector is thinner than on any consumer GPU before it, and the small amount of care above is the difference between a non-issue and a genuine hazard.
Does your PSU need to be ATX 3.0 or 3.1?
Not strictly, but it’s the easier path. ATX 3.0 and 3.1 power supplies were designed around exactly the transient power spikes described above, and they’re the ones that ship with a native 12V-2×6 cable rather than requiring an adapter. If you’re buying a new PSU for an RTX 50-series build, an ATX 3.0 or 3.1 unit is the straightforward choice specifically because it removes the adapter question entirely.
If you already own a solid, reasonably recent PSU that isn’t ATX 3.0/3.1 and meets NVIDIA’s wattage figure for your card, you don’t necessarily need to replace it — the bundled adapter that ships with the GPU is designed for exactly this situation, and PSU engineers consider it a safe option as long as you’re not layering additional third-party adapters on top of it.
What this means per tier
RTX 5090 builds: Buy at least NVIDIA’s 1000W recommendation, but don’t treat it as generous headroom — it’s closer to a floor given the real transient spikes this card produces. A 1200W unit is a more comfortable target, and 1500W with 80 Plus Platinum efficiency is worth considering if you’re running a factory-overclocked card or just want to stop thinking about it. Pay close attention to the connector guidance above specifically for this card.
RTX 5080 builds: NVIDIA’s 850W recommendation has reasonable headroom built in already, since the 5080’s 360W TGP is considerably lower than the 5090’s. An ATX 3.0/3.1 unit is still worth choosing for the native cable, but the safety margin concerns that apply to the 5090 are much less pressing here.
RTX 5070 / 5070 Ti builds: 650–750W covers these cards comfortably in nearly every realistic configuration. If you already own a quality PSU in this range, there’s rarely a reason to upgrade purely for one of these cards.
RTX 5060 / 5060 Ti builds: 550–600W is genuinely sufficient here, and most existing mid-range systems already clear this bar without any PSU changes at all.
FAQ
Is a 1000W PSU enough for an RTX 5090?
It meets NVIDIA’s official recommendation, but given that real-world testing has recorded transient spikes approaching 900–1000W from the GPU alone, a 1000W PSU leaves little margin for the rest of your system. 1200W or higher is a more comfortable target if your budget allows it.
Do I need an ATX 3.0 or ATX 3.1 power supply for RTX 50-series cards?
Not strictly required if your current PSU meets NVIDIA’s wattage figure, since the bundled adapter NVIDIA ships with the card is considered safe by PSU engineers. An ATX 3.0/3.1 unit is the simpler choice for a new build, mainly because it includes a native 12V-2×6 cable and removes the adapter question entirely.
Why do RTX 50-series power connectors still melt if the connector was redesigned?
The 12V-2×6 connector’s shortened sensing pins fixed one specific failure mode — a partially seated connector drawing full power — but documented melting incidents have continued on the RTX 5090 specifically, which draws enough current that even small pin-to-pin imbalances or imperfect seating can create a thermal risk. It’s much less of a concern on lower-power RTX 50-series cards.
Can I use my old 12VHPWR cable with an RTX 50-series card?
Cable manufacturers have specifically advised against using pre-2025 12VHPWR cables with RTX 50-series GPUs, since they weren’t designed for the increased power draw of this generation. Use the adapter or cable that shipped with your card, or a cable explicitly rated for 12V-2×6 and RTX 50-series power levels.
Is it safe to use a PSU adapter with multiple 8-pin connectors for an RTX 5090?
The adapter NVIDIA ships with the card itself is considered reliable. The risk concentrates in third-party adapters and extenders, particularly older or lower-quality ones not specifically rated for this generation’s power draw — avoid stacking adapters, and use a single, properly rated cable wherever possible.