GPU Power & Efficiency Report 2026: the fastest card is one of the least efficient

Modern graphics cards keep getting faster — and a lot hungrier. We pulled every desktop GPU in the SpecPeak database that has both a PassMark G3D score and a rated TDP, 131 cards in all, and looked at how raw performance, power draw and efficiency have moved from 2016 to 2026. Three things stood out, and the last one genuinely surprised us.

Power draw is up about 92% in a decade

The typical card we track pulled 132W in 2016. By 2025 that figure is 253W — a 92% jump. Peak performance roughly tripled over the same window, so we are getting far more for those watts. But the direction of travel is unmistakable: each generation leans a little harder on the power budget, and the flagships lean hardest of all.

The most efficient cards are not the flagships

Performance-per-watt tells a very different story than the top of the leaderboard. The RTX 5090 is the fastest card we track by a clear margin — and it ranks 104th of 131 on efficiency. The cards that actually lead on PassMark-per-watt are mid-range parts: the RTX 4060, the RTX 4070, the RX 9060. Halo silicon buys you frames, not efficiency.

Put bluntly: the 2023 RTX 4060 delivers roughly 2.5 times the performance-per-watt of the 2025 RTX 5090. You are not buying one to replace the other — but it is a sharp reminder that efficiency and raw power are two different races.

A generation that went backwards

Here is the part we did not expect. Newer does not automatically mean more efficient. The 2023 RTX 4060 posts 169.6 PassMark per watt. Its 2025 successor, the RTX 5060, manages 143.1 — the older card is about 18% more efficient. The same pattern shows up a tier up, between the 4060 Ti and the 5060 Ti. If a quiet, cool, power-sane build is the goal, "latest" is not a guarantee of "better per watt." Worth a side-by-side before you upgrade: RTX 4060 vs RTX 5060.

Why this matters for your next build

If you are sizing a power supply, planning a small-form-factor case, or simply paying the electricity bill, raw FPS is only half the decision. The efficiency sweet spot almost always sits a tier or two below the flagship — and as our data shows, sometimes a generation behind it. Decide what you actually need at your resolution first, then pick the most efficient card that clears that bar.

Methodology

Efficiency here is PassMark G3D score divided by rated TDP (manufacturer board power). PassMark is a synthetic benchmark and rated TDP is not the same as measured draw under load, so treat these figures as a consistent comparative proxy rather than a wall-power reading. The dataset covers the 131 desktop GPUs on SpecPeak that have both values, as of June 2026. Our full scoring approach is on the methodology page. You are free to cite this report with a link back.