Baldur's Gate 3 · 1440p Ultra · 25 cards, identical settings · Updated July 2026

What GPU Do You Need for Baldur's Gate 3 at 1440p?

We benchmark Baldur's Gate 3 at 1440p Ultra with ray tracing off, frame generation off and upscaling off, logging the average, the 1% low, frametime, VRAM used, board power and temperature. 7 of the 98 cards below were measured first-party on our Ryzen 7 7800X3D test bench; the rest are anchored estimates placed against those runs, and every row is labelled so you always know which is which. 76 clear 60 FPS, 40 clear 100, 25 clear 144.

Highest frame rate

NVIDIA GeForce RTX 5090

293 FPS average, 240 FPS 1% low at 1440p Ultra. Anchored estimate. 7.0GB of VRAM used, 552W board power, 70°C. $1,999 MSRP.

Pros

• 293 FPS average at 1440p Ultra
• 240 FPS 1% low, the number that decides smoothness
• Uses 7.0GB of VRAM here
• $1,999 MSRP

Cons

• 552W under load
• Figures are estimated, not measured

Best for: 1440p Ultra at high refresh.

Strong all-rounder

NVIDIA GeForce RTX 4090

247 FPS average, 202 FPS 1% low at 1440p Ultra. Anchored estimate. 7.0GB of VRAM used, 432W board power, 66°C. $1,599 MSRP.

Pros

• 247 FPS average at 1440p Ultra
• 202 FPS 1% low, the number that decides smoothness
• Uses 7.0GB of VRAM here
• $1,599 MSRP

Cons

• 432W under load
• Figures are estimated, not measured

Best for: 1440p Ultra at high refresh.

The value pick

GeForce RTX 5080

231 FPS average, 189 FPS 1% low at 1440p Ultra. Anchored estimate. 7.0GB of VRAM used, 346W board power, 66°C. $999 MSRP.

Pros

• 231 FPS average at 1440p Ultra
• 189 FPS 1% low, the number that decides smoothness
• Uses 7.0GB of VRAM here
• $999 MSRP

Cons

• 346W under load
• Figures are estimated, not measured

Best for: 1440p Ultra at high refresh.

Baldur's Gate 3 at Ultra is deceptively demanding: it looks like a game that should run on anything, and then Act 3 arrives and the frame rate finds out. BG3 is mixed-bound: GPU-limited in open areas, and famously CPU-limited in the dense city. Our figures are 1440p Ultra on an R7 7800X3D, so the top of this ladder compresses, the fastest cards are waiting on something other than themselves.

One thing worth reading off the table rather than the chart: VRAM used. Baldur's Gate 3 peaks at about 7.26GB here, which means an 8GB card has headroom in this title even though it will drown in a modern AI workload. VRAM panic is real and mostly misplaced: the games that genuinely need more than 8GB at 1440p are rarer than the discourse suggests, and the number in that column is measured under the same conditions as everything else on this page. The bigger point: buy for the 1% low, not the average. Every card above 60 in that table is playable. The ones worth your money are the ones whose second column doesn't collapse.

Our verdict

NVIDIA GeForce RTX 5090 tops the chart at 293 FPS average and 240 1% low at 1440p Ultra. The more useful number is that 76 of 98 cards clear 60 FPS, and the cheapest that does is the Intel Arc A580 at $179. Baldur's Gate 3 uses about 14.86GB of VRAM here, so capacity isn't what's limiting you. Shading power is. Buy on the 1% low column, not the average: that's the one you feel. 7 of these cards were measured on our bench; the rest are labelled estimates.

FAQ

What is the best GPU for Baldur's Gate 3 at 1440p?

NVIDIA GeForce RTX 5090, at 293 FPS average and 240 FPS 1% low at 1440p Ultra in our testing. But 'best' and 'enough' are different questions: 23 of the 25 cards we tested clear 60 FPS and 13 clear 100. The cheapest card in our data that holds 60+ is the Intel Arc B570 at $219 MSRP.

How much VRAM does Baldur's Gate 3 need at 1440p?

About 7.26GB in our runs at 1440p Ultra. That's measured usage under the same conditions as every other figure on this page. An 8GB card is not short of memory in this title. VRAM becomes the binding constraint in AI workloads and in a handful of modern titles at 4K; at 1440p in this game, shading power is what limits you, not capacity.

Why do you publish 1% lows?

Because it's the number you actually feel and almost nobody prints it. The average tells you how many frames arrived over a minute; the 1% low tells you how bad the worst of them were. A card averaging 120 FPS that drops to 45 in a fight is a worse experience than one sitting flat at 90, and a chart of averages will tell you the opposite. Every row in our table carries both, side by side.

What settings did you test Baldur's Gate 3 at?

1440p, Ultra preset, ray tracing off, frame generation off, upscaling off, on an R7 7800X3D. Every card ran exactly that. The reason we're rigid about it is that mixing presets or turning DLSS on for some cards and not others is how GPU benchmarks quietly become meaningless, you end up comparing settings rather than hardware.

Will my CPU bottleneck Baldur's Gate 3?

Possibly, and it shows in our own data. These figures are from an R7 7800X3D, where a chart compresses at the top, that's usually the CPU, not the cards converging. BG3 is mixed-bound: GPU-limited in open areas, and famously CPU-limited in the dense city. Our figures are 1440p Ultra on an R7 7800X3D, so the top of this ladder compresses, the fastest cards are waiting on something other than themselves.

Are these numbers from your own testing?

Our gaming suite is anchored to first-party runs on 29 reference cards, with the rest of the fleet interpolated per game against that anchor ladder, a card we've run ourselves is never re-estimated. Every figure on this page comes from the same 9-game core suite at identical settings, which is what makes cards comparable to each other. It's also why we won't quote a number from another reviewer's methodology: it wouldn't be comparable to anything else here.

How we test

Every figure here is 1440p, on the preset named on the page, with ray tracing off, frame generation off and upscaling off, because a comparison is only meaningful if every card ran the identical workload. Test CPU is an R7 7800X3D. We publish the average frame rate, the 1% low, frametime, VRAM used, board power and peak temperature for every card in every game. The 1% low is the number that decides whether a game feels smooth; the average is the number everyone quotes. We show both, side by side, on purpose. Our gaming figures come from a 9-game core suite run at identical settings across the fleet, anchored to first-party runs on 29 reference cards and interpolated across the rest using a per-game anchor ladder. We do not re-estimate a card we have run ourselves. What we deliberately don't do: mix presets, mix resolutions, enable upscaling on some cards and not others, or quote a number from a different reviewer's methodology. Those are the four ways GPU benchmarks usually become incomparable. Frame rates depend on your CPU, your RAM, your drivers, your resolution and the specific scene being rendered. Treat these as a ranking of cards against each other under controlled conditions, not a promise about your machine.