Every figure below is a measurement aggregated from published review charts — the CPU's frame-rate ceiling and the frame rate the GPU delivers in this game, combined. The CPU figure is the highest frame rate that CPU sustained in this game with a flagship GPU installed — past that frame rate, the CPU is the limit. Settings: Ultra preset, native rendering, no upscaling, no frame generation. CPU ceiling measured at 1080p, Low preset (reviewers' CPU-isolation method), flagship GPU.
Balanced pairing
Expected frame rate: 54–74 fps (measured range across sources).
The CPU ceiling and the GPU's delivered rate sit within 13.4% of each other — neither part wastes much of the other. The gap is measured between the two figures above: (faster − slower) ÷ faster.
That range suits a 60 Hz monitor; a high-refresh panel would be underused in this game.
Confidence: LOW — based on 3 source measurements for these two parts.
Balanced pairing
Expected frame rate: 54–58 fps (measured range across sources).
The CPU ceiling and the GPU's delivered rate sit within 9.2% of each other — neither part wastes much of the other. The gap is measured between the two figures above: (faster − slower) ÷ faster.
That range suits a 60 Hz monitor; a high-refresh panel would be underused in this game.
Confidence: LOW — based on 3 source measurements for these two parts.
Strong GPU limit
Expected frame rate: 37–37 fps (measured range across sources).
The GPU side is lower by about 43.0% in this game at 4K. The math, from the two measured figures above: the pair runs at the lower one (~37 fps), and the gap to the faster side is (64 − 37) ÷ 64 ≈ 43.0% of the faster side's output. That percentage belongs to this game, this resolution and these settings only — the same two parts score differently in another game, which is why no universal bottleneck percentage exists for a pair of parts. Upgrading the faster part alone would not raise the frame rate.
That range sits under a smooth 60 in this game at the tested preset — expect to trade settings for frame rate.
Confidence: LOW — based on 3 source measurements for these two parts.
Full data: Starfield benchmarks · how verdicts are computed.