gpu blueprint · stagebench

Does math cost the same in every kind of shader?

ReShade effects are mostly pixel shaders, while OpBench measures in compute shaders. StageBench runs OpBench's tests again in pixel shaders and in vertex shaders, so the cost models can follow the kind of shader an effect uses.

1.0 = one fma (a fused multiply-add at full rate), measured as throughput, lower is better. cs = compute shader (OpBench), ps = pixel shader, vs = vertex shader, each from the card's newest report. Highlighted: more than 20% and 0.75 fma away from the compute shader.

Where the kind of shader changes the cost

Tests that differ on at least one card. Each card has three bars on one scale per test: compute pixel vertex, bright where a stage differs from compute. Hover a test for the code it times, a card's bars for the numbers. Left out here, but in the tables below: cards whose clock moved more than 30% during a run, or whose pixel or vertex results differ from compute as a whole by more than 15% (every test would show).

Every StageBench test

Every test on every card in one kind of shader, in fma units on top of the test's base (lower is better, 0 = free), from each card's newest report. Hover a test for the code it times.

One card in detail