Benchmarks

Every number here is measured and traceable. Test machine: HP OMEN, Ryzen 7 8845HS + RTX 4070 Laptop GPU (dedicated) + Radeon 780M (integrated).

Live tests (Subway Surfers, browser)

Pipeline	Game FPS	Output FPS	Notes
RIFE interpolation	~30	63 fps	inference avg 21.0ms, dGPU AI load 0%, 11,500+ frames
RIFE + ESPCN 2x	~16	33 fps	960x540 AI-upscaled (not native), 95-96% pacing
FlowUp	native	45-48 fps	720p native, warp 1.7-2.2ms, E2E 16-18ms, 53,000+ frames

The output is consistently about 2x the game's frame rate.

iGPU inference (Radeon 780M, DirectML, 500 frames, no display)

270p: inference 12.87ms → 74.9 fps.
ESPCN 2x upscale on the iGPU: 3.49ms.

POOLS (Unity, RTX 4070 Laptop, 1080p, 20s)

Baseline average: 59.5 fps.
GlassBreakr real output: 68.5 fps.
Dedicated-GPU impact: 9.6%.

The "2x effective 119" sometimes quoted is a calculation (59.5 × 2), not delivered frames; the real delivered output is 68.5 fps. We do not publish the "≈ RTX 4090" comparison because the 4090 figure is estimated, not measured.

Frame multiplier (per tier) — verified

The engine generates up to 8x frames: FREE 4x, TURBO 6x, ULTRA 8x. Verified by disassembling the shipped binary (multipliers {4,6,8}, no clamp) and by the beta tester's Cyberpunk test below. Honest nuance: generated ≠ displayed — a 180 Hz screen shows ~180 fps (≈4x visible), more on faster displays.

Cyberpunk 2077 — beta tester test

Tier	Game fps	GlassBreakr output	Ratio
FREE	22	120 fps	5.45x — counter artifact, above FREE's 4x cap
TURBO	30	148 fps	4.9x
ULTRA	45	283 fps	6.3x

Proofs (screenshots, beta-tester report PDF + recording, the shipped binary) are in the glassbreakr-proofs repository.