TR
x = (−1)
s
× 1.f × 2
e
; ulp(x) = 2
e−p
(normal, 2
e
≤ |x| < 2
e+1
); ε
32
= 2
−23
, ε
64
= 2
−52
float32 (binary32)
float64 (binary64)
0.0001 s
0.001 s
0.01 s
0.1 s
1024 s