CPU scaling benchmark

W=0 W=1 W=2 W=3 W=4 W=6 W=8 W=9 W=10 W=12 W=16 iters: 100M 500M 1B 2B
workers
16 +1 main
iters total
100M
5882352/stream
elapsed
314.16 ms
total CPU used
2932.44 ms
speedup
9.33×
vs serial
efficiency
54.9%
of 17× ideal
stream spawn ms spawned@ work start@ work end@ work ms reap wait ms
0 (main) 0 66.79 66.8 291.39 224.59 0
1 2.172 2.19 18.76 186.26 167.5 0.19
2 1.864 4.07 16.66 190.99 174.33 0.25
3 2.953 7.04 20.41 189.49 169.08 0.26
4 2.138 9.19 46.19 194.78 148.59 0.28
5 5.745 14.95 49.01 261.63 212.62 0.29
6 3.382 18.35 78.25 210.88 132.63 0.3
7 1.641 20.01 54.81 214.99 160.18 0.32
8 5.437 25.46 78.06 276.6 198.54 0.33
9 4.326 29.8 88.23 288.33 200.1 0.34
10 8.202 38.02 74.21 244.37 170.16 0.35
11 20.28 58.31 100.79 294.53 193.74 4.9
12 2.179 60.5 138.24 310.79 172.55 19.54
13 1.621 62.14 148.23 301.18 152.95 9.94
14 1.545 63.7 128.27 276.97 148.7 0.36
15 1.561 65.27 118.22 293.37 175.15 3.14
16 1.493 66.78 148.23 279.26 131.03 3.11
main
w1
w2
w3
w4
w5
w6
w7
w8
w9
w10
w11
w12
w13
w14
w15
w16
    fork+handshake      CPU work      parent reap wait
what this measures
Each stream runs a tight integer LCG loop — working set is one CPU register, no memory access, no shared data. Speedup = sum(stream CPU time) / wall-clock elapsed. Efficiency = speedup / (workers+1). 100% efficiency means perfect linear scaling; less than 100% is the cost of serial fork setup, reap tail, SMT/core contention.