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
500M
29411764/stream
elapsed
1183.83 ms
total CPU used
17667.13 ms
speedup
14.92×
vs serial
efficiency
87.8%
of 17× ideal
stream spawn ms spawned@ work start@ work end@ work ms reap wait ms
0 (main) 0 49.56 49.57 1085.69 1036.12 0
1 2.127 2.14 32.86 1170.36 1137.5 84.81
2 1.625 3.79 47.7 1125.14 1077.44 39.55
3 1.878 5.68 22.91 1067.51 1044.6 0.14
4 1.654 7.36 30.17 1034.68 1004.51 0.19
5 1.63 9.01 65.61 1152.59 1086.98 72.09
6 1.699 10.72 48.74 1046.73 997.99 0.21
7 1.657 12.4 51.95 1068.28 1016.33 0.22
8 1.67 14.09 54.64 1103.08 1048.44 17.48
9 1.676 15.79 78.54 1133.65 1055.11 53.2
10 1.605 17.41 59.02 1043.63 984.61 0.23
11 8.543 25.97 68.31 1089.05 1020.74 3.47
12 2.01 28 75.84 1079.31 1003.47 0.24
13 1.706 29.73 78.06 1095.56 1017.5 10
14 15.721 45.47 98.57 1158.26 1059.69 75.34
15 2.294 47.78 116.74 1181.05 1064.31 95.48
16 1.753 49.55 151.64 1163.43 1011.79 77.84
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.