After running JetStream on Chrome 56.0.2924.87 / Ubuntu 16.10 (Linux xxx 4.8.0-45-generic #48-Ubuntu SMP Fri Mar 24 11:46:39 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux) some deviations are greater than the average: mandreel-latency 84.26 ± 105.6 mandreel 115.3 ± 142.2 bigfib.cpp 241.7 ± 253.8 box2d 107.2 ± 186.0 splay 121.1 ± 136.3 Just in case, here are all the results: Score 122.58 ± 41.822 Benchmark Average Score Benchmark Average Score Benchmark Average Score Latency 75.410 ± 17.469 regex-dna 128.7 ± 4.499 gcc-loops.cpp 351.2 ± 71.42 3d-cube 59.11 ± 6.473 splay-latency 255.9 ± 129.7 hash-map 119.6 ± 66.36 3d-raytrace 70.21 ± 1.745 tagcloud 65.74 ± 12.12 mandreel 115.3 ± 142.2 base64 69.42 ± 1.424 typescript 51.16 ± 51.08 n-body.c 199.0 ± 119.1 cdjs 63.98 ± 3.945 Throughput 178.59 ± 76.212 navier-stokes 130.5 ± 89.12 code-first-load 62.32 ± 7.358 bigfib.cpp 241.7 ± 253.8 pdfjs 108.6 ± 80.48 code-multi-load 52.85 ± 35.35 box2d 107.2 ± 186.0 proto-raytracer 167.0 ± 47.36 crypto-aes 114.5 ± 7.219 container.cpp 200.5 ± 158.1 quicksort.c 236.1 ± 58.82 crypto-md5 72.58 ± 1.264 crypto 116.1 ± 44.74 regexp-2010 267.4 ± 72.76 crypto-sha1 50.00 ± 1.514 delta-blue 296.4 ± 43.91 richards 167.7 ± 28.27 date-format-tofte 53.28 ± 5.979 dry.c 326.7 ± 49.59 splay 121.1 ± 136.3 date-format-xparb 58.33 ± 10.15 earley-boyer 113.1 ± 4.630 towers.c 159.4 ± 23.60 mandreel-latency 84.26 ± 105.6 float-mm.c 337.9 ± 14.01 zlib 242.5 ± 22.73 n-body 131.5 ± 11.04 gbemu 224.5 ± 18.94 Geometric Mean 122.58 ± 41.822 I think that some calculation is wrong, potentially turning the final score doubtful.
Since these are confidence intervals computed using 3 samples, you can expect some individual scores to have intervals that are larger than the average. To be clear, your complaint would be valid if we were reporting stddev. We're not. We're reporting 95% confints. Without specific evidence suggesting otherwise, I'm gonna say your machine is just noisy. More likely, Chrome just got super noisy with the ignition/turbofan kerfuffle.