Skip to content
This repository has been archived by the owner on Apr 2, 2021. It is now read-only.

In depth SIMD level

Jordan Peck edited this page Apr 7, 2017 · 11 revisions

CPU instruction support

SSE2

  • Intel Pentium 4 - 2001
  • AMD Opteron/Athlon - 2003

SEE4.1

  • Intel Penryn - 2007
  • AMD Bulldozer - Q4 2011

AVX

  • Intel Sandy Bridge - Q1 2011
  • AMD Bulldozer - Q4 2011

AVX2

  • Intel Haswell - Q2 2013
  • AMD Carrizo - Q2 2015

FMA3

  • Intel Haswell - Q2 2013
  • AMD Piledriver - 2012

AVX512

  • Not available to consumers yet

Performance Comparisons

Timings below are x1000 ns to generate 32x32x32 points of noise on a single thread.

  • CPU: Intel Xeon Skylake @ 2.0Ghz
  • Compiler: Intel 17.0 x64
Noise Type AVX512 AVX2 SSE4.1 SSE2 FastNoise FastNoise 2D
White Noise 7 9 16 29 141 111
Value 92 152 324 436 642 361
Value Fractal 290 476 1035 1362 2061 1101
Perlin 147 324 592 795 1002 473
Perlin Fractal 453 989 1818 2420 3209 1493
Simplex 129 294 548 604 1194 883
Simplex Fractal 399 895 1689 1839 2880 2589
Cellular 851 1283 2679 2959 2979 1074
Cubic 615 952 1970 3516 2979 858
Cubic Fractal 1860 2873 5957 10604 9583 2849
CF Sampled 1 360 537 1011 1708
CF Sampled 2 76 146 235 340
CF Sampled 3 37 90 123 142

CF (Cubic Fractal) Sampled {sampling level}: Comparison of performance increases with different levels of sampling

Clone this wiki locally