Rendering on Intel Xeon Phi

CyclesPhi

We have modified the kernel of the Blender Cycles rendering engine and then extended its capabilities to support the HPC environment. We call this version the CyclesPhi and it supports following technologies:

  • OpenMP
  • Intel® Xeon Phi™ with Offload concept
  • Intel® Xeon Phi™ with Symmetric mode

Benchmarks v1

I used the scenes from this page: https://code.blender.org/2016/02/new-cycles-benchmark/

Computed devices:

The time is without preprocessing of the scene ( BVH, … ). Blender 2.77 was compiled with ICC.

Benchmarks v2

I used the scenes from this page: https://code.blender.org/2016/02/new-cycles-benchmark/

Computed devices:

  • IT4Innovations
    • Salomon
      • 2x Intel® Xeon™ E5-2680v3 (HASWELLS)
      • NVIDIA® GeForce® GTX TITAN X
    • Workstation (Windows 10)
      • NVIDIA® GeForce® GTX 970
  • Marconi-A2 CINECA
    • Intel® Xeon Phi™ 7250 (KNL)

Blender 2.79 was compiled with GCC.

Run CyclesPhi

Requirements

  • Intel® Manycore Platform Software Stack (Intel® MPSS)
  • GCC 6.3.0
  • other external libraries are included

Scripts

  • run_blender.sh – run Blender

Settings