Skip to content

Latest commit

 

History

History
55 lines (42 loc) · 1.9 KB

benchmark.md

File metadata and controls

55 lines (42 loc) · 1.9 KB
Raw

Benchmark

Demo Shapes

Data dim: (4000, 64, 64) SOM dim: (8, 8) Rotations: 360 GPU: RTX 2080

Execution s/it
binary with CUDA [^1] 7
binary without CUDA [^2] 135
train.py [^3] 127
colab demo without CUDA 127
colab demo without CUDA @ colab.research.google.com 170

[^1]

./build/bin/Pink --train /data/pink/shapes_v2.bin som.bin --som-width 8 --som-height 8

[^2] --cuda-off

[^3]

./python/pink/train.py /data/pink/shapes/* -v

Radio Galaxy Zoo

Data (176750, 124, 124) SOM (21, 21)

The input data for the SOM training are radio-synthesis images of Radio Galaxy Zoo containing 176750 images of the dimension 124x124. The SOM layout is hexagonal of the dimension 21x21 which has 331 neurons (see image above). The size of the neurons is 64x64. The accuracy for the rotational invariance is 1 degree and the flip invariance is used.

PINK 1 PINK 2
CPU-1 35373
CPU-1 + NVIDIA Tesla P40 3069 909
CPU-1 + 2x NVIDIA Tesla P40 2069 636
CPU-1 + 4x NVIDIA Tesla P40 1891 858
CPU-2 + NVIDIA RTX 2080 673
CPU-3 + NVIDIA GTX 750 Ti 7185
CPU-4 + 2x NVIDIA RTX 2080 SUPER         477

All times are in seconds.

  • CPU-1: Intel Gold 5118 (2 sockets, 12 physical cores per socket)
  • CPU-2: Intel Core i7-8700K (1 socket, 6 physical cores per socket)
  • CPU-3: Intel Core i7-4790K (1 socket, 4 physical cores per socket)
  • CPU-4: Intel Gold 6230 (1 socket, 20 physical cores per socket)