Team Red Build

Build list:

  • 32 gb (2 16 gb dimms) of 3000 MHz GSkill Trident Z
  • Ryzen 7 1700
  • Vega 56 (ASRock blower)
  • Wraith Spire
  • B450 Aorus M
  • 128gb ssd
  • 2 tb hard drive
  • 750 watt power supply
  • Rosewill scm-01 case

Tensorflow benchmarks:

InceptionV3

  • 67.08 Images per second for batch size of 64.
  • Not enough ram at batch size of 80

VGG16

  • 80.57 Images per second for batch size of 64

VGG16 (fp 16)

  • 52.66 Images per second for batch size of 64

Resnet 50

  • 125.27 Images per second for batch size of 64
  • 116.76 Images per second for batch size of 80
  • OOM for batch size of 128

Resnet 50 (fp 16)

  • 138.12 Images per second for batch size of 64

Resnet 50 (fp 16 export TF_ROCM_FUSION_ENABLE=1)

  • 145.19 Images per second for batch size of 64
  • 153.18 Images per second for batch size of 128

Stuff I tried out to get it running outside of docker

Running vgg19 was not possible. Maybe due to tf 1.13.1