Benchmarking distributed tensorflow with affordable GPUs

We extend the benchmark results published in Tensorflow official website to more afforable hardware platforms with consumer grade GPUs, i.e. GTX 1080 cards with PCI switches.

A few image classification models are selected to add the reference points. The details of the methodology is describled in the Tensorflow website.

Test Environment

• Xpander Type: CUBIX
• GPU: 8x NVIDIA GTX 1080
• OS: Ubuntu 16.04 LTS
• CUDA / cuDNN: 8.0 / 5.1
• TensorFlow Version: 1.1
• CPU: 2x Intel Xeon E5-2620
• Disk: Local SSD
• DataSet: ImageNet
• System Topology:

nvidia-smi topo --matrix
        GPU0   GPU1    GPU2     GPU3    GPU4    GPU5    GPU6    GPU7    CPU Affinity
GPU0	 X     PIX     PIX      PIX	SOC	SOC	SOC	SOC	0-5,12-17
GPU1	PIX     X      PIX      PIX	SOC	SOC	SOC	SOC	0-5,12-17
GPU2	PIX    PIX      X       PIX	SOC	SOC	SOC	SOC	0-5,12-17
GPU3	PIX    PIX     PIX       X 	SOC	SOC	SOC	SOC	0-5,12-17
GPU4	SOC    SOC     SOC      SOC	 X 	PIX	PIX     PIX	6-11,18-23
GPU5	SOC    SOC     SOC      SOC	PIX	 X 	PIX     PIX	6-11,18-23
GPU6	SOC    SOC     SOC      SOC	PIX	PIX	 X 	PIX	6-11,18-23
GPU7	SOC    SOC     SOC      SOC	PIX	PIX	PIX	 X 	6-11,18-23