“IBM Softlayer and LeaderGPU appear expensive, mainly due to under-utilisation of their multi-GPU instances. The benchmark was carried out using the Keras frame

“IBM Softlayer and LeaderGPU appear expensive, mainly due to under-utilisation of their multi-GPU instances. The benchmark was carried out using the Keras framework whose multi-GPU implementation was surprisingly inefficient, at times performing worse than a single GPU run on the same machine.“ - this is unacceptable in a benchmark like this. There is an entire software stack that is influencing performance and much of the hardware is dissimilar.