ND and NCv2 virtual machines (VMs), as announced at Build 2017, are now available.
The ND-series, powered by NVIDIA Tesla P40 GPUs based on the new Pascal Architecture, will be excellent for training and inference. These instances will provide more than twice the performance over the previous generation for FP32 (single precision floating point operations), for AI workloads utilizing CNTK, TensorFlow, Caffe, and other frameworks. The ND-series also offer a much larger GPU memory size (24-gigabytes), enabling you to fit much larger neural net models. Also, like our NC-series, the ND-series will offer RDMA and InfiniBand connectivity, so you will be able to run large-scale training jobs spanning hundreds of GPUs.
The next generation of our NC-series, the NCv2, is powered by NVIDIA Tesla P100 GPUs. These new GPUs will provide more than twice the computational performance of our current NC-series. They will also offer InfiniBand networking configurations for workloads that require fast interconnect like Oil and Gas, Automotive, and Genomics to also accelerate scale out capability and improved single instance performance.
Learn more about these new VM series.
Source: Azure Roadmap