Azure NC Virtual Machines Availability – GPU Compute

Kyle Deeds, Program Manager

Azure NC-based instances are powered by NVIDIA Tesla® K80 GPUs and provide the compute power required to accelerate the most demanding high-performance computing (HPC) and AI workloads. Customers can now use these instances to run deep learning training jobs, HPC simulations, rendering, real-time data analytics, DNA sequencing, and many more CUDA® accelerated tasks. Additionally, customers have the option to utilize RDMA (Remote Direct Memory Access) over InfiniBand for scaling jobs across multiple instances. InfiniBand provides close to bare-metal performance even when scaling out to 10s, 100s, or even 1,000s of GPUs across hundreds of machines. This will allow you to submit tightly coupled jobs using frameworks like the Microsoft Cognitive Toolkit (CNTK), Caffe, or TensorFlow, enabling training for natural language processing, image recognition, and object detection.

Instance sizing information is listed in the chart below. For more information on NC-series VMs please see the compute documentation. Pricing details can be found on the Azure Pricing Calculator.

NC instances 

Size  CPU cores  Memory: GiB  Local SSD: GiB  GPU  Maximum data disks 
Standard_NC6  6 56  380 1  8
Standard_NC12  12  112  680  2  16 
Standard_NC24 24  224  1440  4  32 
Standard_NC24r*  24  224 1440  4 32 

1 GPU = one-half K80 card.+

*RDMA capable


Discussion is closed.

Feedback usabilla icon