Using GPUs in Microsoft Azure

Microsoft has recently rolled out two categories of GPU enabled VMs within Azure. The NC series are specialized for compute workloads and use K80 GPUs, and the NV series are specialized for visualization using M60 GPUs.

We will walk thru what it takes to set up an Azure VM with multiple K80s and what is required to be able to create and debug a CUDA program that uses multiple GPUs. The pricing model is based on a per hour cost, but as an example a 24 CPU core VM with 224Gb RAM and the equivalent of 2 K80 devices is currenly listed at about $5.20 per hour that the VM is running. The VM can be stopped, when not needed. If there is sufficient interest, a later meeting can be setup to discuss how to programtically start/stop the VMs for finer grained elasticity of resources and cost control.

We will also briefly talk about some of the templates available for Deep Learning and Data Science Toolkit VMs. We will not dig into these, but just make you aware that the templates exist and can save a lot of setup time and effort.

There will be additional time for open discussions of any topics.

This will be in Keller 3-111. This room is on the "street level" floor of the building. Please enter the building via the main doors which are on the south side of the building.

A link to the parking website can be found here. http://www.pts.umn.edu/park/eventparkingcalendar