By: Michael Feldman

Microsoft has added two of NVIDIA’s premier GPUs into its Azure public cloud offering.  According to a blog post by Azure Director of Program Management Cory Sanders, the company has created two new lines of virtual machine instances, one based on the Tesla K80, the other on the Tesla M60.

The K80 GPU will be used in the VMs belonging to the NC-Series, where the “C” refers to the compute-centric nature of the K80 GPU, a 2.9 teraflop device aimed specifically at HPC work. Software development can be done with the usual NVIDIA-supported GPU frameworks, namely CUDA and OpenCL.  Sanders calls out applications like crash simulations, ray-traced rendering, and deep learning as areas of interest, although any software that can be accelerated by GPUs and is fair game.

GPU virtualization is accomplished via Microsoft’s Discrete Device Assignment (DDA) feature, which is a hardware past-through mechanism that provides something close to bare metal performance – an important attribute for the kind of compute-intensive applications these instances are aimed at.

Here are the specs for the NC-Series line-up:

	NC6	NC12	NC24
Cores	6 (E5-2690v3)	12 (E5-2690v3)	24 (E5-2690v3)
GPU	1 x K80 GPU (1/2 Physical Card)	2 x K80 GPU (1 Physical Card)	4 x K80 GPU (2 Physical Cards)
Memory	56 GB	112 GB	224 GB
Disk	380 GB SSD	680 GB SSD	1.44 TB SSD

 

The M60 GPU will be the basis for the NV-Series. In this case, the “V” stands for visualization, which is the main capability this particular GPU was designed for.  More specifically, the M60 is designed to support the visualization capabilities of virtual desktops in a cloud setting. Here FLOPS don’t matter so much; rather, performance is measured by the number of desktop displays that can be handled. In the case of the M60, up to 36 streams of 1080 displays can be supported.

The application set is skewed toward users with more visually demanding applications, such as AutoCAD and Adobe Premier Pro, the idea being that once you’ve done the modeling and simulation work on the K80 instance, you can switch to the M60 instance for the visualization. For CAD/CAE types of workloads, real-time modeling and visualization is possible.

The specs for the NV-Series instances are as follows:

	NV6	NV12	NV24
Cores	6 (E5-2690v3)	12 (E5-2690v3)	24 (E5-2690v3)
GPU	1 x M60 GPU (1/2 Physical Card)	2 x M60 GPU (1 Physical Card)	4 x M60 GPU (2 Physical Cards)
Memory	56 GB	112 GB	224 GB
Disk	380 GB SSD	680 GB SSD	1.44 TB SSD

 

Sanders says this preview “offers the first public cloud support for this bleeding edge of specialized hardware.” However, that’s not strictly true since IBM recently added both the K80 and M60 into its own public cloud, including into the SoftLayer bare metal servers. Presumably, once NVIDIA’s Pascal GPUs become generally available, both Microsoft and IBM will add those into their respective cloud offerings as well.

The Microsoft NC and NV instances will be available in preview for Azure developers, something the company had originally planned to have ready by December 2015. For whatever reason, the creation of these GPU virtual machines took a good deal longer than expected. The current plan is make the preview initially available in the South Central region and expand to additional regions in the next couple of months. General availability is scheduled before the end of the year. Pricing has not been made public.