Microsoft is lobbing a bunch of GPUs into its Azure cloud to try and attract HPC-type workloads.
The GPU-enabled VM option was one of two additions to the Azure lineup at its AzureCon conference – the other is Azure DV2, based on Intel's Haswell processor.
The GPU option, Azure N, targets remote visualisation and “compute intensive with Remote Direct Memory Access (RDMA)” workloads.
As The Register's HPC sister-site The Platform notes, Nvidia's Tesla M60 GPUs (with Grid 2.0 virtualised graphics) target the visualisation market, while its K80s are for compute workloads.
Exactly how this is integrated into Redmond's existing kit hasn't yet been detailed, but The Platform's Timothy Prickett-Morgan takes a shot: “we think Microsoft is adding GPUs to its Xeon-based compute nodes by adding adjacent GPU trays that sit alongside the CPU trays in the Open Cloud Server.”
Microsoft's corporate veep for Azure Jason Zander told AzureCon attendees the N instances will be in preview “within a few months” and will go general-availability early in 2016.
Zander said the Azure N virtual machines' RDMA will reduce latency between notes.
The DV2 instances are based on the Haswell Xeon E5 processors, and The Platform says the E5-2673 v3 processors “appear to be a custom part with sixteen cores with a Turbo Boost speed bump to 3.2 GHz and a base clock speed of 2.4 GHz”.
More at The Platform here. ®