Category Archives: GPU virtualisation

AWS bolsters GPU-accelerated instances

AWS is updating its GPU-accelerated cloud instances

AWS is updating its GPU-accelerated cloud instances

Amazon has updated its family of GPU-accelerated instances (G2) in a move that will see AWS offer up to times more GPU power at the top end.

Announced on the tail end of 2013, AWS teamed up with graphics processing specialist Nvidia to launch the Amazon EC2 G2 instance, a GPU-accelerated instance specifically designed for graphically intensive cloud-based services.

Each Nvidia Grid GPU offers up to 1,536 parallel processing cores and give software as a service developers access to higher-end graphics capabilities including fully-supported 3D visualization for games and professional services.

“The GPU-powered G2 instance family is home to molecular modeling,  rendering, machine learning, game streaming, and transcoding jobs that require massive amounts of parallel processing power. The Nvidia Grid GPU includes dedicated, hardware-accelerated video encoding; it generates an H.264 video stream that can be displayed on any client device that has a compatible video codec,” explained Jeff Barr, chief evangelist at AWS.

“This new instance size was designed to meet the needs of customers who are building and running high-performance CUDA, OpenCL, DirectX, and OpenGL applications.”

The new g2.8xlarge instance, available in US East (Northern Virginia), US West (Northern California), US West (Oregon), Europe (Ireland), Asia Pacific (Singapore), and Asia Pacific (Tokyo), offers four times the GPU power than standard G2 instances including: 4 GB of video memory and the ability to encode either four real-time HD video streams at 1080p or eight real-time HD video streams at 720P; 32 vCPUs; 60 GiB of memory; 240 GB (2 x 120) of SSD storage.

GPU virtualisation is still fairly early on in its development but the technology does open up opportunities for the cloudification of a number of niche applications in pharma and engineering, which have a blend of computational and graphical requirements that have so far been fairly difficult to replicate in the cloud (though bandwidth constraints could still create performance limitations).