AWS Bulks Up Cloud-Based Graphics Processing With New G3 Family


Printer-friendly version Email this CRN article

Amazon Web Services again pushed forward the limits for processing graphics in the cloud with the introduction on Friday of a new family of GPU-powered instances.

The cloud leader says the new G3s offer more compute and memory resources complimenting its graphics processors than any comparable cloud service. AWS partners told CRN the G3s could be the key to bringing to the masses next-generation applications like virtual reality.

The G3s are a good fit for companies running three-dimensional rendering and visualization workloads, or doing anything with virtual reality, video encoding and server-side graphics "that need a massive amount of parallel processing power," blogged Jeff Barr, AWS' chief evangelist.

[Related: Amazon Cuts Cloud Prices: New AWS Price Reductions Target Reserved Instances]

Amazon introduced GPUs to its public cloud back in 2013 with the launch of the G2 instances – the first to give customers high-performance graphics-processing capabilities.

The G3s take graphics acceleration to the next level. They come in three specific flavors, with varying numbers of GPU and CPU cores, memory and bandwidth.

The three specific G3 types range from 1, 2 or 4 GPU cores running Nvidia's Tesla M60 chip, complimented by four times the number of virtual CPUs. Each GPU supports 8 GiB of memory and 2048 parallel processing cores, according to Barr.

The instances come with licenses for the Nvidia GRID graphics virtualization platform.

"Our customers have told us that they are looking forward to visualizing large 3D seismic models, configuring cars in 3D, and providing students with the ability to run high-end 2D and 3D applications," Barr said.

The G3s could prove an important step toward mainstreaming virtual reality technology, said Greg Baker, director of DevOps at Pythian, an Ottawa, Ontario AWS partner.

And they will immediately help partners like Pythian migrate data analytics, 3D visualization and machine learning workloads out of corporate data centers and into the cloud.

Printer-friendly version Email this CRN article