AWS Launches Next-Generation GPU Instances
Amazon Web Services (AWS) this week unveiled a new Elastic Compute Cloud (EC2) instance family designed for graphics-intensive workloads.
The new G3 instances "make it easy to procure a powerful combination of central processing (CPU) and random-access memory (RAM)," according to the AWS announcement on Friday.
Available out of AWS' Northern Virginia, Northern California, Ohio, Oregon, Ireland and U.S. GovCloud regions, the G3 joins the P2 and F1 families in AWS' lineup of instances designed for "accelerated computing" purposes.
The G3 is specifically optimized for graphics-heavy, server-side applications, including 3-D rendering, virtual reality and visual encoding. It's ideal for organizations that perform a lot of "complex modeling and 3D visualization analyses, such as medical image processing, computer-aided design, and seismic visualization jobs," AWS said.
The new instances run on Intel Xeon E5-2686 v4 processors and use NVIDIA Tesla M60 GPUs. According to a blog post by AWS evangelist Jeff Barr, G3 users receive an "NVIDIA GRID Virtual Workstation License and can make use of the NVIDIA GRID driver without purchasing a license on your own."
Currently, the G3 comes in three sizes. The smallest, g3.4xlarge, features a single GPU, 16 virtual CPUs, 122GiB of host memory and 8GiB of GPU memory.
The g3.8xlarge has twice as much capacity, featuring two GPUs, 32 vCPUs, 244GiB of host memory and 16GiB of GPU memory.
The g3.16xlarge is the largest, with four GPUs, 64 vCPUs, 488GiB of host memory and 32GiB of GPU memory.
"AWS was the first to offer cloud instances customized for graphics-intensive applications. In 2010, we launched the CG1 instance type to provide a cost-effective, high-performance instance for graphics-heavy applications, and today, G3, our third generation GPU-powered instance, serves the most demanding graphics workloads such as 3D rendering and data visualization," said Matt Garman, vice president of EC2 at AWS, in a prepared statement.