AWS Launches Elastic GPUs for EC2 Instances
After a nine-month preview period, Amazon Web Services' new bolt-on GPU capability is now generally available.
Amazon EC2 Elastic GPUs, which AWS first announced at its re:Invent conference last November, let AWS customers add incremental amounts of GPU power to their existing EC2 instances for a temporary boost in graphics performance.
They're ideal for workloads such as gaming, high-performance computing visualizations, virtual desktops and industrial design, which sometimes require small hikes in GPU power beyond what a user's set EC2 instance can support.
"Elastic GPUs are not part of the hardware of your instance," explained AWS evangelist Jeff Barr in a blog post Tuesday. "Instead they're attached through an elastic GPU network interface in your subnet which is created when you launch an instance with an Elastic GPU."
Elastic GPUs come in four sizes -- 1GiB, 2GiB, 4GiB and 8GiB -- with prices ranging from $0.05 per hour to $0.40 per hour. They provide an affordable option for users who don't require the full power of AWS' GPU-optimized instances, the P2 and the G3.
"If we attach [the smallest] Elastic GPU to a t2.medium ($0.065/hour) we pay a total of less than 12 cents per hour for an instance with a GPU. Previously, the cheapest graphical workstation (G2/3 class) cost 76 cents per hour. That's over an 80% reduction in the price for running certain graphical workloads," Barr said.
Elastic GPUs are currently available out of AWS' Virginia and Ohio regions. They support OpenGL 3.3 API standards and earlier, though the company plans to expand support in the future.