Services offer flexible GPU acceleration, simple virtual private servers and improved compute.
Amazon Web Services continued to push its infrastructure offering forward on Wednesday with the launch of upgrades to its existing instance types and new tools for simplifying and accelerating computation tasks in its public cloud.
AWS CEO Andy Jassy unveiled a new Elastic GPUs feature for the Elastic Compute Cloud (EC2) that will let developers add flexible acceleration using parallel processing for their applications.
On top of that, Amazon launched a new Lightsail service that’s aimed at giving customers an easy way to spin up a straightforward virtual private server without having to orchestrate a fleet of different AWS services. Finally, the company also unveiled upgrades to its burstable, memory intensive, high I/O, and compute intensive instances.
All of these announcements are aimed at helping Amazon maintain its leading position in the infrastructure-as-a-service market. The company has positioned its IaaS offering as foundational to its cloud platform.
The Elastic GPU service will let developers attach GPUs to different EC2 instances, so they can get the graphical acceleration they need without having to pay for specific, GPU-accelerated instances. When users first spin up an instance, they’ll be able to attach up to 8GiB of GPU vRAM. If they need to change the amount of vRAM available, developers will be able to shut down, modify and then spin up the instance again.
It’s a service that will be useful for a variety of applications, like gaming, graphical rendering and machine learning. The GPUs will be available for all of Amazon’s instance types, once they launch officially in the future. Right now, the amount of vRAM that an instance needs determines which of Amazon’s existing GPU-attached instances customers have to use.
“Developers will love having slices of GPU, instead of paying for it all the time,” IDC Program Director Al Hilwa said in an email.
It’s similar to functionality that Google announced earlier this month, which will let companies attach GPUs to any of the compute instances running in its cloud. Amazon’s OpenGL library will automatically detect and use Elastic GPUs, which will be a boon to developers.
A pricing chart shows the different pricing available for Amazon’s new Lightsail VPS service.
Lightsail will allow people to easily spin up a virtual private server on AWS without having to understand how to start up a compute instance, connect it to storage, handle networking and more. Instead, users make a few clicks to select what they want the server to do, pick out its hardware capabilities and give it a name.
Lightsail servers start at $5 per month, which gets users 512MB of memory, 1 vCPU, 20GB of SSD storage, at 1TB of data transfer. Users can go all the way up to $80 per month, with 8 GB of memory, 2 vCPUs, 80GB of SSD, and 5TB of data transfer.
Jassy said users that want to then take advantage of additional AWS services will be able to go “under the hood” with a Lightsail instance and connect with other AWS services.
On top of all that, Amazon also announced refreshes to a host of different instance types that it already has available. New R4 instances are an upgrade to Amazon’s memory-focused R3 instances. They sport up to 488GiB of DDR4 memory and up to 64 vCPUs. That’s twice the memory available with the previous instances, and twice the number of virtual processors.
Amazon’s burstable T2 instances got new t2.xlarge and t2.2xlarge types, which provide twice and four times the number of virtual CPUs and memory compared to the t2.large instance type, respectively.
The high I/O I3 instance type, which will be available in the future, will support configurations with up to 488GiB of memory, 3.3 million IOPS, 15.2TB of NVMe SSD storage and 64 vCPUs. On top of all that, the forthcoming compute-intensive C5 instances will support up to 72 vCPUs and 144GiB of memory.
All of these infrastructure revisions are part of Amazon’s announcements at its Re:Invent conference. The company also announced a new instance type that lets users access field-programmable gate arrays to help accelerate particular applications.