AWS launches elastic GPUs for EC2, FPGA-backed F1 instances, R4 and refreshed T2, C5 and I3 coming in Q1

At its fifth annual re:Invent user conference in Las Vegas today, public cloud market leader Amazon Web Services (AWS) announced the launch of new virtual machine (VM) instances that will be available for developers to rent out by the hour to run their applications.

This is the most predictable announcement of the 2016 re:Invent conference, as AWS tends to unveil new VM types at this event, in sync with Intel’s rhythm of releases of new chips.

Of course, the other big public clouds, Microsoft Azure and the Google Cloud Platform, also regularly introduce new VM types for customers. Prices vary and all three cut VM prices from time to time, with Azure continuing to maintain a commitment to match AWS’ prices.

Rather than just releasing new VM instances backed with graphics processing units (GPUs), AWS chief executive Andy Jassy said that AWS will be exposing “elastic GPUs for EC2,” a way for people to attach GPU resources to their existing VM instances. This will be useful in cases when people need “just a little GPU” rather than an entire GPU, Jassy said.

Also, there are new F1-branded VM instances that are accelerated with field-programmable gate arrays (FPGAs). This announcement comes after Microsoft said it was accelerating Azure instances with FPGAs. The AWS service lets you write code, package it up as an image, and then run it as custom logic on the FPGAs.

Additionally there are several new regular VM instance types on the way.

First up is the new R4 instance family, with DDR4 RAM and twice the RAM of the high end of the existing R3 instance family. There’s also a larger L3 cache and twice as many vCPUs, Jassy said.

The T2 family for smaller general-purpose workloads is being refreshed, too. The T2.xlarge has 16GiB and 2 vCPUs, while the T2.2xlarge has 32GiB and 2vCPUs.

Coming in Q1 are two other new instance families: I3 and C5.

The I3 family of instances for IO-heavy workloads will offer twice as much RAM and vCPU and more than nine times the input/output operations per second (IOPS) of the top I2 instances, Jassy said.

Finally, the new C5 instance family for compute-intensive applications will run on seventh-generation Intel Skylake Xeon chips.