A New Generation of EC2 Instances for Compute-Intensive Workloads

Many AWS customers run CPU-bound, compute-intensive workloads on Amazon EC2, often using parallel processing frameworks such as Hadoop to distribute work and collect results. This includes batch data processing, analytics, high-performance scientific computing, 3D rendering, engineering, and simulation.

To date these needs have been met by the existing members of our compute-optimized instance families — the C1 and CC2 instance types. When compared to EC2’s general purpose instance types, the instances in this family have a higher ratio of compute power to memory.

Hello C3
Today we are introducing the C3 family of droids instances. Compared to C1 instances, the C3 instances provide faster processors, approximately double the memory per vCPU and SSD-based instance storage.

As the newest member of our lineup of compute-optimized instances, the C3’s were designed to deliver high performance at an economical price. The C3 instances feature per-core performance that bests that provided by any of the other EC2 instance types, at a price-performance ratio that will make them a great fit for many compute-intensive workloads.

Use the Cores
Each virtual core (vCPU) on a C3 instance type is a hardware Hyper-Thread on a 2.8 GHz Intel Xeon E5-2680v2 (Ivy Bridge) processor. There are five members of the C3 family:

Prices are for Linux instances in US East (Northern Virginia).

Protocols
If you launch C3 instances inside of a Virtual Private Cloud and you use an HVM AMI with the proper driver installed, you will also get the benefit of EC2’s new enhanced networking. You will see significantly higher performance (in terms of packets per second), much lower latency, and lower jitter.

Update: Read the documentation on Enabling Enhanced Networking on Linux Instances in a VPC to learn how to do this.

Getting Technical
As you may have noticed, we are specifying the underlying processor type for new instance types. Armed with this information, you can choose to make use of specialized instructions or to tune your application to exploit other characteristics (e.g. cache behavior) of the actual processor. For example, the processor in the C3 instances supports Intel’s AVX (Advanced Vector Extensions) for efficient processing of vector-oriented data in 256-bit chunks.

Some Numbers
In order to measure the real-world performance of the new C3 instances, we launched a 26,496 core cluster and evaluated it against the most recent Top500 scores. This cluster delivered an Rmax of 484.18 teraflops and would land at position 56 in the June 2013 list. Notably, this is over twice the performance of the last cluster that we submitted to Top500. We also built an 8,192 cluster, which delivered an Rmax of 163.9, putting it at position 210 on the Top500 list.

Launch One Now
The C3 instances are available today in the US East (Northern Virginia), US West (Oregon), EU (Ireland), Asia Pacific (Singapore), Asia Pacific (Tokyo), and Asia Pacific (Sydney) Regions. You can choose to launch C3 instances as On-Demand, Reserved Instances, or Spot Instances.

— Jeff;