Blogging for Amazon Web Services, Chief Evangelist Jeff Barr announced the public debut of “Haswell”-powered C4 instances, further boosting the HPC horsepower of the EC2 engine. First revealed at the cloud innovator’s re:Invent developer conference in November, the compute-optimized offering is now available in seven AWS regions with more to be added. The new instance types can be purchased On-Demand, as Reserved Instances, or via the Spot Market. They will also be available in the AWS Marketplace.
As we learned in November, the C4 instances are based on a semi-custom version of the latest generation Intel Xeon “Haswell” processors, called E5-2666 v3. Optimized for the massive scale of EC2, the exclusive chips operate with a base speed of 2.9 GHz and can reach clock speeds as high as 3.5 GHz “under optimal conditions” thanks to Intel Turbo Boost.
The new instances provide users with the highest number of virtual cores yet: the c4.xlarge instance packs 36 virtual cores, compared to 32 on offer from the c3.xlarge. Barr offers the caveat that some operating systems do not support more than 32 vCPUs and thus may not be compatible with the largest C4 instance type.
The C4 lineup that was previewed at re:Invent remains unchanged, except for the addition of pricing information and a few more technical details.
The above pricing scheme is applicable to the US East (Northern Virginia) and US West (Oregon) regions. The new instances are also available to Europe (Ireland), Asia Pacific (Tokyo), US West (Northern California), Asia Pacific (Singapore), and Asia Pacific (Sydney) regions.
Barr explains that EBS Optimization adds 500 Mbps to 4,000 Mbps of additional dedicated throughput to Amazon Elastic Block Store (EBS) I/O in order to minimize contention. For C4 users, this feature is enabled by default, at no additional charge. Barr also calls out “Enhanced Networking,” available on both C3 and C4 instances, where it works to increase packet per second (PPS) performance, lower network jitter, and lower network latency. Users also have the option of putting two or more C4 instances within a placement group to support low-latency connectivity.
The following chart shows the current generation of compute-optimized instances with pricing for the US East region.