Amazon recently let slip the news that its new Amazon EC2 C4 instances, based on Intel’s Haswell processor, would soon be available—only to kill the post shortly afterward. Although the post has disappeared from aws.amazon.com, some minor detective work revealed a remarkably similar post by AWS Chief Evangelist Jeff Barr. “Private: Now Available—New C4 Instances” is still available in full at http://reinvent.kinvey.com/h/i/39267818-private-now-available-new-c4-instances, dated December 31, 2014. In the post, Barr explains:
The new C4 instances are based on the Intel Xeon E5-2666 v3 (code name Haswell) processor. This custom processor, optimized for EC2, runs at a base speed of 2.9 GHz, and can achieve clock speeds as high as 3.5 GHz with Turbo Boost. These instances are designed to deliver the highest level of processor performance on EC2.
Chip watchers will know that the Xeon E5-2666 v3 designation is not on the standard Intel processor sheet. Instead, Intel has built a custom processor designed to meet Amazon’s requirements and optimized for EC2 technology. The 2.9 GHz processor is based on the chipmaker’s Haswell architecture. Built using the company’s latest 22 nanometer process, it splits the difference between the off-the-shelf 2.6 GHz E5-2660 and the 3.2 GHz E5-2667. This isn’t the first time Intel has customized a processor for a big customer. In 2013, Intel designed fifteen custom CPUs for customers including Facebook and eBay. In 2014, it developed custom chips for more customers, including Oracle, for use in its big Exadata X4-8 Database Machines, which debuted in July. Intel has also been working on hybrid technologies that combine the Xeon E5 processor with a field-programmable gate array (FPGA). Intel and Amazon are not saying whether the E5-2666 v3 is using FPGAs or not, although it is unlikely that general-purpose EC2 machines like this would do so.
The complete lineup of C4 instances is as follows:
Instance Name | vCPU Count | RAM | Network Performance | Dedicated EBS Throughput | Linux On-Demand Price |
c4.large | 2 | 3.75 GiB | Moderate | 500 Mbps | $0.116/hour |
c4.xlarge | 4 | 7.5 GiB | Moderate | 750 Mbps | $0.231/hour |
c4.2xlarge | 8 | 15 GiB | High | 1,000 Mbps | $0.462/hour |
c4.4xlarge | 16 | 30 GiB | High | 2,000 Mbps | $0.924/hour |
c4.8xlarge | 36 | 60 GiB | 10 Gbps | 4,000 Mbps | $1.848/hour |
C4 instances are available as On-Demand Instances, or for purchase as Reserved Instances or via the Spot market.
Pricing is up 10% across the board compared to current-generation C3 instances. However, all things being equal, existing code should run 30 to 50% faster on the new Haswell architecture chips than on previous-generation hardware, making the transition to C4 more-or-less equivalent to a 17% to 26% drop in cost. And with the c4.8xlarge instance’s sporting thirty-six vCPUs (assuming that the chart is correct)—up from thirty-two in the current generation’s c3.8xlarge—even more gains may be had. New applications that can take advantage of Intel’s Advanced Vector Extensions 2 (AVX2) instruction set can double floating-point performance on some workloads, which will result in further savings.
There is no official news as to when the Amazon EC2 C4 instances will come on-stream, but expect to see them any time in the next week or so.