Almost by definition, cloud services are about delivering applications at scale. And given that scale, it makes sense that cloud service providers are focused on high degrees of datacenter optimization to deliver the best performance for the application per TCO (total cost of ownership). Our team’s mission has been to optimize technology for cloud service providers (CSPs) ranging from the CPUs to the systems they design to the software they run and the facilities they use.
We have a long history of working closely with AWS on optimization since the original EC2 instances. We have delivered high performance custom Intel Xeon E5-v3 for AWS C4 instances that provided over 3x performance improvement over the previous generation. We collaborated on custom solutions for the M4 instances announced earlier this year to deliver a greater higher performance/TCO through a highly optimized solution. This trend has defied the conventional wisdom that cloud computing is just about “commodities”.
Today this collaboration reached a new level of performance as AWS released their new X1 instances based on Intel “Haswell-EX” Xeon E7 v3 processors. With the ability to exceed over 100 virtual CPUs and 2TB of memory, the X1 is the first major instance of an IaaS designed around a four socket system. This processor is capable of running highly complex workloads to deliver insights in real time and is well suited for intensive analytic workloads.
At Intel we have seen a trend that with each new generation of Intel processors over 80% of the CPU volume to the top cloud service providers migrated higher performing SKU’s beyond the generational improvement. Top CSPs also move sooner to our next generation technology to gain competitive advantage.
This is only logical since to leverage all the investment in the cloud infrastructure it makes sense to get more performance and capacity from the system. Hyperscale CSPs have become very good at this and that’s what leads Intel to customization and even to include customer IP in our products. In fact many of the top CSPs are driving some form of custom Si and all buying ahead of market launch in order to reap the benefits of the improved performance as soon as they possibly can.
The need for performance and optimization in the cloud segment is continuing to accelerate and the Amazon X1 instance is the latest major step forward. To support the industry need for performance, we are looking forward to delivering a steady stream of innovation. As an example Intel announced its 3DxPoint memory technology which will enable dramatically larger memory footprints. You can expect that this will be complemented by FPGAs, high speed fabrics, custom Si, and several other innovations.