Announcing New Intel Xeon Phi Processors and Intel Scalable System Framework (SSF) Updates
Meeting the demands of the machine learning and HPC industries with outstanding performance, scalability, programmability, and investment protection.

As data becomes increasingly complex and exponentially more abundant, the need for next-generation hardware, software, and architectures that can provide deep insights into machine learning and artificial intelligence analytics performance and accelerate new discoveries and business innovation is growing.

Intel has launched the new Intel® Xeon Phi™ processor as a solution to enable these deep insights.

The Intel Xeon Phi family, a key element of the Intel® Scalable System Framework (Intel SSF – Intel’s reference system architecture for HPC), combines key technologies to form high-performance clusters and provides a complete solution for easy deployment.

The new Intel Xeon Phi processor is Intel's first bootable host processor, specifically designed for highly parallel workloads, and the first to combine memory and fabric technologies. As a bootable x86 CPU, the Intel Xeon Phi processor can effectively scale without limitations, unlike GPU accelerators that must rely on the PCIe bus. By eliminating these dependencies, Intel Xeon Phi processors deliver superior scalability and the performance to manage a wider range of workloads and specifications than any other accelerator family.

Additionally, this product family combines 16GB of high-bandwidth memory, providing up to 500GB/s of sustained memory bandwidth for memory-intensive workloads2, and when combined with the dual-port Intel® Omni-Path Architecture (Intel OPA), can further reduce solution costs, power consumption, and footprint. As a general-purpose CPU developed based on open standards, the Intel Xeon Phi processor enables future-proofing of existing software investments.


Intel Xeon Phi processors deliver exceptional efficiency and scalability to accelerate the most demanding computational tasks across a wide range of environments and applications, from machine learning to high-performance computing (HPC). 3 This family also enables new innovations using high-performance modeling and simulation, virtualization, and data analytics.

A wide range of vendors today unveiled systems featuring the new Xeon Phi processor family, with products expected to be available around September of this year. Intel has shipped tens of thousands of units to vendors so far and expects to sell more than 100,000 units this year. This processor will be supported by a broad ecosystem including over 50 OEMs, ISVs, and middleware partners.

Machine Learning Gets Even More Powerful with Intel Xeon Phi Processors

Implementing machine learning requires extremely powerful computing power to run computational algorithms and process massive amounts of data. To meet this need, Intel has launched the Intel Xeon Phi processor family, further expanding the range of technologies available for machine learning.

The Intel Xeon Phi processor delivers powerful performance for training machine learning models and, as a bootable host processor, offers the flexibility to handle multiple analytics workloads. Additionally, clusters based on Intel SSF, powered by Intel Xeon Phi processors, and the combined Intel OPA, enable data scientists to run complex neural networks and train models in a fraction of the time. The Intel Xeon Phi family delivers up to 1.38x better scalability than GPUs on a 32-node infrastructure, and up to 50x faster model training on a 128-node infrastructure.3

The Intel Xeon Phi family can also deliver enhanced performance when combined with the Intel Xeon processor E5 family, the most widely adopted infrastructure for machine learning. The Intel Xeon processor E5 v4 family is optimized for machine learning scoring models and delivers outstanding performance and value for a wide range of data center workloads. The combination of Intel Xeon Phi and the Intel Xeon processor E5 v4 family will give developers a consistent programming model for training and scoring, and a common architecture for high-performance computing, data analytics, and machine learning workloads.