White PaperHigh Performance Computing

Business BriefHigh Performance Computing

Interest in Artificial intelligence (AI) is rising fast, as applications in speech recognition, image analysis and other complex pattern recognition tasks have now surpassed human accuracy. Practical solutions are emerging across a wide range of use cases, including energy, healthcare, physics, finance, fraud detection, and many others. A recent report by Goldstein Research estimates that the AI industry will be worth USD 14.2 billion by 2023, up from USD 525 million in 20151.

AI offers enormous potential across many industries, yet training an AI model is a compute- and data-intensive task that can require the performance and scale of a high performance computing (HPC) system. A balanced system design is important, since AI training workloads stress every aspect of an HPC cluster, from memory-bandwidth and capacity, to message latency and network bandwidth.

Intel® Scalable System Framework (Intel® SSF) is designed to help organizations address these workload requirements at lower cost and with less effort. It provides a blueprint for balanced, high-performing HPC clusters that are easier to deploy, easier to use, and can scale almost without limit to accommodate AI and other HPC workloads. In combination with the Intel® AI technology portfolio, this scalable HPC framework provides end-to-end hardware and software support for developing, deploying, and growing AI solutions, and for integrating them with other business and technical applications.

Scaling AI for Virtually Unlimited GrowthThe Intel® Scalable Systems Framework (Intel® SSF)

Intel SSF offers a simpler and more flexible HPC platform to address the demands of AI training workloads. Intel SSF is designed to eliminate bottlenecks and improve utilization by providing balanced high performance at every layer of the solution stack—compute, memory, storage, fabric, and software.

This holistic, system-level approach simplifies the design of optimized clusters and helps organizations take advantage of disruptive new technologies with less effort and lower risk. Importantly, it also provides flexible support for the full range of HPC workloads, so integrating AI with other business and technical applications is simpler, cost models are improved, and there is reduced need to move large data sets from one specialized system to another.

The most important elements of Intel SSF are discussed below.

Powerful Processors and AcceleratorsIntel SSF supports a wide range of Intel processor and accelerator options, so organizations can configure a cluster to address specific requirements. They can also upgrade and expand their system with new and alternative compute options without having to rearchitect the cluster, a major advantage in the complex world of HPC.

with Intel® Scalable System Framework and the Intel® AI Technology Portfolio

Ramp Up Your AI Revolution

“ Intel® Scalable System Framework provides flexible support for the full range of HPC workloads, so integrating AI with other business and technical applications is simpler, cost models are improved, and there is reduced need to move large data sets from one specialized system to another.”

Figure 2. Intel® Scalable System Framework simplifies the design of efficient, high-performing clusters that optimize the value of HPC investments.

Business Brief | Ramp Up Your AI Revolution 2

Choosing the right processors is important. Current options include:

• Intel® Xeon® Scalable processors for inference engines and for many training workloads. With more cores and more memory bandwidth than previous-generation Intel® Xeon® processors, plus ultra-wide 512-bit vector support, the new Intel Xeon Scalable processors provide up to 2.2x higher neural network training performance than their predecessors2. They are also available with integrated fabric controllers. These processors are ideal for deploying AI inference engines at scale and, in many cases, for tackling the heavier demands of neural network training. With their broad interoperability, they also provide an agile foundation for integrating AI solutions with other business and technical applications.

• Intel® Xeon Phi™ processors for training large and complex neural networks. With up to 72 cores, 288 threads, and 512-bit vector support—plus integrated high bandwidth memory—these processors offer performance and scalability advantages for neural network training and for many other highly-parallel workloads. Since they function as host processors and run standard x86 code, they simplify implementation and eliminate the inherent latencies of PCIe-connected accelerator cards.

• Optional accelerators for neural network training and inference. Intel offers a range of workload-specific accelerators, including programmable Intel® FPGAs for flexible and scalable acceleration of high-volume inference solutions. These optional add-ons for Intel Xeon processor-based servers open new doors for innovation, performance, and energy efficiency across a wide range of workloads and configurations.

In addition to these offerings, Intel recently announced the Intel® Nervana™ Neural Network Processor, which is flexible enough to handle deep learning workloads and scalable enough to handle high-intensity computation requirements. High-speed memory and powerful interconnects will be built into each chip to deliver extreme performance that can be scaled across multiple chips and multiple chassis without performance loss.

Efficient, Scalable ClustersAs Intel processor performance has increased over the years, memory and storage technologies have lagged far

behind, leading to data access bottlenecks that throttle performance for many applications. Intel SSF brings together multiple technologies to help resolve these bottlenecks and to simplify the design of flexible, scalable HPC clusters. This system-level integration helps to ensure that clusters based on Intel SSF can efficiently support the full range of AI workloads, from big data preprocessing to scalable training and fast inference as well as traditional workloads like simulation and modeling.

• A breakthrough in storage capacity and performance. Intel® Optane™ Solid State Drives (SSDs), built with 3D XPoint™ memory media, offer a major leap forward in storage performance. Intel Optane SSDs are designed to deliver 5-8x higher performance than NAND-based SSDs3. They can be used as ultra-fast storage, storage cache, or extended memory. In combination with Intel® 3D NAND SSDs and traditional hard drives, they make it easier to address the extreme data access requirements of AI and other HPC applications, while containing overall costs.

• Massive scalability for big data. The open source Lustre* file system provides a next-generation, software defined storage (SDS) solution that addresses the key challenges of large neural networks and other data-intensive HPC workloads. By scaling metadata on separate servers and striping object data across multiple storage drives and servers, Lustre enables extreme performance at almost any scale.

• A fast, affordable fabric. In a large, distributed neural network, data must move almost as efficiently between nodes as it does within nodes. Intel® Omni-Path Architecture (Intel® OPA) offers the same 100 Gbps link speeds as today’s fastest InfiniBand* fabrics, and provides higher port densities per chip for superior scalability and cost models4. Intel OPA uses streamlined software that helps to maintain low latency even at extreme scale, so high performance can be sustained under heavier loads. Additional cost, density, and performance advantages can be achieved by using Intel Xeon and Intel Xeon Phi processors with integrated Intel OPA controllers.

• Complete, validated system software. Maintaining a stable and optimized HPC software environment can be a complex and time-consuming task, even for HPC experts. OpenHPC, a collaborative community effort, is working to aggregate the common ingredients required to manage HPC Linux clusters. Intel contributes to the OpenHPC software project to enable the entire ecosystem with up-to-date, open source software ingredients that simplify deploying and operating HPC clusters.

Democratizing AI through End-to-End SupportThe Intel® AI Technology Portfolio

The Intel AI technology portfolio provides comprehensive resources to help organizations develop and deploy AI solutions, integrate them with existing applications, and manage them in production environments.

Business Brief | Ramp Up Your AI Revolution 3

Hardware options extend all the way from the cloud and data center to the edge of the IoT. Software options include a broad set of open source and proprietary components, including development tools, frameworks, libraries, and middleware. These components are highly optimized for Intel® architecture. They can help organizations compress development cycles and boost performance, in some cases by orders of magnitude5.

A Launching Pad for AI Innovation The next wave of AI innovation will require simpler development tools and enormous new computing capability. Intel SSF and the Intel AI technology portfolio provide comprehensive support and a unified platform to help organizations of every size develop and scale AI applications.

As innovation heats up, the advantages will grow. Intel is working with academic, open source, and industry leaders to extend AI solutions and usage models. These and many other advances will ultimately be reflected in the Intel AI technology portfolio and will help to provide an increasingly powerful and flexible platform for AI. Meanwhile, Intel SSF will help AI innovators scale their neural networks and other back-end solutions more affordably and on a flexible hardware architecture that can efficiently support the widest range of business and technical workloads.

These foundational resources will help organizations ride the wave of escalating performance and functionality, so they can focus on driving deeper and more useful intelligence into everything they do.

Learn MoreFor more information about Intel technologies and solutions for HPC and AI, visit intel.com/HPC.

To learn more about Intel AI solutions, go to intel.com/AI.

Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to http://www.intel.com/performance/datacenter.No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document. You may not use or facilitate the use of this document in connection with any infringement or other legal analysis concerning Intel products described herein. You agree to grant Intel a non-exclusive, royalty-free license to any patent claim thereafter drafted which includes subject matter disclosed herein.Normalized performance is calculated by assigning a baseline value of 1.0 to one benchmark result, and then dividing the actual benchmark result for the baseline platform into each of the specific benchmark results of each of the other platforms, and assigning them a relative performance number that correlates with the performance improvements reported.Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software or service activation. Performance varies depending on system configuration. No computer system can be abso-lutely secure. Check with your system manufacturer or retailer or learn more at www.intel.com.The products described may contain design defects or errors known as errata which may cause the product to deviate from published specifications. Current characterized errata are available on request.Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of perfor-mance, course of dealing, or usage in trade.Copyright © 2017 Intel Corporation. All rights reserved. Intel, the Intel logo, Xeon, and Xeon Phi are trademarks of Intel Corporation and its subsidiaries in the U.S. and other countries.*Other names may be trademarks of their respective owners.

102017/RJMJ/RJ/PDF Please Recycle 336609-001US

Figure 3. The Intel® Artificial Intelligence portfolio offers end-to-end hardware and software support for developing, deploying, and managing AI solutions.

1 Global Artificial Intelligence (AI) Market Outlook 2024: Global Opportunity and Demand Analysis, Market Forecast, 2016-2024, published July 28, 2017. https://www.goldsteinresearch.com/report/global-artificial-intelligence-ai-market-outlook-2024-global-opportunity-and-demand-analysis-market-forecast-2016-2024

2 For details, see https://www.intel.com/content/www/us/en/processors/xeon/scalable/xeon-scalable-platform.html.

3 Common Configuration - Intel® 2U Server System, OS CentOS* 7.2, kernel 3.10.0-327.el7.x86_64, CPU 2 x Intel® Xeon® E5-2699 v4 @ 2.20GHz (22 cores), RAM 396GB DDR @ 2133MHz. Configuration – Intel® Optane™ SSD DC P4800X 375GB and Intel® SSD DC P3700 1600 GB. Performance ―measured under 4K 70-30 workload at QD1-16 using FIO 2.15.

4 Switches based on Intel® Omni-Path Architecture use a 48-port switch chip versus the 36-port chips used in today’s InfiniBand* switches. Clusters can potentially be built using fewer switches and fewer switch hops, which helps to optimize cost, scalability, and performance.

5 For example, see “Up To Orders of Magnitude More Performance with Intel’s Distribution of Python,” by Rob Farber, August 17, 2016 at http://www.techenablement.com/orders-magnitude-performance-intel-distribution-python/ and “Bridging Advanced Analytics and Artificial Intelligence with Bid,” by Jason Dai, March 14, 2017 at https://itpeernetwork.intel.com/advanced-analytics-artificial-intelligence-bigdl/