Fact Sheet: Unleashing the Power of Artificial Intelligence in Healthcare with Intel Technologies

Artificial Intelligence: Machine Learning, Deep Learning and More

Artificial intelligence (AI) is now all around us, as machines increasingly gain the ability to sense, learn, reason, act and adapt in the real world. Much of the AI spotlight is focused on deep learning, a branch of machine learning that uses neural networks to comprehend complex and unstructured data (training) and use that understanding to classify, recognize and process new inputs (inference). Deep learning is delivering breakthroughs in areas like image recognition, speech recognition, natural language processing and other complex tasks.

Processing Deep Learning Workloads in Healthcare

Deep learning workloads require a tremendous amount of computational power to run complex mathematical algorithms and process huge amounts of data. While GPUs have been used for some deep learning training applications, they don’t necessarily have a performance edge for deep learning inference. In healthcare in particular, there are many examples of applications that run better on CPUs than GPUs.

That’s why for healthcare organizations running AI workloads, Intel® Xeon® Scalable processors provide an ideal computational foundation. Intel Xeon Scalable processors are optimized for AI, scale up quickly and seamlessly for 2.1x faster deep learning performance over previous generations1, and offer server-class reliability and workload flexibility1.

Versatility

Many organizations that are implementing AI don’t need to train models 24/7. Those that do need to train models around the clock require a dedicated accelerator, but the majority of organizations simply need to train and then deploy a model. If an organization invests thousands of dollars in building a dedicated acceleration stream, it can’t do anything else with that infrastructure beyond training – it will just sit idle.

Intel Xeon Scalable processors are more agile. They are designed to flex with business needs to support the workload needs of the moment, allowing organizations to leverage their data center infrastructure for AI applications and a wide range of other workloads.

Increased Utilization

Most organizations have at least 35 percent free utilization of processing capacity, if not more. This means they have could be obtaining better ROI from their infrastructure investments. With Intel Xeon Scalable processors, they can leverage this unused capacity to implement AI while still meeting the needs of other applications.

Technical Advantages

GPUs are often leveraged as an accelerator, but they have limitations when it comes to building models to tackle healthcare issues. These limitations begin with memory. GPUs are limited to only 12 to 16 gigabytes of memory on the chip itself, and that restricts the size and capabilities of the model that can be built. This memory limitation is a particular problem in healthcare AI work, which often involves building massive models – for example, to support a giant CT scan.

For today’s AI networks deployed on a standard GPU, the image size for a scan that can be developed is usually 256 x 256 pixels. At the most, that image could be 1000 x 1000 pixels, but that’s pushing it. If clinicians have a 4K image where they actually need all of those pixels to make an analysis — for example, to detect a tumor in a tiny portion of the image — then they have a problem. When they scale that 4K image down to 256 x 256 pixels, they lose all of the fine temporal resolution.

The same model can be built with Intel Xeon Scalable processors without limiting the size and resolution because the Intel Xeon Scalable family offers a terabit of on-load memory. Intel Xeon processors may process the image more slowly than a GPU, but when the resolution is needed for medical imaging, Intel Xeon processors deliver.

A Processor Designed for Deep Learning

Moving forward, Intel has also developed the Intel® Nervana™ Neural Network Processor (NNP), the world’s first processor specifically designed from the ground up for deep learning. The Intel Nervana NNP promises to further enhance medical imaging and other healthcare applications.

Using the Intel Nervana platform, healthcare organizations can develop entirely new classes of AI applications that maximize the amount of data processed and lead to greater insights. For example, AI will allow for earlier diagnosis and greater accuracy, helping make the impossible possible by advancing research on cancer, Parkinson’s disease and other brain disorders2.

The Intel Advantage for AI

Intel has 50 years of experience in helping its customers make their data valuable. Today, Intel is using that experience to help healthcare organizations address every aspect of their workflows through the successful implementation of AI.

From hardware to software and data science, Intel brings its full suite of products and expertise to AI. Intel has the technical expertise not only to help organizations build the right infrastructure – from data to storage and network – but also has the data scientists to understand and model data and the application developers to help make the data useful. In short, Intel’s expertise to implement AI goes well beyond the CPU.

Driving the Age of AI

At Intel, we recognize the age of AI is upon us, and we know our technologies will help drive the future of AI. We are also motivated by a desire to use our technology to help advance society and tackle the world’s big challenges, so we’re particularly excited about our work in healthcare. We want to help channel AI for societal good – we call it AI with a purpose.

With these thoughts in mind, we’re pleased to be engaged with healthcare providers and researchers. We know that our collective work is helping move the needle – with even more promise and unlimited possibilities for the future.

Benchmark results were obtained prior to implementation of recent software patches and firmware updates intended to address exploits referred to as “Spectre” and “Meltdown”.  Implementation of these updates may make these results inapplicable to your device or system.

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 complete information visit www.intel.com/benchmarks.

1INFERENCE using FP32 Batch Size Caffe GoogleNet v1 256  AlexNet 256.

Performance estimates were obtained prior to implementation of recent software patches and firmware updates intended to address exploits referred to as “Spectre” and “Meltdown.”  Implementation of these updates may make these results inapplicable to your device or system. 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 complete information visit http://www.intel.com/performance  Source: Intel measured as of June 2017 Optimization Notice: Intel’s compilers may or may not optimize to the same degree for non-Intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice.

Configurations for Inference throughput

Processor :2 socket Intel(R) Xeon(R) Platinum 8180 CPU @ 2.50GHz / 28 cores HT ON , Turbo ON Total Memory 376.46GB (12slots / 32 GB / 2666 MHz).CentOS Linux-7.3.1611-Core , SSD sda RS3WC080 HDD 744.1GB,sdb RS3WC080 HDD 1.5TB,sdc RS3WC080 HDD 5.5TB , Deep Learning Framework caffe version: f6d01efbe93f70726ea3796a4b89c612365a6341 Topology :googlenet_v1 BIOS:SE5C620.86B.00.01.0004.071220170215 MKLDNN: version: ae00102be506ed0fe2099c6557df2aa88ad57ec1 NoDataLayer. Measured: 1190 imgs/sec vs Platform: 2S Intel® Xeon® CPU E5-2699 v3 @ 2.30GHz (18 cores), HT enabled, turbo disabled, scaling governor set to “performance” via intel_pstate driver, 256GB DDR4-2133 ECC RAM. CentOS Linux release 7.3.1611 (Core), Linux kernel 3.10.0-514.el7.x86_64. OS drive: Seagate* Enterprise ST2000NX0253 2 TB 2.5″ Internal Hard Drive. Performance measured with: Environment variables: KMP_AFFINITY=’granularity=fine, compact,1,0‘, OMP_NUM_THREADS=36, CPU Freq set with cpupower frequency-set -d 2.3G -u 2.3G -g performance. Deep Learning Frameworks: Intel Caffe: (http://github.com/intel/caffe/), revision b0ef3236528a2c7d2988f249d347d5fdae831236. Inference measured with “caffe time –forward_only” command, training measured with “caffe time” command. For “ConvNet” topologies, dummy dataset was used. For other topologies, data was stored on local storage and cached in memory before training. Topology specs from https://github.com/intel/caffe/tree/master/models/intel_optimized_models (GoogLeNet, AlexNet, and ResNet-50), https://github.com/intel/caffe/tree/master/models/default_vgg_19 (VGG-19), and https://github.com/soumith/convnet-benchmarks/tree/master/caffe/imagenet_winners (ConvNet benchmarks; files were updated to use newer Caffe prototxt format but are functionally equivalent). GCC 4.8.5, MKLML version 2017.0.2.20170110. BVLC-Caffe: https://github.com/BVLC/caffe, Inference & Training measured with “caffe time” command.  For “ConvNet” topologies, dummy dataset was used. For other topologies, data was st ored on local storage and cached in memory before training  BVLC Caffe

Configuration for training throughput:

Processor :2 socket Intel(R) Xeon(R) Platinum 8180 CPU @ 2.50GHz / 28 cores HT ON , Turbo ON Total Memory 376.28GB (12slots / 32 GB / 2666 MHz).CentOS Linux-7.3.1611-Core , SSD sda RS3WC080 HDD 744.1GB,sdb RS3WC080 HDD 1.5TB,sdc RS3WC080 HDD 5.5TB , Deep Learning Framework caffe version: f6d01efbe93f70726ea3796a4b89c612365a6341 Topology :alexnet BIOS:SE5C620.86B.00.01.0009.101920170742 MKLDNN: version: ae00102be506ed0fe2099c6557df2aa88ad57ec1 NoDataLayer. Measured: 1023 imgs/sec vs Platform: 2S Intel® Xeon® CPU E5-2699 v3 @ 2.30GHz (18 cores), HT enabled, turbo disabled, scaling governor set to “performance” via intel_pstate driver, 256GB DDR4-2133 ECC RAM. CentOS Linux release 7.3.1611 (Core), Linux kernel 3.10.0-514.el7.x86_64. OS drive: Seagate* Enterprise ST2000NX0253 2 TB 2.5″ Internal Hard Drive. Performance measured with: Environment variables: KMP_AFFINITY=’granularity=fine, compact,1,0‘, OMP_NUM_THREADS=36, CPU Freq set with cpupower frequency-set -d 2.3G -u 2.3G -g performance. Deep Learning Frameworks: Intel Caffe: (http://github.com/intel/caffe/), revision b0ef3236528a2c7d2988f249d347d5fdae831236. Inference measured with “caffe time –forward_only” command, training measured with “caffe time” command. For “ConvNet” topologies, dummy dataset was used. For other topologies, data was stored on local storage and cached in memory before training. Topology specs from https://github.com/intel/caffe/tree/master/models/intel_optimized_models (GoogLeNet, AlexNet, and ResNet-50), https://github.com/intel/caffe/tree/master/models/default_vgg_19 (VGG-19), and https://github.com/soumith/convnet-benchmarks/tree/master/caffe/imagenet_winners (ConvNet benchmarks; files were updated to use newer Caffe prototxt format but are functionally equivalent). GCC 4.8.5, MKLML version 2017.0.2.20170110. BVLC-Caffe: https://github.com/BVLC/caffe, Inference & Training measured with “caffe time” command.  For “ConvNet” topologies, dummy dataset was used. For other topologies, data was st ored on local storage and cached in memory before training  BVLC Caffe (http://github.com/BVLC/caffe), revision 91b09280f5233cafc62954c98ce8bc4c204e7475 (commit date 5/14/2017). BLAS: atlas ver. 3.10.1.

2Intel editorial by Brian Krzanich, chief executive officer of Intel Corporation, “Announcing Industry’s First Neural Network Processor,” October 17, 2017.

Intel, the Intel logo, Xeon, and Nervana are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other countries.

Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software or service activation. Performance varies depending on system configuration. Check with your system manufacturer or retailer or learn more at www.intel.com.

About Intel
Intel (NASDAQ: INTC) expands the boundaries of technology to make the most amazing experiences possible. Information about Intel can be found at newsroom.intel.com and intel.com.

Intel and the Intel logo are trademarks of Intel Corporation in the United States and other countries.

*Other names and brands may be claimed as the property of others.