HPC Cloud: Accelerate Time to Discovery

Cloud-Based HPC and Intel:

  • Intel delivers optimized hardware and software capabilities at the cloud service provider (CSP) level that help ensure you get exceptional performance per cost.

  • Intel works closely with AWS, Google Cloud, Microsoft Azure, and Oracle to integrate the latest technologies into their HPC cloud offerings.

  • The same tools that developers use on-premises—like the Intel® C++ Compiler, Intel® MPI Library, and Intel® Cluster Checker—can be used in the cloud, thereby helping customers who want portability between the cloud and on-premises.

  • The Intel® HPC Platform Specification for applications and cloud services like AWS ParallelCluster help customers get the best performance on Intel-based cloud infrastructure.



Running HPC Workloads in the Cloud

HPC cloud computing has come a long way in terms of performance level and availability. Organizations that need to run HPC workloads can leverage cloud services to take on their most-complex and -challenging compute and storage-intensive requirements. HPC in the cloud can also improve time to results, so researchers spend their time researching, not waiting for the job queue. Many of the Intel® technologies that enable on-premises HPC deployments can also be found in the HPC cloud. Intel works closely with CSPs, using key toolkits and hardware-based security and acceleration to optimize HPC workloads on Intel® architecture.

Cloud HPC Architecture Considerations

In the cloud, customers pay by the hour and prioritize their budget based on time to results. Architecture plays a key role in delivering the performance necessary to crunch HPC workloads within desired time frames, ultimately contributing to the customer’s bottom line. Intel leads the way in working with CSPs to deploy performant architecture where it’s most impactful for HPC cloud instances. Key frameworks and technologies include the following:

  • Intel® Xeon® Scalable processors are the beating heart of HPC cloud servers, delivering the performance and memory capacity for the most compute-intensive workloads. The Intel® Xeon® Scalable processor platform also supports several key technologies enumerated below that enable specific HPC use cases including AI convergence. 3rd Generation Intel® Xeon® Scalable processors, soon to be deployed in major CSP HPC cloud offerings, will deliver more memory bandwidth1 and higher instructions per clock2 over the previous generation processor. Intel® Speed Select Technology on Intel® Xeon® Scalable processors allows for multiple configurations in one server to meet the needs of diverse workloads. These enhancements can help customers achieve better cost per performance and time to results.
  • Intel® Software Guard Extensions (Intel® SGX) on select Intel® Xeon® Scalable processors is an extremely desirable feature for running and helping secure HPC workloads in the cloud. While many security technologies focus on protecting data at rest, Intel® SGX helps protect data during the critical moment when it is being processed. In multitenant HPC cloud environments, Intel® SGX also helps protect containers and VMs by using memory enclaves to isolate cryptographic keys and data in transit.
  • Intel® Advanced Vector Extensions 512 (Intel® AVX-512) is a set of processor instructions that boost vector-intensive compute workload performance. Intel® AVX-512 is especially well suited to vector/matrix operations on large data sets and is a competitive differentiator for Intel® Xeon® Scalable processors. Researchers and data scientists can use Intel® AVX-512 to help increase the performance for AI/DL workloads, DNA sequencing, simulations, financial analytics, and 3D modeling.
  • Intel® Deep Learning Boost (Intel® DL Boost) in Intel® Xeon® Scalable processors includes a new set of Vector Neural Network Instructions (VNNI) that extends Intel® AVX-512. VNNI can also help reduce the number and complexity of convolutional operations required for AI inference, resulting in lower power and memory requirements for HPC cloud systems.3 Intel® DL Boost can accelerate convolutional neural network loops and increase AI operations in HPC cloud instances, with 3.4x greater performance.3
  • Intel® oneAPI is a unified programming model specifically designed for heterogeneous HPC infrastructure. This model includes key performance libraries such as the Intel® Distribution for Python and Intel® MKL, which help optimize and accelerate HPC workloads on Intel architecture. Intel® MPI is a differentiated offering available in many CSP marketplaces that enables developers to easily deploy complex applications to multiple clusters, optimize code for high performance, and use automatic tuning to achieve low latency and high bandwidth. Customers and CSPs use these frameworks to help ensure they achieve the most impact from their HPC investments.
  • Intel® HPC Platform Specification is a set of minimum requirements for compute, memory, storage, and fabric, as well as compatible applications for HPC infrastructure. Customers and businesses can rely on this specification for assurance that HPC cloud service provider offerings will satisfy a high standard of quality for their HPC workloads.

Intel-Based HPC Cloud Service Providers

Intel works closely with leading HPC cloud service providers, including AWS, Google Cloud Platform, Microsoft Azure, and Oracle. Each CSP offers their own cloud instances with a strong foundation of Intel® Xeon® Scalable processors, which are optimized for Intel® MPI and offer built-in Intel® DL Boost. Additionally, each CSP has its own marketplace of Intel and third-party solutions to help businesses get started quickly on Intel-based HPC instances.

Simplifying Intel-Based CSP Onboarding

As businesses consider using Intel-enabled CSPs, choosing the correct HPC cloud offerings may be a complex and daunting endeavor. Fortunately, third-party cloud service integration partners are available that can help businesses choose the best-fit offerings and streamline the onboarding process. These partners are typically smaller organizations that assist with setting up workloads, enabling nonstandard features, offering unique insights via rich UI dashboards, or even discovering ways to replicate workload processing in a manner consistent with on-premises usage models. Key examples of these types of technology partners include RescaleRONIN, Six Nines, and OnScale. HPC in the cloud offers immense choice and variety in service offerings, but that choice can cut both ways and increase the difficulty in selecting the right solution. Partners help by offering guidance, expertise, and specialization.

Leading the Way for HPC Cloud

Many IT decision-makers are aware of the role that Intel fills in providing expertise for designing on-premises HPC architecture. However, Intel can fill the same role as a trusted adviser for HPC in the cloud. Any organization looking for a point of entry can start with Intel and benefit from a global ecosystem of CSPs and technology partners.