• <المزيد على Intel.com

Get Consistent Platform-Level Services for Well Bound Accelerators

للأسف فإن ملف PDF هذا متوافر فقط بهيئة قابلة للتنزيل

Get Consistent Platform-Level Services for Well Bound Accelerators

White Paper: Intel® QuickAssist Technology Accelerator Abstraction Layer Computationally-Intensive Applications

Enabling Consistent Platform-Level Services for Tightly Coupled Accelerators New software abstraction layer simplifies the use and deployment of accelerators

Executive Summary: Industries, such as financial services, energy, manufacturing and chemical production, are exploring the use of Field Programmable Gate Array (FPGA)-based accelerators for computation-intensive algorithms. FPGA accelerators can speed up domain specific algorithms by implementing these algorithms directly in hardware. Typically, accelerators are programmed via a “direct” API that essentially wraps a driver interface. The programming paradigm offered by a direct API is of a device tightly bound to an individual application.

However, this paradigm lacks the flexibility to easily accelerate functions across multiple users or abstract the accelerator interconnect specifics, which is needed to avoid software changes when accelerators with different busses are deployed. In today’s data center environment, where scarce resources are virtualized and accelerator technology advances at a rapid rate, a direct API doesn’t offer sufficient agility. Further, as the underlying hardware technology of the accelerator changes (e.g., different FPGAs, capacities, properties and attach technologies), necessary modifications inevitably “leak through” the API to the end user, requiring application software revisions.

Easing accelerator system integration, the Intel® QuickAssist Technology Accelerator Abstraction Layer (AAL) solves the management and virtualization problem, providing a uniform way for applications to communicate with accelerators and manage them as resources. This paper outlines some of the platform-level service needs of accelerator users and how introducing a new software framework for acceleration increases platform agility.

Read the full Get Consistent Platform-Level Services for Well-Bound Accelerators White Paper.