September 08, 2017
FPGA-as-a-Service: How to develop efficient FPGA accelerators for Amazon F1 in ½ day with Accelize


On September 8th in Gent, Belgium, Accelize will hold a tutorial showing how to develop FPGA accelerators to target Amazon F1 using QuickPlay. Registration to the event can be done here:

Following the success of last year’s QuickPlay presentation during the FSP workshop at FPL 2016, we decided to create a dedicated tutorial for all engineers and scientists eager to understand how they can take part in the FPGA-as-a-Service revolution using our framework.

Over the last 6 years, Accelize has developed QuickPlay and QuickStore to uniquely enable the creation of FPGA accelerators to be sold as-a-Service by Cloud Service Providers such as Amazon with their F1 instance.

The attendees of this tutorial will learn how to go through a complete FPGA accelerator design using 3rd Party IP cores, integrating their own algorithms using C/C++, targeting the Amazon F1 instance, and running/testing their design on the Amazon F1 instance all within the course of this tutorial, and this regardless of their HW/FPGA expertise.

Amazon, Microsoft, Alibaba Cloud, OVH and many others have announced the availability of FPGAs in their Cloud infrastructures. This represents a major opportunity for the industry but also a major challenge as the success of these offerings will greatly depend on the availability of a large selection of ready-to-use efficient FPGA accelerators that Cloud end users can load onto their FPGA instances seamlessly (without even caring about the fact that these are FPGA designs).

The creation of this large catalog of FPGA accelerators requires a broad ecosystem of “accelerator developers” that can develop these accelerators very quickly and efficiently leveraging a large selection of 3rd Party IP cores without having to pay high upfront cost for these cores. Accelize has developed the QuickPlay/QuickStore framework to deliver just this:

QuickPlay is a software-defined FPGA design platform that enables a broad ecosystem of IP providers and accelerator developers. It offers a high-level C/C++ data-flow programming model that enables the integration of plug & play 3rd party IP blocks for users to choose from QuickStore. QuickPlay’s pure functional modeling fully abstracts the target hardware, while its cloud compilation fully abstracts FPGA tools.  QuickPlay is built on a unique correct-by-construction methodology that relies on source-level debug and therefore entirely eliminates the need for low-level hardware debug and logic simulation.  Finally, QuickPlay & QuickStore align FPGA acceleration monetization with the as-as-service cloud business model, through a unique pay-per-use mechanism based on any measurable metric such as time or data volumes.

Similar to the Apple Apps Store (and Google Play store) which democratized the development of applications for the smart phones, QuickPlay and QuickStore enables many companies, individuals and universities to develop valuable and efficient accelerators with this unique and fast growing tool flow, whether it is for research purposes or to be sold as a service on Amazon or other Cloud Services.

Tags: FPGA accelerators, Amazon F1, QuickPlay, FPGA-as-a-Service, QuickStore