Advanced Computing in the Age of AI | Saturday, April 20, 2024

A Cloud-Based Marketplace Designed to Ease FPGA Workload Acceleration 

A French “acceleration-as-a-service” company has joined the array of companies forming partnerships and ecosystems that push FPGA technology into the data center mainstream with the launch of a cloud-based marketplace that it says makes FPGA workload speed-ups more accessible to cloud users who are not familiar with FPGAs.

Accelize has introduced AccelStore, offerering ready-to-use accelerator functions provided by an ecosystem of nearly 20 third party IP providers, design houses and ISV’s (including Intel and Xilinx) that can be deployed on different cloud FPGA instances. According to the company, AccelStore hosts data compression, search-and-replace, image encoding and security functions on both the Amazon Web Services and France-based OVH public cloud services platforms.

Accelize claims performance improvements from 10x to more than 1000x compared to traditional pure software implementations..

Stephane Monboisset, director of marketing and business development for Accelize, told EnterpriseTech that AccelStore is designed to simplify testing and deploying accelerator functions on host applications. Using APIs, cloud users launch AccelStore accelerator functions on any supported cloud FPGA instance within minutes, speeding up evaluation of FPGA acceleration for specific workloads. The company said API functions also can be integrated into the end application, providing acceleration with minimal modification.

Initial accelerator functions include:

  • High compression Gzip, from CAST, providing compression speed-ups of 30x, according to Accelize.
  • Text-based find-and-replace capability from Nagase, “with expected acceleration factors over 1000x.”
  • JPEG Encoding, from A.L.S.E., for image encoding in JPEG format.
  • TRNG, a true random number generator from Secure-IC that provides acceleration in random number generation and “the highest levels of security.” This is the only cloud option that has been validated against NIST FIPS SP-800 90 and BSI AIS 31 statistical tests, industry standards for true random number generation.

“By connecting users with cloud service, and IP providers with a seamless workflow, Accelize addresses the key technical and business challenges of using FPGAs for application acceleration,” said Karl Freund, senior analyst, ML and HPC at analyst firm Moor Insights & Strategy. “Accelize’s design environment and app store-like business model can deliver impressive performance gains to users who are new to FPGAs.”

Stephane Monboisset of Accelize

Monboisset said additional accelerator functions are under development by partners and will be available in the coming weeks. He said Accelize will add support for other cloud service provider’s FPGA instances soon.

Accelerator function pricing is based on the volume of data processed – a pricing structure the company said provides accuracy and predictability.  The company also offers custom accelerator function development for cloud users needing functions not available on AccelStore.

“Having Cloud users be able to evaluate ready-to-use accelerator functions faster than it takes to get a cup of coffee is a major benefit for us to demonstrate and disseminate our technology among Cloud users,” said Nikos Zervas CEO of CAST. “Through AccelStore Cloud users can now experience the benefit of FPGA accelerated GZIP compression and deploy it in their Cloud application instantly. The data driven pay-per-use model also provides predictability on cost which is so important when deploying new technologies.”

EnterpriseAI