Accelerating the pace of machine learning

“Wave Computing is solving the performance and scalability challenges imposed by traditional solutions used in today’s advanced machine learning applications,” said Dr. Chris Nicol, CTO at Wave Computing. “Our dataflow-based compute appliance improves the performance of machine learning training and inferencing by orders of magnitude compared to CPU- or co-processor-based offerings. We selected Synopsys’ ZeBu Server because it delivered four times faster performance than other emulation solutions. ZeBu helped to further speed development of our dataflow software, part of our overall system offering, which began in 2012.”

Synopsys’ ZeBu Server emulation system helps enable faster verification and earlier software bring-up of complex SoC designs. Utilising Synopsys Verification Continuum’s Unified Compile with VCS technology, Wave Computing was able to quickly add ZeBu to the verification environment for its Dataflow Processing Unit SoC. Wave’s complex 16nm design features 6.7GHz performance, six billion transistors and high speed access to HMC memories. With ZeBu’s virtualisation technology, Wave Computing’s software teams were able to develop and debug its entire software stack, including the WaveFlow Software Development Kit (with compilers, linkers, simulators and debuggers), WaveFlow Dataflow Agent Library and WaveFlow Execution Engine (with a resource monitor, drivers and more), in its native Linux environment prior to taping out commercial silicon.

“The industry’s leading SoC design, verification and software teams increasingly require a high performance emulation solution to achieve aggressive time-to-market goals,” said Rohit Vora, Vice President of Engineering for the Synopsys Verification Group. “We continue our R&D collaboration with these leading teams to deliver high performance emulation technology, focused on reducing time-to-market by months.”