Esperanto Technologies™, a developer of high-performance, energy-efficient machine learning (ML) inference accelerators based on the RISC-V instruction set, revealed new details of the company’s new ET-SoC-1 “supercomputer-on-a-chip” at the Hot Chips 33 conference on August 24.
Featuring over a thousand RISC-V custom processor cores, Esperanto’s ML inference accelerator is designed to be the highest performance commercial RISC-V chip. Designed to meet the high-performance but still air-cooled, low-power requirements of large-scale datacenter customers, Esperanto’s RISC-V-based inference chip is a general purpose, parallel processing solution that can accelerate many parallelizable workloads.
The Esperanto chip is designed to operate at under 20 watts in order to fit within enterprise customers’ demanding system power constraints. The chip includes over a thousand energy-efficient RISC-V, ET-Minion™ processor cores each with its own custom vector / tensor unit, four high performance out-of-order RISC-V, ET-Maxion™ processor cores, and a high-performance memory system.
The Esperanto chip is designed to excel at machine learning recommendation models, one of the most important types of AI workloads in many large datacenters. Additional details about the new chip will be provided at the Hot Chips 33 Conference.
The company will begin an Early Access Program for qualified customers later this year.
“I’m impressed with the energy efficiency and memory bandwidth of this new AI chip, which is the first and fastest RISC-V design we have seen for high-performance data center workloads,” said Karl Freund, founder and principal analyst at Cambrian AI Research. “In addition, there are significant benefits to the programmer of using standard RISC-V cores, including flexibility and the ability to easily optimize code.”
“Esperanto’s chip is focused on high throughput with energy efficiency,” said Rich Wawrzyniak, principal market analyst at Semico Research Corporation. “With so many RISC-V cores operating in unison, the ability to process very large models operating using their low-voltage approach really delivers on the performance per watt.”
