Abstract: On April 27, according to foreign media reports, Esperanto Technologies, a chip design start-up headquartered in California, said that Samsung's IT services department and a number of unnamed companies cooperated to evaluate its AI inference accelerator ET-SoC-1.
April 27 news, according to foreign media reports, California-based chip design startup Esperanto Technologies said that Samsung's IT services department and a number of unnamed companies and other cooperation evaluated its AI inference accelerator ET-SoC-1.
According to reports, ET-SoC-1 contains 1088 RISC-V architecture 64-bit low-power cores, with vector/tensor operation units, application of accelerated machine learning operation functions, and more than 160 million Bytes of SRAM memory, as well as connecting NAND and external DRAM.
Esperanto Technologies notes that with this architecture, ET-SoC-1 will be the fastest RISC-V architecture chip. Combined with low-power capabilities, the ET-SoC-1 meets the performance needs of large data center customers while also reducing power consumption and total cost of ownership. ET-SoC-1 can run on any machine learning workload. Hyperscale data center operators such as Facebook's parent company, Meta, and Amazon often use machine learning to execute their plans.
Esperanto Technologies was founded in 2014 by Dave Ditzel, a veteran developer in the semiconductor industry who previously held research positions at Intel's parallel computing architecture and led sparrch CPU instruction set architecture development at Sun Microsystems earlier in his career. Esperanto Technologies is currently led by Art Swift, who has served as CEO of several semiconductor executives such as Wave Computing.
△Esperanto Technologies core team (Source: Core Zhixun shooting)
Esperanto Technologies said it is looking for companies to work with, Esperanto Technologies allows users to put ET-SoC-1 into a variety of off-the-shelf AI models, and the program allows users to test up to 32 cluster models, data types, batch sizes and compute configurations, integrating them into innovative ecosystem developments.
Editor: Xin ZhiXun - Langke Sword