ET-SoC-1
RISC-V system on a chip for AI inference
The ET-SoC-1 is an artificial intelligence accelerator designed by Esperanto Technologies. It uses the RISC-V instruction set and features over 1,000 cores per chip. The device targets low-power generative AI workloads in data centres.
- Runs generative AI models at power levels as low as 25W per chip
- Contains over 1,000 64-bit RISC-V CPUs on each PCIe card
- Supports Meta’s Open Pre-Trained Transformer models
- Delivers up to 16,000 RISC-V CPUs per evaluation server
What it is
The ET-SoC-1 is a system on a chip developed by Esperanto Technologies for artificial intelligence tasks. It is built around the open RISC-V instruction set architecture. The chip is designed to handle massively parallel workloads, including machine learning and general-purpose computing. Its primary focus is on delivering high performance with low energy consumption for AI inference. The company states that it enables developers to run generative AI models efficiently.
How it came about
Esperanto Technologies announced in April 2023 that it had ported several generative AI models to the ET-SoC-1 hardware. This milestone included running versions of Meta’s Open Pre-Trained Transformer model. The company made this capability available to the RISC-V research community to accelerate development. The effort supports a strategy to extend RISC-V benefits from cloud to edge applications. Commercial customers already use the accompanying software development kit.
Inside the chip
Each PCIe card containing the ET-SoC-1 silicon holds more than 1,000 64-bit RISC-V central processing units. These cores are paired with attached vector and tensor units to handle mathematical operations required for AI. This many-core design allows for massive parallelism. The architecture aims to provide high throughput while maintaining low power usage. The system supports multiple precision levels and context sizes for large language models.
How it was made
Specific details regarding the manufacturing process node, transistor count, or foundry partner are not provided in the available information. The chip is integrated into PCIe cards for use in standard server form factors. Esperanto ships evaluation servers that contain these cards alongside host processors. The physical production details remain unspecified in the public announcements.
Where it went
The ET-SoC-1 is deployed in AI evaluation servers for commercial customers. These servers fit in a standard 2U-high rack mount and include dual Xeon host processors. Each server can hold eight or sixteen ET-SoC-1 PCIe cards, providing up to 16,000 RISC-V CPUs. The hardware is used for running industry-standard AI models and custom data sets. It targets applications where low power and reduced total cost of ownership are critical.
Versions and family
The hardware is described as a unified solution for AI inference and general-purpose parallel workloads. No predecessor or successor chips are mentioned in the provided text. The focus remains on the current generation of RISC-V based accelerators.
Why it matters
The chip represents an effort to democratize AI development through open standards. By supporting RISC-V, it offers researchers an alternative to proprietary platforms. Partners such as Penguin Computing help deploy these low-power solutions to broader markets. The technology aims to differentiate developers through customization and energy efficiency. It signals growing industry interest in RISC-V for high-performance computing tasks.
Problems and reception
Industry analysts have noted the potential for RISC-V to drive innovation in generative AI. Leaders from RISC-V International praised the move towards accessible development tools. The announcement was framed as a milestone for the semiconductor sector. Researchers are invited to evaluate the hardware under open-source license terms. Commercial partners highlight the benefits of reduced operational costs.
Questions people ask about ET-SoC-1
What instruction set does the ET-SoC-1 use?
The ET-SoC-1 is based on the RISC-V instruction set architecture. This open standard allows for customization and collaboration in chip design. Esperanto Technologies uses it to build high-performance, energy-efficient AI solutions.
How much power does the chip consume during AI tasks?
The company states that the chip can run AI inferencing workloads at power levels as low as 25 watts. This low energy draw is a key feature for reducing the total cost of ownership in data centre deployments.
How many processing cores are on the device?
Each PCIe card featuring the ET-SoC-1 contains over 1,000 64-bit RISC-V CPUs. These cores are equipped with vector and tensor units to accelerate machine learning calculations and parallel processing tasks.
Good reading and sources
There is no English Wikipedia article on ET-SoC-1 yet, so the facts on this page come from the sources listed above: esperanto.ai/News/generative-ai-models-running-on-et-soc-1/, markets.financialcontent.com/prnews/article/bizwire-2023-4-19-esperant. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryChip wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.