Tokyo, June 12, 2023 /PRNewswire/ — EdgeCortix® Inc., an innovative edge artificial intelligence (AI) platform company focused on delivering best-in-class computing efficiency and ultra-low latency for AI inference, has announced that Molex’s BittWare will power EdgeCortix’s SAKURA-I Announced that it was selected as an excellent product. Adopts a class energy-efficient AI co-processor as an edge inference solution.
Dr. Sakyasingha Dasgupta, Founder and CEO of EdgeCortix, poses for photos with the latest SAKURA AI processor-powered machine learning system for power-efficient inference at the edge, from vision to generative AI applications.
EdgeCortix SAKURA-I has been benchmarked against leading GPU platforms in various power modes. The Edge GPU is expected to be a TSMC 7nm device while SAKURA-I is a TSMC 12nm device. End-to-end latency is measured with a batch size of 1 in all cases. All models are deployed to INT8 without pruning. Edge GPU measurements were compiled and optimized with the vendor’s latest tools. All SAKURA-I measurements were compiled and deployed on early access SAKURA boards using EdgeCortix MERA v1.4 software.
“We are delighted that BittWare is able to offer our products. Sakura eye “The AI chip is coming to market as part of a new artificial intelligence and machine learning inference acceleration program,” said Sakyasingha Dasgupta, Founder and CEO of EdgeCortix. We offer a complete edge AI acceleration solution for computer boxes: an edge inference server that can be used as a “drop-in” high-efficiency solution in place of power-hungry typical GPUs. From a performance standpoint, BittWare customers see very low inference latency and high data throughput while consuming less than 10 watts and achieving up to 8x performance improvement over leading GPUs. We can expect great things from SAKURA-I. We are excited to make our solutions available at scale through BittWare. ”
“For over 30 years, the BittWare brand has been trusted to bring the best acceleration technology to the market. ASICsA base acceleration solution optimized for AI inference is a natural step for both our business and our brand.” Craig Petrie, Vice President of BittWare. “Customers are already recognizing the wide range of opportunities for this solution, from surveillance and security applications, to quality assurance at manufacturing sites, to data closets in office buildings, and these broad edge use cases are well positioned for SAKURA-I. We are excited to partner with EdgeCortix to bring an edge-focused SAKURA-I acceleration solution to the market. ”
The SAKURA-I AI coprocessor is an AI-focused TSMC 12nm FinFET accelerator device designed for fast computation and power efficient inference processing in small to medium form factor systems. SAKURA-I is particularly well suited for real-time applications that use streaming data such as high-resolution camera and LIDAR data, real-time control of autonomous systems, radio frequency (RF) signal processing, generative AI applications and natural language processing. As a future 5G-AI integrated system. SAKURA-I coprocessors lead the industry across power and mobility sensitive edge applications such as airborne, underwater, ground-based autonomous or manned platforms, smart cities, smart manufacturing, security and surveillance, visual and text/voice. Provides AI inference capabilities. Combined(multimodal) systems, generative AI systems, real-time defense applications, next-generation AI-5G integrated systems. The SAKURA-I AI processor is specifically optimized for inference with real-time, streaming, high-resolution data. It supports mixed numerical precision for more accurate solutions and can handle the latest generation of AI models such as Transformers used in large language models and generative AI applications. SAKURA-I is equipped with a maximum 40 trillion operations per second (TOPS) computing engine called the Dynamic Neural Accelerator.® (DNA). DNA, EdgeCortix’s proprietary neural processing engine, with built-in run-time reconfigurable data paths, connects all compute engines to reduce power and increase compute utilization. DNA enables the SAKURA-I AI coprocessor to run multiple processors, multimodal It enables deep neural network models with ultra-low latency while maintaining outstanding TOPS utilization and low power consumption, key market differentiators.
Overview of SAKURA-I edge AI processor (product version):
- Up to 40 TOPS (INT8 single chip) and 20 TFLOPS (BF16 single chip). Multiple chips can be combined on one board.
- Real-time, low-latency inference using both generative and non-generative machine learning models, including convolution- and transformation-based neural networks.
- Powered by EdgeCortix’s patented Runtime Reconfigurable Data Path – Dynamic Neural Accelerator IP.
- On-chip dual 64-bit LPDDR4x – up to 16 GB on a single board.
- PCIe Gen 3 Up to 16 GB/s bandwidth.
- SAKURA Board TDP (10W to 15W based on form factor)
- The dynamic power for running modern neural network models on a chip is in the range of about 1W to 5W.
About EdgeCortix Inc.
Edge Cortix is an edge artificial intelligence (AI) platform as a service company pioneering the future of the connected intelligent edge. The company has the novel idea of taking a software-first approach while designing a run-time reconfigurable processor purpose-built for artificial intelligence from the ground up, using a technique called “hardware and software co-exploration.” was established in 2019. Target advanced computer vision applications first with unique hardware and software IP on existing processors such as FPGAs or custom designs. ASICsThe company aims to aggressively disrupt fast-growing AI hardware sectors such as defense, aerospace, smart cities, Industry 4.0, self-driving cars, and robotics.
© Copyright 2023 EdgeCortix, Inc. All rights reserved. EdgeCortix and all EdgeCortix logos are trademarks or registered trademarks of EdgeCortix, Inc. (or its group companies) in the United States and other countries. EdgeCortix Inc. is a wholly owned subsidiary of EdgeCortix Pte.Ltd.
Source Edgecortix, Inc.
