Is this your business? Claim it to manage your IP and profile
The Chimera GPNPU is a versatile neural processing unit designed to meet the extensive range of demands in modern AI applications. It centralizes matrix, vector, and scalar operations into a single execution pipeline, thereby eliminating the need for splitting workloads between multiple processor types, as commonly seen with conventional SoCs. The GPNPU’s single semiconductor IP design allows developers to seamlessly execute complex machine learning models alongside traditional C++ code, making it a universally applicable solution for a variety of market sectors.<br><br>This neural processing unit is highly scalable, ranging from modest computational power at a single-core level to expansive setups exceeding 864 TOPs for heavier workloads, catering to both power-users and intricate automotive applications. The architecture supports a wide range of standard and emerging machine learning models, from the traditionally-used backbones to the latest advancements in Vision Transformers and Large Language Models. Its fully programmable nature empowers developers to continually optimize model performance and adapt to unseen AI trends without the constraints of hardware limitations.<br><br>Notably, the Chimera GPNPU architecture prioritizes efficiency alongside programmability. It incorporates a hybrid model mixing Von Neumann computing with SIMD matrix capabilities, further enhanced by a comprehensive instruction set optimized for deep learning. This amalgamation fosters compatibility with a diverse set of applications, ensuring developers can implement dynamic problem-solving strategies and customize SoC solutions that anticipate future technological innovations.
The Chimera Software Development Kit (SDK) is a robust environment engineered to harness the power of Quadric’s Chimera GPNPU for the creation and optimization of complex AI-driven applications. It facilitates an effortless integration of both traditional machine learning algorithms and contemporary C++ code into a unified development framework. Users can exploit its versatile functionalities via the Quadric DevStudio, either online or through local deployment, ensuring flexibility to match various development setups.<br><br>The SDK embodies a drag-and-drop interface allowing developers to seamlessly blend pre-optimized C++ kernels and machine learning models, fostering an ideal ecosystem for experimentation and iterative testing. This integrated approach simplifies the application code chain, catering specifically to advancements in ML models like vision transformers and large language models. Moreover, a Docker image of the full Chimera SDK is available for secure deployment on private systems, enabling teams to compile proprietary codes and ML models using specialized tool flows unique to Quadric.<br><br>Central to the SDK’s power is the Chimera Graph Compiler, which supports input from leading ML training frameworks like TensorFlow and PyTorch. It analyzes and optimizes ML graphs before converting them into efficient C++ code, which may then be compiled by the Chimera LLVM C++ compiler. This important capability ensures developers can achieve optimal application performance by simulating real-world workloads efficiently, leveraging the cycle-accurate Instruction Set Simulator offered within the kit.
Join the world's most advanced semiconductor IP marketplace!
It's free, and you'll get all the tools you need to evaluate IP, download trial versions and datasheets, and manage your evaluation workflow!
To evaluate IP you need to be logged into a buyer profile. Select a profile below, or create a new buyer profile for your company.
