High performance. High-efficiency. Inclusive
This new processor addresses an industry gap by providing customers with high-performance, high-efficiency deep learning compute choices for both training workloads and inference deployments in the data center while lowering the AI barrier to entry for companies of all sizes.
Gaby Hayon, executive vice president of R&D at Mobileye
Chetan Paul, vice president for Technology Innovation, Government Health and Safety Solutions at Leidos
Charles Liang, president and CEO of Supermicro
Deep Learning Training Efficiency
Gaudi2 offers significant improvements in training performance, with a process leap from 16nm to 7nm, boosting compute, memory, and networking capabilities.
It provides a substantial increase in in-package memory capacity (96GB of HBM2E) and integrates 24 x 100GbE RoCE RDMA NICs, enabling scalability.
Gaudi2 provides higher-performance deep learning training compared to GPU-based acceleration, resulting in cost savings and faster training times.
It accelerates vision modeling in various applications, including autonomous vehicles, medical imaging, defect detection in manufacturing, and natural language processing.
Networking Capacity and Flexibility
Gaudi2 simplifies scaling and configuration with integrated industry-standard RoCE. It allows easy integration with a wide array of Ethernet switching and networking equipment, reducing costs and avoiding vendor lock-in.
Simplified Model Build and Migration
The SynapseAI software suite optimizes deep learning model development and migration from GPU-based models to Gaudi hardware. It supports training on Gaudi2 and inference on various targets. Developers receive extensive documentation, tools, and community support for seamless migration.
Our team is primed and eager to provide assistance, be it for individual components or the ambitious endeavor of constructing a groundbreaking supercomputer on a global scale.