The Intel® Gaudi®2 AI accelerator is redefining deep learning capabilities with improved price-performance and operational efficiency. This powerhouse for AI and large language models (LLMs) is designed for scalable deployment, from cloud-based applications to local data centers. The Gaudi2 stands on the shoulders of its predecessor with advanced architectural features like 7nm process technology, 24 Tensor Processor Cores, and an impressive 96 GB HBM2E memory onboard, ensuring a robust and efficient AI processing environment.
For cloud applications, the Gaudi2 offers ease of use and high performance on the Intel Developer Cloud and will soon be available on the Genesis Cloud. Data centers can leverage its price-performance benefits through solutions from partners like Supermicro and IEI.
Intel Gaudi2's training and inference performance are notable, with MLPerf Training 3.0 results from June 2023 showcasing it as the sole viable alternative to H100 for training large language models such as GPT-3. It also performs well in other third-party evaluations.
Its 24x 100 Gigabit Ethernet ports per accelerator facilitate massive, flexible scalability, allowing performance to scale efficiently from a single unit to thousands. Furthermore, the SynapseAI Software Stack, optimized for the Gaudi platform, simplifies model development and migration, providing access to a vast library of over 50,000 models through the Hugging Face hub.