Intel Habana Gaudi3 baseboard with 8 HL-325L OAM cards

• 1.8 PFlops FP8 and BF16 Compute
• 128 GB HBM2e Memory
• 3.7 TB/s HBM Bandwidth
• 8 MME and 64 TPC Engines
• TSMC 5nm Process
• High Power Efficiency

• Deep Learning Workloads
• Artificial Intelligence Inference
• Generative AI Models
• Large Language Models
• Image Generation
• Standard Object Recognition
• Voice Dubbing Applications

The Intel Habana Gaudi3 baseboard with 8 HL-325L OAM cards is designed for high-performance deep learning and AI workloads. It features the Intel Gaudi3 AI Accelerator, manufactured using TSMC's 5nm process, offering advancements in power efficiency and compute capacity with 1.8 PFlops of FP8 and BF16 performance. The baseboard integrates 8 MME engines, 64 TPC engines, and 24 RDMA NIC ports for robust connectivity and processing. It also includes 128 GB of high bandwidth HBM2e memory, supporting up to 3.7 TB/s bandwidth, ideal for generative AI applications, including LLMs, diffusion models, and standard recognition tasks.