40G UCIe IP Advantages for AI Applications
By Aparna Tarde, Sr. Technical Product Manager and Manuel Mota, Sr. Product Manager - Synopsys
The deployment of generative AI in the devices we use every day is growing, driving demand for large language model sizes and higher compute performance. According to a presentation by Yole Group at the 2024 OCP Regional Summit, ‘For training on GPT-3 with 175 billion parameters, we estimate that between 6,000 and 8,000 A100 GPUs would have required up to a month to complete.’ Growing HPC and AI compute performance requirements are driving the deployment of multi-die designs, integrating multiple heterogeneous or homogenous dies in a single standard or advanced package. For AI workloads to be processed reliably at a fast rate, the die-to-die interface in multi-die designs must be robust, low latency, and most importantly high bandwidth. This article outlines the need for 40G UCIe IP in AI data center chips leveraging multi-die designs.
To read the full article, click here
Related Semiconductor IP
- UCIe Controller baseline for Streaming Protocols for ASIL B Compliant, AEC-Q100 Grade 2
- UCIe D2D Adapter
- UCIe Die-to-Die Chiplet Controller
- Simulation VIP for UCIE
- UCIe Controller add-on CXL3 Protocol Layer
Related White Papers
- 40G UCIe IP Advantages for AI Applications
- Menta eFPGA IP for Edge AI
- Generative AI for Analog Integrated Circuit Design: Methodologies and Applications
- Boosting RISC-V SoC performance for AI and ML applications
Latest White Papers
- CRADLE: Conversational RTL Design Space Exploration with LLM-based Multi-Agent Systems
- On the Thermal Vulnerability of 3D-Stacked High-Bandwidth Memory Architectures
- OmniSim: Simulating Hardware with C Speed and RTL Accuracy for High-Level Synthesis Designs
- Balancing Power and Performance With Task Dependencies in Multi-Core Systems
- LLM Inference with Codebook-based Q4X Quantization using the Llama.cpp Framework on RISC-V Vector CPUs