Overcoming Memory Bottlenecks in AI
ABSTRACT
As genAI scales, memory constraints become a key challenge. This keynote explores how AMD's GPU platforms overcome these limits with advanced 3D packaging for ultra-high memory density and bandwidth, enabling larger models with fewer GPUs. Through innovation and open-source collaboration, AMD offers the flexibility and performance AI leaders need to scale efficiently in a rapidly evolving field.
BIOGRAPHY
Ramine Roane currently holds the position of Corporate Vice President of AI Product Management at AMD in San Jose, CA, where he works on AI integration across all AMD platforms. In 2022, he was the VP of Data Center acceleration within the Adaptive and Embedded Computing Group. Before the merger between AMD and Xilinx, Ramine served as VP of Software & AI product management at Xilinx, overseeing projects like Vivado Design Suite, Vitis, and Vitis AI. His career has seen him at EDA companies including Synopsys and Mentor Graphics. He holds a Masters in Electronic Engineering and Computer Science from the National Polytechnic Institute of Grenoble, France