Seminar Overview
In this session, Fixstars engineers will dive into AI performance engineering from a systems perspective—demonstrating how to detect and eliminate inefficiencies across GPU, CPU, and interconnect layers in both training and inference pipelines. You’ll learn how to perform low-level hardware profiling, interpret utilization metrics, and apply targeted kernel-level and I/O optimizations that translate directly into measurable throughput and cost efficiency gains.
Through real-world case studies—including a 5× acceleration in Llama-3 training and a 50% reduction in inference costs—we’ll illustrate concrete optimization paths using Fixstars AIBooster, a performance engineering platform for analyzing workload characteristics and scaling performance tuning. A live demonstration will show end-to-end profiling, bottleneck analysis, and optimization workflows that engineers can reproduce within their own AI infrastructure environments.
Speaker
Sam is a full-stack software engineer at Fixstars Solutions, who specializes in the Frontend Engineering. His work includes working with React, Golang, and Python. He plays a key role in the development of Fixstars AIBooster. He pursued his education at California State University Long Beach and University of California Irvine and has a B.S in Computer Engineering Technology and a Minor in Computer Science.
Agenda
- What is AI Performance Engineering?
- Common Practices in AI Performance Engineering
- Practical Optimization Techniques for AI Workloads
- How Fixstars AIBooster Solves These Challenges
- Q&A Session
Total 40 minutes
* You can enter and leave the session at any time.
* Please note that the schedule and content may change without prior notice.
AIBooster is a performance engineering platform for continuously observing and improving the performance of AI workloads.
Through comprehensive dashboards, users can visualize the utilization efficiency of various hardware resources—including CPU, GPU, interconnect, and storage—as well as identify software bottlenecks to analyze AI workload performance characteristics. Furthermore, by applying optimization frameworks specifically designed for AI workloads, users can achieve efficient performance improvements.
Date and time
Wednesday, October 30, 2025
12:00 PM - 12:40 PM PDT
2:00 PM - 2:40 PM CDT
3:00 PM - 3:40 PM EDT
Location
Zoom
Target Audience
- AI Infrastructure Managers and Engineers
- Technical Leaders of AI Development Teams
- Project Managers aiming for cost-efficient AI projects
- Data Scientists and Researchers interested in performance engineering
Participation fee
Free
