Tessera system unlocks heterogeneous GPUs for AI workloads

A new research paper introduces Tessera, a system designed to optimize AI workload performance on heterogeneous GPU clusters. Tessera operates at the kernel granularity, extracting inter-kernel dependencies to ensure correctness and overlapping communication with computation. Evaluations show Tessera can improve serving throughput by up to 2.3x and cost efficiency by 1.6x compared to existing methods, even outperforming homogeneous GPU setups in some cases. AI

IMPACT Optimizes AI inference on heterogeneous hardware, potentially reducing costs and increasing throughput.

RANK_REASON Research paper detailing a new system for AI infrastructure. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• graphics processing unit
• Parallel Thread Execution
• tessera
• Tiancheng Hu

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →