A100 study shows high GPU utilization masks low tensor core efficiency
tokenbender · x · 2026-08-19
Relying on nvidia-smi utilization metrics is misleading; it only indicates time with at least one kernel running. A DATE 2024 paper analyzing TensorFlow workloads on an A100 found that while GPU-level utilization was high, the average instruction issue rate was below 50%, and tensor-core instructions were below 5.2%. This highlights why newer architectures like Blackwell add features like TMEM to optimize data movement.
More from Infra
- White House AI chief discusses Empire AI, New York's shared compute infrastructure — mkratsios47 · 2026-10-16
- FT warns of 'Darwin moment' for AI borrowers as CoreWeave bonds sink to distress levels — firstadopter · 2026-10-16
- Phone as interface, home servers as the lab: a two-workflow local AI setup — EAccelerate_42 · 2026-10-16
- "A datacenter is a bank": CoreWeave drama revives calls for算力 supervision — arthurcolle · 2026-10-16
- DOT-MoE: converting dense models into sparse MoE via optimal transport — burkov · 2026-10-16
- Stanford CS336: A 7B model needs 14GB for weights but 84GB to train — stanfordnlp · 2026-10-16