The Definitive Guide toAI Data Centers
Ask the GuideAboutAccount
GuideGlossaryModel FLOPs Utilization

Model FLOPs Utilization · MFU

How much of a chip's theoretical compute a training run actually uses; the headline training-efficiency metric.

Current numbers

~15–30%AMD Instinct hardware price discount to NVIDIA SXM — real only if the realized-MFU gap on your workload is smalleras of 2025-2026 · register ↗
37–66%MI300X realized inference throughput vs H100/H200 (2025), despite ~1.3x paper FLOPS — the historical realized-MFU gapas of 2025 · register ↗
35–55%typical training MFU band for a well-tuned dense transformer on a mature stack; immature stacks roughly halve itas of 2025 · register ↗

← All terms