The Definitive Guide toAI Data Centers
Ask the GuideAboutAccount
GuideGlossaryQuantization

Quantization

Using lower numerical precision (FP8, FP4, INT8) to cut memory and boost throughput, trading some accuracy for speed.

Current numbers

93.3% (~15x)KV-cache memory reduction from attention architecture and quantization; DeepSeek-V2 reported a 93.3% reduction (~15×) from MLA versus its DeepSeek-67B baselineas of 2024-05 · register ↗

← All terms