FP8 cache doubles Kimi capacity as Cloudflare hardens AI serving

Source: 4sysops.com· IT News· August 3, 2026
FP8 cache doubles Kimi capacity as Cloudflare hardens AI serving

Cloudflare has doubled the context capacity of Kimi K2.6 on its H200 infrastructure by storing the inference KV cache in FP8 instead of BF16. The change lets one deployment handle twice as many cached tokens and reach 2,192 tokens per second at high concurren…

This story was reported by 4sysops.com. Read the full original article:
Read on 4sysops.com

More in Enterprise & Adoption

View all