Kimi says MI355X serving changes cut p99 TTFT by 3.2× and lift throughput 7.7%
AnushElangovan · x · 2026-07-23
Kimi and AMD say they rebuilt the serving stack for Kimi K2.6 on AMD Instinct MI355X, targeting agent workloads with long contexts, bursty traffic, and frequent tool calls.
The update adds scheduler-aware multi-tier KV caching and claims up to 3.2× lower p99 TTFT plus 7.7% higher total-token throughput, with no accuracy loss. The post points readers to the technical blog for details.
More from Infra
- Compute is the Spice: Meta Eyes $10B Compute Lease Deal with Anthropic — thealexbanks · 2026-07-23
- Pinokio 8.0.40 hotfix checks Hugging Face token validity after login revocations — cocktailpeanut · 2026-07-23
- Developer Reflects on Local AI Hardware: Prepaid Workstations Reduce Experiment Friction — generativist · 2026-07-23
- Kimi-K3 is estimated at 50.4B active params, with 2.8T models needing 8 B200s — StefanoGogioso · 2026-07-23
- Open-source models may shift margins from models to infrastructure — matt_slotnick · 2026-07-23
- Graphsignal auto-tunes vLLM and SGLang flags using GPU profiles and telemetry — l0g1cs · 2026-07-23