NTN 520 -   Escalar IA Duele

NTN 520 - Escalar IA Duele

From No Tiene Nombre by elbruno

May 19, 2026 · 23 min

About this episode

The episode discusses the complexities of scaling large language models compared to traditional web applications.

Servir un LLM no es como escalar una web app normal. Hablamos de tokens, GPUs, KV cache, latencia, costos y por qué “agregar más pods” ya no salva a nadie. Fuentes: • https://www.dheeth.blog/llm-serving-is-not-normal-web-serving/ • https://arxiv.org/abs/2309.06180 • https://kubernetes.io/docs/concepts/workloads/autoscaling/horizontal-pod-autoscale/

People in this episode

Host: elbruno

Topics covered

  • LLM serving
  • scaling
  • latency
  • costs
  • GPU
  • KV cache

Keywords

  • LLM
  • scaling
  • tokens
  • GPUs
  • latency
  • costs
  • KV cache

Mentioned in this episode

Organizations: dheeth.blog, arxiv, kubernetes.io

More episodes of No Tiene Nombre

Explore listener stats, chart rankings, contacts and more on the No Tiene Nombre podcast page.