hard20 pts · first answer countsllm-fundamentalsmodel-selection
GPUs needed for peak traffic
One GPU server can generate about 2,500 output tokens per second in total across users. At peak you expect 400 users streaming at once, each needing 15 tokens per second. How many servers do you need at minimum, ignoring headroom?
Discussion 0
Join free to post in the discussion.
Related questions
- Tokens in a 3,000-word reportAI Engineer (LLM & Generative AI)
- How a reply is generatedAI Engineer (LLM & Generative AI)
- A base model ignores the questionAI Engineer (LLM & Generative AI)
- What 8 billion parameters meansAI Engineer (LLM & Generative AI)
- Do examples in a prompt train the modelAI Engineer (LLM & Generative AI)
- Does the reply use context spaceAI Engineer (LLM & Generative AI)
- Hello and hello are not the sameAI Engineer (LLM & Generative AI)
- What happens to a token id firstAI Engineer (LLM & Generative AI)