Llm Inference Serving easy
Your 70B-class model servers each handle about 40 concurrent chats with 4K-token contexts. Product raises the maximum context to 32K tokens, and many users start using it. What is the main capacity consequence?
- A Little change, because the KV cache is small next to the model weights, which dominate GPU memory at any length.
- B Far fewer concurrent requests fit, because KV cache grows linearly with length, about 8x per long request.
- C Throughput rises, because longer prompts give each forward pass more parallel work and better GPU utilization.
- D Only TTFT changes, because longer prompts affect prefill time while decode memory per request stays constant.
Answer: B. Far fewer concurrent requests fit, because KV cache grows linearly with length, about 8x per long request.
KV cache costs hundreds of KB **per token** for a 70B-class model, so a 32K-token request needs roughly 10 GB or more of KV and crowds out other requests. Treating the KV cache as small is a common mistake. It only holds at short contexts, where the weights dominate memory. The tradeoff is longer context for each user against fewer concurrent users on each GPU.
One system design question a day, with spaced repetition so the ones you miss come back.
Get the appRelated questions
- A 70B-class model (80 layers, grouped-query attention with 8 KV heads of dimension 128) is served in fp16 on one 8-GPU node with about 640 GB of total HBM. Product wants 256 concurrent sessions, each holding a full 32K-token context. What is the binding constraint, and what does it imply?
- Your server reserves a contiguous KV region sized for max_tokens = 8,192 for every request at admission, but the average sequence ends near 600 tokens. The node reports out-of-KV-memory at about 20 concurrent requests while GPU compute sits at 30%. Which change best raises concurrency without restricting legitimate long outputs?
- You must summarize 10 million documents overnight, and no user is waiting on any single result. How should you configure the serving stack compared with your interactive chat deployment?
- You serve a 7B-parameter model in fp16 (about 14 GB of weights) for a chat app. Users say single-conversation token generation is too slow, and you are choosing new GPU hardware. Which hardware spec most directly increases per-sequence decode speed?
- Your RAG workload sends 8K-token prompts and gets ~150-token answers from a 70B-class model. You plan to move prefill and decode onto separate GPU pools so long prefills stop stalling decode. What new resource must you size carefully?
- Speculative decoding with a small draft model cut inter-token latency by about 2x at low traffic. At peak, with about 200 sequences batched per replica, aggregate throughput is 15% lower than with speculation disabled. What explains this, and what is the fix?