Llm Inference Serving medium
Every request to your assistant API starts with the same 3,000-token system prompt, followed by a ~200-token user message and a ~150-token reply. GPUs spend most of their time in prefill and TTFT is high. Which change gives the largest TTFT improvement, and what does it require?
- A Raise the maximum batch size so more prefills run together, amortizing the system prompt across requests in each batch.
- B Quantize weights to 4-bit so each prefill moves fewer bytes, which proportionally shortens the 3,200-token prefill.
- C Enable prefix caching so the system prompt's KV is computed once and reused; the shared prefix must be byte-identical.
- D Stream the first tokens sooner by flushing after the system prompt is prefetched, which hides the prefill cost from users.
Answer: C. Enable prefix caching so the system prompt's KV is computed once and reused; the shared prefix must be byte-identical.
**Prefix caching** skips about 94% of the prefill work here. The cost is GPU memory held by cached KV blocks and a stable prefix, so placing a timestamp or user ID at the top of the prompt breaks reuse. Bigger batches (A) still compute the shared prefix once for every request and raise latency, and 4-bit weights (B) mainly help bandwidth-bound decode, not compute-bound prefill. Larger batches would be the better lever only if prompts were unique and the GPU were underutilized.
One system design question a day, with spaced repetition so the ones you miss come back.
Get the appRelated questions
- 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 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?