Llm Inference Serving easy
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?
- A Higher peak FP16 compute (TFLOPS), because each generated token requires a full forward pass through the model.
- B More GPUs running data-parallel replicas, so each request gets more dedicated hardware throughput per token.
- C Higher memory bandwidth, because each decode step must stream all of the model weights from GPU memory.
- D A larger on-chip L2 cache, so frequently used attention heads stay resident between consecutive decode steps.
Answer: C. Higher memory bandwidth, because each decode step must stream all of the model weights from GPU memory.
**Decode is memory-bandwidth-bound**: each token reads every weight once, so per-sequence speed is roughly bandwidth divided by model bytes (about 1 TB/s / 14 GB, or roughly 70 tokens/s). Extra TFLOPS is the tempting answer, but it mainly speeds **prefill**, which is compute-bound, so it is the right lever when long prompts dominate. Extra replicas raise total throughput, not the speed of a single sequence.
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?
- 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?