---
title: "Comment by Kartik N V J K on \"vLLM Deep Dive: How PagedAttention, Continuous Batching and the V1 Engine Serve LLMs in Production\" | Hashnode"
description: "Fifty conversations at full 8k context is the number I'm taking from the worked example. I had been reading an 80 GB card as a token-throughput question, and wa"
url: "https://hashnode.com/posts/vllm-deep-dive-how-pagedattention-continuous-batching-and-the-v1-engine-serve-llms-in-production/6af1a0c4a32bf9f449f76bd3/comment/6af3551ff452f9a8a3de3f6e"
---

[Discussion](https://hashnode.com/posts/vllm-deep-dive-how-pagedattention-continuous-batching-and-the-v1-engine-serve-llms-in-production/6af1a0c4a32bf9f449f76bd3)

Comment [vLLM Deep Dive: How PagedAttention, Continuous Batching and the V1 Engine Serve LLMs in Production](https://hashnode.com/posts/vllm-deep-dive-how-pagedattention-continuous-batching-and-the-v1-engine-serve-llms-in-production/6af1a0c4a32bf9f449f76bd3)

[Kartik N V J K](https://hashnode.com/@kartik-nvj)

[3d ago](https://hashnode.com/posts/vllm-deep-dive-how-pagedattention-continuous-batching-and-the-v1-engine-serve-llms-in-production/6af1a0c4a32bf9f449f76bd3/comment/6af3551ff452f9a8a3de3f6e)

Fifty conversations at full 8k context is the number I'm taking from the worked example. I had been reading an 80 GB card as a token-throughput question, and walking the arithmetic down from 72 usable GB to 54 GB of KV makes concurrency the real budget line. Checking the KV cache size vLLM prints at startup also beats my habit of estimating capacity from the model card and hoping.