<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/0unqVexS2" -->

---
title: Qwen 4 Isn't Out. I Ran Its Rehearsal on One 32 GB Card...
description: Qwen 3.8, a 27 billion parameter model, was run on a 32 GB card using VLLM as a rehearsal for its unreleased Qwen 4 successor. Quantized weights load at 13.2...
canonical: https://daily.dev/posts/0unqvexs2
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: Qwen 4 Isn't Out. I Ran Its Rehearsal on One 32 GB Card - 81 tok/s, Measured | daily.dev
og:description: Qwen 3.8, a 27 billion parameter model, was run on a 32 GB card using VLLM as a rehearsal for its unreleased Qwen 4 successor. Quantized weights load at 13.2...
og:url: https://daily.dev/posts/0unqvexs2
og:image: https://api.daily.dev/og/posts/0unqVexS2.png
og:image:alt: Post cover image
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Qwen 4 Isn't Out. I Ran Its Rehearsal on One 32 GB Card - 81 tok/s, Measured

**[Qwen](https://daily.dev/sources/qwen)** · [@nerdalytics](https://daily.dev/nerdalytics) · 11 upvotes · 1 comments

## Summary

Qwen 3.8, a 27 billion parameter model, was run on a 32 GB card using VLLM as a rehearsal for its unreleased Qwen 4 successor. Quantized weights load at 13.2 GiB, with 11.87 GB for KV cache, supporting up to 555,000 tokens of context. Measured throughput hit 81 tokens per second for a single request, above the 46 tokens per second bandwidth-based ceiling, which the author calls wrong by a factor of two. Eight concurrent requests shared 173 tokens per second total, dropping per-request speed to 29 to 53. The setup used VLLM 0.29, NVFP4 weights, 4-bit KV cache, and 262,000 context with seven draft tokens. The author expects Qwen 4 to need a similar 13.2 GB file size and plans to re-measure KV cache budget once it launches.

## Full article

daily.dev links to this article rather than hosting it. Read it at the original source: <https://www.youtube.com/watch?v=sf5oqxPu3Hk>

## Community discussion

Top comments from developers on daily.dev.

**@skepagent** · 1 upvotes

> 27B 在 32GB 卡上跑出 81 tok/s 确实挺惊人，不过实测超过了带宽上限 46 tok/s 的两倍，draft token 和 vLLM 的调度应该占了大头。好奇并发 8 个请求时 per-request 延迟的分布——29 到 53 这个区间跨度挺大，KV cache 在并发下的争抢值得再细看看。

## Similar posts on daily.dev

- [I dropped Qwen and Gemma for a 27B model that runs smoothly on my 8GB graphics card](https://daily.dev/posts/i-dropped-qwen-and-gemma-for-a-27b-model-that-runs-smoothly-on-my-8gb-graphics-card-h5gi3rwyo) · XDA Developers · 0 upvotes · 0 comments

---

Tags: [#data-science](https://daily.dev/tags/data-science), [#ai-inference](https://daily.dev/tags/ai-inference), [#local-ai](https://daily.dev/tags/local-ai), [#vllm](https://daily.dev/tags/vllm), [#qwen](https://daily.dev/tags/qwen)

[View this post on daily.dev](https://daily.dev/posts/0unqvexs2)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/0unqvexs2","headline":"Qwen 4 Isn't Out. I Ran Its Rehearsal on One 32 GB Card - 81 tok/s, Measured","text":"Shared: Qwen 4 Isn't Out. I Ran Its Rehearsal on One 32 GB Card - 81 tok/s, Measured","url":"https://daily.dev/posts/0unqvexs2","datePublished":"2026-11-01T23:00:42.503Z","dateModified":"2026-11-01T23:01:43.174Z","author":{"@type":"Person","name":"nerdalytics","url":"https://daily.dev/nerdalytics","image":"https://avatars.githubusercontent.com/u/97166791?v=4","description":"Husband. Cat and dog dad. Nerd.","interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"EndorseAction"},"userInteractionCount":14280}},"image":"https://media.daily.dev/image/upload/s--1KxV4ohY--/f_auto/v1722860400/public/Placeholder%2007","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":11},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"sharedContent":{"@type":"WebPage","url":"https://api.daily.dev/r/y2ZU6ZHrU"},"comment":[{"@type":"Comment","text":"27B 在 32GB 卡上跑出 81 tok/s 确实挺惊人，不过实测超过了带宽上限 46 tok/s 的两倍，draft token 和 vLLM 的调度应该占了大头。好奇并发 8 个请求时 per-request 延迟的分布——29 到 53 这个区间跨度挺大，KV cache 在并发下的争抢值得再细看看。","datePublished":"2026-11-02T02:55:52.525Z","url":"https://daily.dev/posts/0unqVexS2#c-deLcEOCPX","author":{"@type":"Person","name":"kris gu","url":"https://daily.dev/skepagent","image":"https://lh3.googleusercontent.com/a/ACg8ocJnnwFKNx7PJNhb7W6mFWr2k2SV_VHW_9ainY71DX_dv79frQ=s96-c"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1}}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/squads/qwen","name":"Qwen"}}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Qwen","item":"https://daily.dev/squads/qwen"},{"@type":"ListItem","position":3,"name":"Qwen 4 Isn't Out. I Ran Its Rehearsal on One 32 GB Card - 81 tok/s, Measured"}]}
```

