Alibaba Cloud235B (22B active)Q4_K_M 下约 143 GB 显存

Qwen 3 235B-A22B (MoE) — 显存、速度与本地部署

作者: Jakub Rusinowski · 最后更新:

The flagship Qwen 3 model. 235B total with only 22B active per token — comparable compute cost to a dense 22B model. Tops open-source leaderboards. Requires Apple Silicon 128GB+ or multi-GPU.

Qwen 3 235B-A22B (MoE) 在 Q4_K_M 下约需 143 GB 显存——量化权重加框架开销,不含 KV 缓存。在 Apple Silicon 上,这部分来自统一内存。

逻辑99
创意92
编程99

按量化级别的显存与速度

计算基准:NVIDIA RTX 4090 (24 GB)。包含 8K 上下文的 KV 缓存,因此数值高于上方的主数字。

量化显存速度(估算)适配
Q2_K
2.63 bpw
79.6 GB—放不下
Q3_K_M
3.41 bpw
102.5 GB—放不下
Q4_K_M
4.83 bpw
144.3 GB—放不下
Q5_K_M
5.67 bpw
168.9 GB—放不下
Q6_K
6.56 bpw
195.1 GB—放不下
Q8_0
8.50 bpw
252.1 GB—放不下
F16
16.00 bpw
472.4 GB—放不下

黑色标记 = NVIDIA RTX 4090 (24 GB) 上的可用显存。 估算来自内存带宽屋顶线模型,详见 方法说明页. Qwen 3 235B-A22B (MoE) 显存计算器 →

运行 Qwen 3 235B-A22B (MoE)

目录中能运行 Qwen 3 235B-A22B (MoE) 的最便宜 GPU 是 Apple M3 Ultra (512 GB).

购买此硬件 Apple Mac Studio M3 Ultra — 512 GB VRAM · 60 W board power立即云端部署 RunPod

或在 Vast.ai 比较

作为亚马逊联盟成员,我们从符合条件的购买中获得收入。云 GPU 链接为推荐链接——我们可能获得佣金,您无需额外付费。

联盟营销声明: 本页部分链接为联盟推广链接——如果你通过它们购买,LLM Configurator 可能会获得佣金,而你无需支付任何额外费用。作为亚马逊联盟成员(Amazon Associate),LLM Configurator 会从符合条件的购买中获得收益。
Apple Mac Studio M3 Ultra
512 GB VRAM · 60 W board power
2026年价格波动较大——请以当前商品页价格为准。

如何运行 Qwen 3 235B-A22B (MoE)

安装 Ollama,然后运行:

ollama run qwen3:235b-a22b
Hugging Face 上的权重: Qwen/Qwen3-235B-A22B ↗

规格

Curated — A hand-written entry from before this catalogue recorded its sources. The figures are long-standing but their provenance is not on file.

参数量
235 Billion (22B active)
上下文窗口
128,000
架构
MoE
提供商
Alibaba Cloud
许可证
Apache 2.0
Practical deployment
Workstation (large unified memory or pro GPU)
规格量化
Q4_K_M
系统内存
128 GB
记录更新于
2025-04-28
许可证Apache-2.0允许商业使用

Commercial use permitted. No usage restrictions beyond attribution.

质量与使用场景

评分由模型作者或独立评测方发布——衡量质量而非吞吐量,并非我们实测。

最适合frontier reasoningcomplex codingresearchenterprise

我的 GPU 能运行 Qwen 3 235B-A22B (MoE) 吗?

Qwen 3 的其他尺寸

Qwen 3 235B-A22B (MoE) — 常见问题

How much VRAM does Qwen 3 235B-A22B (MoE) need?

About 143 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.

Does Qwen 3 235B-A22B (MoE) run on an RTX 4090 (24 GB)?

No. Qwen 3 235B-A22B (MoE) needs about 143 GB at Q4_K_M, more than a single RTX 4090's 24 GB. It needs a larger card, several GPUs, or Apple Silicon with enough unified memory — or it runs with part of the weights offloaded to system RAM, which is much slower.

How do I run Qwen 3 235B-A22B (MoE) locally?

Install Ollama and run `ollama run qwen3:235b-a22b`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.

What other sizes does Qwen 3 come in?

Qwen 3 8B (6 GB), Qwen 3 14B (10 GB), Qwen 3 32B (21 GB), Qwen 3 30B-A3B (MoE) (19 GB), Qwen 3 235B-A22B (MoE) (143 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.