NVIDIA16BBF16 下约 10 GB 显存

Cosmos 3 Nano — 显存、速度与本地部署

Name: LLM Configurator — GPU VRAM Checker
Author: LLM Configurator

作者： Jakub Rusinowski · 最后更新： 2026年7月21日

The mid-tier workhorse — a 16B-parameter two-tower MoT on a dense 8B backbone, tuned for high-quality video and action reasoning in a fraction of a second. Nano is the balance point: much cheaper to serve than Super while still leading open models (NVIDIA reports it matches the second-best result on R-Bench). Weights are ~32 GB at BF16, but NVIDIA recommends an RTX PRO 6000 / 96 GB-class GPU for full-context BF16 inference because video generation activations are memory-hungry. Omnimodal I/O (text, image, video, audio, action). OpenMDW-1.1. Specs from launch coverage and the Hugging Face model card — verify before relying on them.

Cosmos 3 Nano 在 BF16 下约需 10 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85

创意88

编程32

按量化级别的显存与速度

计算基准：NVIDIA RTX 4090 (24 GB)。仅含权重与开销：该模型架构未公开，因此未计入 KV 缓存。

量化	显存	速度（估算）	适配
Q2_K 2.63 bpw	6.1 GB	~150 tok/s	可运行
Q3_K_M 3.41 bpw	7.6 GB	~130 tok/s	可运行
Q4_K_M 4.83 bpw	10.5 GB	~104 tok/s	可运行
Q5_K_M 5.67 bpw	12.1 GB	~93 tok/s	可运行
Q6_K 6.56 bpw	13.9 GB	~84 tok/s	可运行
Q8_0 8.50 bpw	17.8 GB	~69 tok/s	可运行
F16 16.00 bpw	32.8 GB	~6 tok/s	需卸载

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. Cosmos 3 Nano 显存计算器 →

运行 Cosmos 3 Nano

目录中能运行 Cosmos 3 Nano 的最便宜 GPU 是 Intel Arc B580 (12 GB).

购买此硬件 Intel Arc B580 12GB — 12 GB VRAM · 190 W board power 立即云端部署 RunPod 上的 RTX 4090 — 低至 $0.34/小时 · 价格核实于 2026-07

或在 Vast.ai 比较，低至 $0.35/小时 (typical low · varies)

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。

Intel Arc B580 12GB

12 GB VRAM · 190 W board power

2026年价格波动较大——请以当前商品页价格为准。

在亚马逊查看价格

如何运行 Cosmos 3 Nano

安装 Ollama，然后运行：

$ollama run cosmos-3

Hugging Face 上的权重： nvidia/Cosmos3-Nano ↗

规格

Preview — The model is released, but these specs are thin or rest on a single source. Individual fields may be wrong.

参数量: 16B total (8B dense transformer backbone)
上下文窗口: Omnimodal I/O (video/action)
架构: Two-tower Mixture-of-Transformers — autoregressive reasoner (VLM) + diffusion generator, shared multimodal attention
提供商: NVIDIA
许可证: OpenMDW-1.1
Practical deployment: Laptop
规格量化: BF16
系统内存: 96 GB
记录更新于: 2026-07-21

许可证OpenMDW-1.1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roboticsworld simulationvideo generationaction reasoningsynthetic data

基准	分数	来源
R-Bench (Image-to-Video, open-source)	2 / 2 2nd best (rank)	厂商声称 · NVIDIA Cosmos 3 (launch)

我的 GPU 能运行 Cosmos 3 Nano 吗？

Cosmos 3 的其他尺寸

Cosmos 3 Nano — 常见问题

How much VRAM does Cosmos 3 Nano need?

About 10 GB at BF16 — 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 Cosmos 3 Nano run on an RTX 4090 (24 GB)?

Yes. Cosmos 3 Nano needs about 10 GB at BF16, inside a 24 GB card, at an estimated 104 tokens/sec.

How do I run Cosmos 3 Nano locally?

Install Ollama and run `ollama run cosmos-3`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.

What other sizes does Cosmos 3 come in?

Cosmos 3 Super (39 GB), Cosmos 3 Nano (10 GB), Cosmos 3 Edge (3 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.