Deep Cogito70BQ4_K_M 下约 43 GB 显存

Cogito v1 70B — 显存、速度与本地部署

作者: Jakub Rusinowski · 最后更新:

Largest Cogito variant for multi-GPU workstations. Scores 65.0% on AIME 2025 — comparable to DeepSeek-R1 671B on math benchmarks. Requires ~40 GB at Q4 (dual RTX 3090 or A6000). Best open-source reasoning model for teams with workstation hardware.

Cogito v1 70B 在 Q4_K_M 下约需 43 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92
创意84
编程89

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
26.5 GB~4 tok/s需卸载
Q3_K_M
3.41 bpw
33.3 GB~3 tok/s需卸载
Q4_K_M
4.83 bpw
45.7 GB~3 tok/s需卸载
Q5_K_M
5.67 bpw
53.1 GB~2 tok/s需卸载
Q6_K
6.56 bpw
60.9 GB—放不下
Q8_0
8.50 bpw
77.9 GB—放不下
F16
16.00 bpw
143.5 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. Cogito v1 70B 显存计算器 →

运行 Cogito v1 70B

目录中能运行 Cogito v1 70B 的最便宜 GPU 是 NVIDIA DGX Spark (128 GB).

购买此硬件 NVIDIA DGX Spark (128GB) — 128 GB VRAM · 150 W board power立即云端部署 RunPod 上的 NVIDIA A40 — 低至 $0.44/小时 · 价格核实于 2026-08

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。
NVIDIA DGX Spark (128GB)
128 GB VRAM · 150 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 Cogito v1 70B

安装 Ollama,然后运行:

ollama run cogito:70b
Hugging Face 上的权重: deepcogito/Cogito-v1-70B-Instruct ↗

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
70 Billion
上下文窗口
128,000
架构
Hybrid Reasoning Transformer
提供商
Deep Cogito
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
64 GB
记录更新于
2026-03-20
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reasoningmathresearchmulti gpu workstation
基准分数来源
MMLU88.9 / 100 %已发布
MATH83.7 / 100 %已发布
AIME 202565 / 100 %已发布
HumanEval88.4 / 100 %已发布

我的 GPU 能运行 Cogito v1 70B 吗?

Cogito v1 的其他尺寸

Cogito v1 70B — 常见问题

How much VRAM does Cogito v1 70B need?

About 43 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 Cogito v1 70B run on an RTX 4090 (24 GB)?

No. Cogito v1 70B needs about 43 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 Cogito v1 70B locally?

Install Ollama and run `ollama run cogito:70b`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.

What other sizes does Cogito v1 come in?

Cogito v1 3B (3 GB), Cogito v1 8B (6 GB), Cogito v1 14B (9 GB), Cogito v1 32B (20 GB), Cogito v1 70B (43 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.