Nemotron 3 Super 120B-A12B — 显存、速度与本地部署
作者: Jakub Rusinowski · 最后更新:
120B total / 12B active MoE. The low active-parameter count means throughput closer to a 12B model than a 120B one, but the full weight set still has to be resident — roughly 65 GB at Q4_K_M, so dual RTX 3090/4090 or a single 80GB datacenter card. Q8_0 and BF16 tags are published for workstations with more headroom. NVIDIA releases open training data and recipes with this family, which is unusual among open-weight releases.
Nemotron 3 Super 120B-A12B 在 Q4_K_M 下约需 73 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。
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 | 显存 | 速度(估算) | 适配 |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 40.2 GB | ~16 tok/s | 需卸载 | |
| Q3_K_M 3.41 bpw | 52 GB | ~14 tok/s | 需卸载 | |
| Q4_K_M 4.83 bpw | 73.3 GB | — | 放不下 | |
| Q5_K_M 5.67 bpw | 85.9 GB | — | 放不下 | |
| Q6_K 6.56 bpw | 99.2 GB | — | 放不下 | |
| Q8_0 8.50 bpw | 128.3 GB | — | 放不下 | |
| F16 16.00 bpw | 240.8 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. Nemotron 3 Super 120B-A12B 显存计算器 →
运行 Nemotron 3 Super 120B-A12B
目录中能运行 Nemotron 3 Super 120B-A12B 的最便宜 GPU 是 NVIDIA DGX Spark (128 GB).
或在 Vast.ai 比较,低至 $0.77/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。
如何运行 Nemotron 3 Super 120B-A12B
安装 Ollama,然后运行:
ollama run nemotron-3-super:120b-a12b规格
Preview — The model is released, but these specs are thin or rest on a single source. Individual fields may be wrong.
Preview — The model is released, but these specs are thin or rest on a single source. Individual fields may be wrong.
- 参数量
- 120 Billion (12B active)
- 上下文窗口
- 128,000
- 架构
- Mixture-of-Experts (reasoning-trace)
- 提供商
- NVIDIA
- 许可证
- NVIDIA Open Model License
- 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
- 2026-08-15
Weights are downloadable and commercial use is permitted, subject to the licence’s acceptable-use terms.
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。
我的 GPU 能运行 Nemotron 3 Super 120B-A12B 吗?
- Nemotron 3 Super 在 AMD Ryzen AI Max+ 395 上
- Nemotron 3 Super 在 Apple M1 Max 上
- Nemotron 3 Super 在 Apple M1 Pro 上
- Nemotron 3 Super 在 Apple M1 Ultra 上
- Nemotron 3 Super 在 Apple M2 Max 上
- Nemotron 3 Super 在 Apple M2 Pro 上
- Nemotron 3 Super 在 Apple M3 Max 上
- Nemotron 3 Super 在 Apple M3 Pro 上
- Nemotron 3 Super 在 Apple M4 上
- Nemotron 3 Super 在 Apple M4 Max 上
- Nemotron 3 Super 在 Apple M5 上
- Nemotron 3 Super 在 Apple M5 Max 上
- Nemotron 3 Super 在 Apple M5 Pro 上
- Nemotron 3 Super 在 NVIDIA A100 80GB (PCIe) 上
- Nemotron 3 Super 在 NVIDIA DGX Spark 上
- Nemotron 3 Super 在 NVIDIA GeForce RTX 5090 上
- Nemotron 3 Super 在 NVIDIA H100 80GB (PCIe) 上
- Nemotron 3 Super 在 NVIDIA L40S 上
- Nemotron 3 Super 在 NVIDIA RTX 6000 Ada Generation 上
- Nemotron 3 Super 在 NVIDIA RTX PRO 6000 Blackwell 上
Nemotron 3 Super 120B-A12B — 常见问题
How much VRAM does Nemotron 3 Super 120B-A12B need?
About 73 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 Nemotron 3 Super 120B-A12B run on an RTX 4090 (24 GB)?
No. Nemotron 3 Super 120B-A12B needs about 73 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 Nemotron 3 Super 120B-A12B locally?
Install Ollama and run `ollama run nemotron-3-super:120b-a12b`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.