证据: 基准测试结果由厂商自行报告
Granite Switch 4.1 30B — 显存、速度与本地部署
作者： Jakub Rusinowski · 最后更新：
在 Granite 4.1 30B 基础模型上采用相同的组合结构：一个检查点共 30.0B 参数，66 层，上下文 128K。提供官方 GGUF。
Granite Switch 4.1 30B 在 Q4_K_M 下约需 19 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)。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 | 显存 | 速度（估算） | 适配 |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 12.9 GB | ~60 tok/s | 可运行 | |
| Q3_K_M 3.41 bpw | 15.8 GB | ~48 tok/s | 可运行 | |
| Q4_K_M 4.83 bpw | 21.1 GB | ~36 tok/s | 可运行 | |
| Q5_K_M 5.67 bpw | 24.3 GB | ~5 tok/s | 需卸载 | |
| Q6_K 6.56 bpw | 27.6 GB | ~4 tok/s | 需卸载 | |
| Q8_0 8.50 bpw | 34.9 GB | ~3 tok/s | 需卸载 | |
| F16 16.00 bpw | 63 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. Granite Switch 4.1 30B 显存计算器 →
运行 Granite Switch 4.1 30B
目录中能运行 Granite Switch 4.1 30B 的最便宜 GPU 是 AMD Radeon RX 7900 XTX (24 GB).
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。
如何运行 Granite Switch 4.1 30B
No Ollama tag. Use IBM's official GGUF with llama.cpp.
规格
Verified — Checked against the primary source — the model card or the vendor spec page — and corroborated by a second independent source.
- 参数量
- 30.0 Billion (base + adapters + encoder)
- 上下文窗口
- 131,072
- 架构
- Dense Transformer (GQA) with aLoRA adapters
- 提供商
- IBM
- 许可证
- Apache-2.0
- Practical deployment
- Single consumer GPU
- 规格量化
- Q4_K_M
- 系统内存
- 32 GB
- 记录更新于
- 2026-11-03
Commercial use permitted. No usage restrictions beyond attribution.
限制与注意事项
- A composed checkpoint: the adapters are reached through control tokens.
- Kept out of general chat rankings; compare it with Granite 4.1 30B for plain chat.
另请参阅 Granite 4.1 30B
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。
IBM Granite Switch 4.1 的其他尺寸
Granite Switch 4.1 30B — 常见问题
How much VRAM does Granite Switch 4.1 30B need?
About 19 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 Granite Switch 4.1 30B run on an RTX 4090 (24 GB)?
Yes. Granite Switch 4.1 30B needs about 19 GB at Q4_K_M, inside a 24 GB card, at an estimated 36 tokens/sec.
How do I run Granite Switch 4.1 30B locally?
No Ollama tag. Use IBM's official GGUF with llama.cpp. Running the published tag would send your prompts to a hosted GPU rather than your own machine.
What other sizes does IBM Granite Switch 4.1 come in?
Granite Switch 4.1 8B (6 GB), Granite Switch 4.1 30B (19 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.