How much VRAM does Qwen3.8 27B need?
Qwen3.8 27B needs 15.7 GB for its weights at 4-bit, 17.4 GB with an 8K context and 18.9 GB with 32K. At 8-bit, 29.2 GB.
- Parameters
- 27.8B
- Longest context
- 256K
- Released
- Aug 2026
- License
- Apache 2.0
- Strongest at
- Chat, Coding, Agents, Docs
Memory at each precision
| Precision | Weights | With 8K | With 32K | With 128K |
|---|---|---|---|---|
| 16-bit | 51.7 GB | 53.5 GB | 55.0 GB | 61.0 GB |
| 8-bit | 27.5 GB | 29.2 GB | 30.7 GB | 36.7 GB |
| 6-bit | 21.2 GB | 22.9 GB | 24.4 GB | 30.4 GB |
| 5-bit | 18.4 GB | 20.1 GB | 21.6 GB | 27.6 GB |
| 4-bit | 15.7 GB | 17.4 GB | 18.9 GB | 24.9 GB |
| 3-bit | 12.6 GB | 14.4 GB | 15.9 GB | 21.9 GB |
| 2-bit | 9.7 GB | 11.4 GB | 12.9 GB | 18.9 GB |
| 1.6-bit | 5.3 GB | 7.1 GB | 8.6 GB | 14.6 GB |
One conversation, 16-bit context cache, runtime overhead included.
What runs it
| Machine | Best fit | Context | Speed | First reply |
|---|---|---|---|---|
| RTX 3060 12 GB | Too big | |||
| RTX 4060 Ti 16 GB | 3-bit | 20K | 15 tok/s | 2.5 s |
| RTX 3090 | 5-bit | 48K | 32 tok/s | 1.6 s |
| RTX 4090 | 5-bit | 48K | 35 tok/s | 0.7 s |
| RTX 5090 | 8-bit | 24K | 41 tok/s | 0.5 s |
| 2× RTX 3090 | 8-bit | 256K | 22 tok/s | 0.9 s |
| RTX PRO 6000 Blackwell | 8-bit | 256K | 41 tok/s | 0.4 s |
| M4 Max 128 GB | 8-bit | 256K | 12 tok/s | 8.9 s |
| M5 Max 128 GB | 8-bit | 256K | 13 tok/s | 2.6 s |
| Ryzen AI Max+ 395 128 GB | 8-bit | 256K | 5.3 tok/s | 5.1 s |
| DGX Spark 128 GB | 8-bit | 256K | 6.7 tok/s | 1.8 s |
| M3 Ultra 256 GB | 8-bit | 256K | 17 tok/s | 5.0 s |
| M5 Ultra 512 GB | 8-bit | 256K | 24 tok/s | 1.6 s |
| H100 SXM | 8-bit | 256K | 71 tok/s | 0.4 s |
| DGX Station GB300 | 8-bit | 256K | 131 tok/s | 0.1 s |
| 8× H200 NVL | 8-bit | 256K | 97 tok/s | 0.1 s |
Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.
How to run it
- Ollama
ollama run qwen3.8:27b- llama.cpp
llama-server -hf unsloth/Qwen3.8-27B-GGUF- vLLM
vllm serve Qwen/Qwen3.8-27B- MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen3.8-27B-4bit
Questions
How much VRAM does Qwen3.8 27B need?
Qwen3.8 27B needs 15.7 GB for its weights at 4-bit, 17.4 GB with an 8K context and 18.9 GB with 32K. At 8-bit, 29.2 GB.
Can Qwen3.8 27B run on an RTX 4090?
Yes. At 5-bit it holds 48K of context and writes about 35 tokens a second.