How much VRAM does DeepSeek V4 Flash need?
DeepSeek V4 Flash needs 155 GB for its weights as released (FP4 + FP8), 156 GB with an 8K context and 156 GB with 32K.
- Parameters
- 284B
- Active per token
- 13B
- Longest context
- 1M
- Released
- Jul 2026
- License
- MIT
- Strongest at
- Docs
Memory at each precision
| Precision | Weights | With 8K | With 32K | With 128K |
|---|---|---|---|---|
| As released (FP4 + FP8) | 155 GB | 156 GB | 156 GB | 157 GB |
| 3-bit | 129 GB | 130 GB | 130 GB | 131 GB |
| 2-bit | 99.2 GB | 100 GB | 100 GB | 101 GB |
| 1.6-bit | 54.6 GB | 55.4 GB | 55.6 GB | 56.1 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 | Too big | |||
| RTX 3090 | Too big | |||
| RTX 4090 | Too big | |||
| RTX 5090 | Too big | |||
| 2× RTX 3090 | Too big | |||
| RTX PRO 6000 Blackwell | Too big | |||
| M4 Max 128 GB | Too big | |||
| M5 Max 128 GB | Too big | |||
| Ryzen AI Max+ 395 128 GB | Too big | |||
| DGX Spark 128 GB | 2-bit | 1M | 25 tok/s | 2.8 s |
| M3 Ultra 256 GB | FP4 + FP8 | 1M | 24 tok/s | 7.7 s |
| M5 Ultra 512 GB | FP4 + FP8 | 1M | 33 tok/s | 2.4 s |
| H100 SXM | Too big | |||
| DGX Station GB300 | FP4 + FP8 | 1M | 157 tok/s | 0.2 s |
| 8× H200 NVL | FP4 + FP8 | 1M | 122 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
- llama.cpp
llama-server -hf unsloth/DeepSeek-V4-Flash-0731-GGUF- vLLM
vllm serve deepseek-ai/DeepSeek-V4-Flash-0731- MLX (Mac)
mlx_lm.chat --model mlx-community/DeepSeek-V4-Flash-0731-2.4bit-mixed
Questions
How much VRAM does DeepSeek V4 Flash need?
DeepSeek V4 Flash needs 155 GB for its weights as released (FP4 + FP8), 156 GB with an 8K context and 156 GB with 32K.
Can DeepSeek V4 Flash run on an RTX 4090?
Not on one card. The smallest machine here that runs it is the DGX Spark 128 GB.