How to Deploy DeepSeek V3.2 Locally
DeepSeek V3.2 is a MoE model with 671B total parameters and 37B active per token. At Q4 it needs 376GB of VRAM to sit entirely on the GPU; with experts offloaded to system RAM it needs only 12GB of VRAM plus 365GB of RAM. The cheapest card that runs it today is the Mac Studio M3 Ultra 512GB at about $19,315, at roughly 19.4 tokens/s. In offload mode the cheapest card that runs it is the Tesla V100 16GB, at roughly 4 tokens/s.
DeepSeek V3.2 details
How much VRAM does DeepSeek V3.2 need for local deployment?
MoE: all 671B parameters must stay resident in memory, while speed is set by the 37B active per token.
| Quantization | WeightsGB | Total at 8KGB | Total at 32KGB | Total at 128KGB |
|---|---|---|---|---|
| 1-bit | 157.5est. | 159 | 160.6 | 167.1 |
| 2-bit | 249.3est. | 250.9 | 252.5 | 258.9 |
| 3-bit | 321.5est. | 323.1 | 324.7 | 331.1 |
| 4-bit | 374est. | 375.5 | 377.2 | 383.6 |
| 5-bit | 452.8est. | 454.3 | 455.9 | 462.3 |
| 8-bit | 695.5est. | 697.1 | 698.7 | 705.1 |
| FP16 | 1,312.3est. | 1,313.9 | 1,315.5 | 1,321.9 |
Total = weights + KV cache (FP16) + 1GB runtime overhead. Weights marked GGUF are measured file sizes from the repository; the rest are estimated from bytes per parameter. Contexts beyond this model's 160K limit show a dash. Each tier takes whichever file of that bit width the repository has, preferring unsloth dynamic quants, so the exact file name varies by model.
Which GPUs can run DeepSeek V3.2?
Xianyu
| Device | VRAMGB | Price | Est. t/s | Suggested context |
|---|---|---|---|---|
| Mac Studio M3 Ultra 512GB | 512 | $19,3156 listings | 19.4 | 128K |
Fits when weights + KV cache + runtime overhead is at or below usable memory, single card. "Offload" on MoE models means it runs with expert weights in system RAM (only attention layers and KV cache stay in VRAM); that speed assumes 70 GB/s RAM bandwidth. The context column is the largest tier that fits.
Recommended GPUs for DeepSeek V3.2 with CPU/GPU offloading
Xianyu
Expert weights live in system RAM (needs at least 365 GB); VRAM holds only attention layers, shared experts and the KV cache, at least 12 GB at Q4. Speed is bound by RAM bandwidth (estimated at 70 GB/s) and is far slower than a full-VRAM setup.
Deploy DeepSeek V3.2 locally with Ollama, llama.cpp or vLLM
The Ollama library has no official tag for it yet.
vLLM serves the original-precision weights, which needs far more memory than GGUF and usually more than one GPU.
vllm serve deepseek-ai/DeepSeek-V3.2
File names follow that week's model-library snapshot; the definitions are in the methodology. Methodology
FAQ
How much VRAM does DeepSeek V3.2 need?
At Q4_K_M with an 8K context DeepSeek V3.2 needs about 376GB of VRAM; 452GB leaves comfortable headroom.
What is the cheapest GPU that runs DeepSeek V3.2?
The Mac Studio M3 Ultra 512GB: 512GB of VRAM, currently about $19,315, at roughly 19.4 tokens/s.
Can you run DeepSeek V3.2 with Ollama?
Not yet: there is no official Ollama tag and no public GGUF build for it.