DeepSeek V3 0324
671.2B total, 37.7B active · 160k context · MIT License · released 2025-03-24
chatcodingreasoningagentic
In short
DeepSeek V3 0324 is an open model from DeepSeek with 671.2B parameters (37.7B active per token). At Q4_K_M, DeepSeek V3 0324 needs about 378 GB of memory at 8k context and 382 GB at 64k, counting the weights, the KV cache and runtime buffers. Nothing we price in the US runs DeepSeek V3 0324 fully in memory at 64k context at a usable speed; see the table for what comes closest.
Updated . Speeds are estimates; how they are worked out.
What to run it on in the US
Fully in memory at 64k context, priced at retailers today.
Something we price holds it fully in memory, but none reaches 15 tok/s, our bar for recommending it. The table below has the speeds.
All hardware that runs it
Everything we price, best value first. Two cards split the model by layers.
| Hardware | Quant | Fits | Decode | Price | tok/s per 1,000 |
|---|---|---|---|---|---|
| Mac Studio M3 Ultra 512GB | Q4_K_M | fits | 12 tok/s | $23,000 | 0.5 |
39 other cards and machines we price cannot hold it at this context.
Memory needed
Weights, KV cache and runtime buffers at each context length (KV cache at fp16).
| Quant | File | @8k | @32k | @64k | @128k |
|---|---|---|---|---|---|
| Q4_K_Mgguf | 376.7 GB | 377.9 GB | 379.7 GB | 382.0 GB | 386.7 GB |
| mlx-5bitmlx | 429.8 GB | 431.1 GB | 432.8 GB | 435.1 GB | 439.8 GB |
| Q5_K_Mgguf | 442.7 GB | 444.0 GB | 445.8 GB | 448.1 GB | 452.8 GB |
| Q6_Kgguf | 513.0 GB | 514.3 GB | 516.0 GB | 518.3 GB | 523.0 GB |
| Q8_0gguf | 664.3 GB | 665.6 GB | 667.3 GB | 669.7 GB | 674.3 GB |
Run it
Copy a command for the runtime you use.
- Q4_K_M
- Ollama
ollama run hf.co/unsloth/DeepSeek-V3-0324-GGUF:Q4_K_M - LM Studio
lms get unsloth/DeepSeek-V3-0324-GGUF@q4_k_m - llama.cpp
llama-server -hf unsloth/DeepSeek-V3-0324-GGUF:Q4_K_M
- Ollama
- mlx-5bit
- LM Studio
lms get mlx-community/DeepSeek-V3-0324-5bit - MLX
mlx_lm.server --model mlx-community/DeepSeek-V3-0324-5bit
- LM Studio
- Q5_K_M
- Ollama
ollama run hf.co/unsloth/DeepSeek-V3-0324-GGUF:Q5_K_M - LM Studio
lms get unsloth/DeepSeek-V3-0324-GGUF@q5_k_m - llama.cpp
llama-server -hf unsloth/DeepSeek-V3-0324-GGUF:Q5_K_M
- Ollama
- Q6_K
- Ollama
ollama run hf.co/unsloth/DeepSeek-V3-0324-GGUF:Q6_K - LM Studio
lms get unsloth/DeepSeek-V3-0324-GGUF@q6_k - llama.cpp
llama-server -hf unsloth/DeepSeek-V3-0324-GGUF:Q6_K
- Ollama
- Q8_0
- Ollama
ollama run hf.co/unsloth/DeepSeek-V3-0324-GGUF:Q8_0 - LM Studio
lms get unsloth/DeepSeek-V3-0324-GGUF@q8_0 - llama.cpp
llama-server -hf unsloth/DeepSeek-V3-0324-GGUF:Q8_0
- Ollama
Notes
Same architecture as DeepSeek-V3. The card says Hugging Face Transformers does not directly support it yet. Excludes 13.4B parameters of multi-token-prediction layers, which local runtimes do not load.
Questions
- How much VRAM does DeepSeek V3 0324 need?
- At Q4_K_M, DeepSeek V3 0324 needs about 378 GB of memory at 8k context and 382 GB at 64k, counting the weights, the KV cache and runtime buffers.
- Can I run DeepSeek V3 0324 on a 16 GB GPU?
- No: it does not fit in 16 GB at 64k context, even with system RAM.
- Can I run DeepSeek V3 0324 on a 24 GB GPU?
- No: it does not fit in 24 GB at 64k context, even with system RAM.
- How do I run DeepSeek V3 0324 locally?
- With Ollama: ollama run hf.co/unsloth/DeepSeek-V3-0324-GGUF:Q4_K_M (the Q4_K_M build). LM Studio and llama.cpp commands are on this page.
- What licence is DeepSeek V3 0324 released under?
- MIT License, which allows commercial use.