← CanIRunLocal · All hardware · All models
DeepSeek · 671B parameters (37B active) · reasoning.
Ollama weight files: Q4_K_M 404 GB (deepseek-r1:671b); Q8_0 713 GB (deepseek-r1:671b-q8_0).
Yes on 4 machines in this list. The smallest is 2x Mac Studio M5 Ultra (80c GPU) 256GB: Q4_K_M, about 380 GB, roughly 32 tokens per second, via MLX. exo # on every Mac (brew install --cask exo); load mlx-community/DeepSeek-R1-0528-4bit with Tensor sharding + RDMA
exo # on every Mac (brew install --cask exo); load mlx-community/DeepSeek-R1-0528-4bit with Tensor sharding + RDMA| Hardware | Runtime | Quant | Memory | Fit | Speed | Command |
|---|---|---|---|---|---|---|
| 2x Mac Studio M5 Ultra (80c GPU) 256GB | MLX | MLX 4bit | 380 GB | fits in fast memory | ~32 tok/s | exo # on every Mac (brew install --cask exo); load mlx-community/DeepSeek-R1-0528-4bit with Tensor sharding + RDMA |
| Mac Studio M3 Ultra (80c GPU) 512GB | LM Studio | MLX 4bit | 380 GB | fits in fast memory | ~16 tok/s | lms get mlx-community/DeepSeek-R1-0528-4bit |
| 4x DGX Spark 128GB | llama.cpp | Q4_K_M GGUF | 407.3 GB | fits in fast memory | ~5 tok/s | llama-server -hf unsloth/DeepSeek-R1-0528-GGUF:Q4_K_M --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node |
| 4x Mac Studio M5 Ultra (80c GPU) 256GB | llama.cpp | Q8_0 GGUF | 715.6 GB | fits in fast memory | ~9 tok/s | llama-server -hf unsloth/DeepSeek-R1-0528-GGUF:Q8_0 --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node |
Estimates, not benchmarks. Model sizes are ollama.com download sizes; memory needed adds headroom for ~8k tokens of context; speeds are rough figures from memory bandwidth. Data updated 2026-10-16.
A project by OnDevice Labs · llms.txt