← CanIRunLocal · All hardware · All models
Alibaba · 32.8B parameters · coding.
Ollama weight files: Q4_K_M 20 GB (qwen2.5-coder:32b); Q5_K_M 23 GB (qwen2.5-coder:32b-instruct-q5_K_M); Q8_0 35 GB (qwen2.5-coder:32b-instruct-q8_0).
Yes on 44 machines in this list. The smallest is RTX 4060 Ti 16GB: Q4_K_M, about 22.2 GB, roughly 4 tokens per second, via Ollama. ollama run qwen2.5-coder:32b
ollama run qwen2.5-coder:32b-instruct-q8_0ollama run qwen2.5-coder:32b-instruct-q8_0ollama run qwen2.5-coder:32b-instruct-q8_0ollama run qwen2.5-coder:32b-instruct-q8_0lms get unsloth/Qwen2.5-Coder-32B-Instruct-GGUF@Q8_0llama-server -hf unsloth/Qwen2.5-Coder-32B-Instruct-GGUF:Q8_0 -c 8192vllm serve Qwen/Qwen2.5-Coder-32B-Instruct --max-model-len 8192 --gpu-memory-utilization 0.9| Hardware | Runtime | Quant | Memory | Fit | Speed | Command |
|---|---|---|---|---|---|---|
| RTX 4060 Ti 16GB | Ollama | Q4_K_M | 22.2 GB | partial offload (slower) | ~4 tok/s | ollama run qwen2.5-coder:32b |
| RTX 4070 Ti Super 16GB | Ollama | Q4_K_M | 22.2 GB | partial offload (slower) | ~5 tok/s | ollama run qwen2.5-coder:32b |
| RTX 4080 16GB | Ollama | Q4_K_M | 22.2 GB | partial offload (slower) | ~5 tok/s | ollama run qwen2.5-coder:32b |
| RTX 5080 16GB | Ollama | Q4_K_M | 22.2 GB | partial offload (slower) | ~6 tok/s | ollama run qwen2.5-coder:32b |
| RX 7800 XT 16GB | Ollama | Q4_K_M | 22.2 GB | partial offload (slower) | ~5 tok/s | ollama run qwen2.5-coder:32b |
| RX 9070 XT 16GB | Ollama | Q4_K_M | 22.2 GB | partial offload (slower) | ~5 tok/s | ollama run qwen2.5-coder:32b |
| RTX 3090 24GB | Ollama | Q4_K_M | 22.2 GB | fits in fast memory | ~28 tok/s | ollama run qwen2.5-coder:32b |
| RTX 4090 24GB | Ollama | Q4_K_M | 22.2 GB | fits in fast memory | ~30 tok/s | ollama run qwen2.5-coder:32b |
| RX 7900 XTX 24GB | Ollama | Q4_K_M | 22.2 GB | fits in fast memory | ~29 tok/s | ollama run qwen2.5-coder:32b |
| M1 Pro 32GB | Ollama | Q4_K_M | 22.2 GB | fits in fast memory | ~8 tok/s | ollama run qwen2.5-coder:32b |
| M2 Pro 32GB | Ollama | Q4_K_M | 22.2 GB | fits in fast memory | ~8 tok/s | ollama run qwen2.5-coder:32b |
| M4 32GB | Ollama | Q4_K_M | 22.2 GB | fits in fast memory | ~5 tok/s | ollama run qwen2.5-coder:32b |
| M3 Pro 36GB | Ollama | Q5_K_M | 25.2 GB | fits in fast memory | ~5 tok/s | ollama run qwen2.5-coder:32b-instruct-q5_K_M |
| Mac Studio M4 Max (32c GPU) 36GB | Ollama | Q5_K_M | 25.2 GB | fits in fast memory | ~14 tok/s | ollama run qwen2.5-coder:32b-instruct-q5_K_M |
| Mac Studio M5 Max (32c GPU) 36GB | Ollama | Q5_K_M | 25.2 GB | fits in fast memory | ~16 tok/s | ollama run qwen2.5-coder:32b-instruct-q5_K_M |
| RTX 5090 32GB | Ollama | Q5_K_M | 25.2 GB | fits in fast memory | ~47 tok/s | ollama run qwen2.5-coder:32b-instruct-q5_K_M |
| M4 Pro 48GB | Ollama | Q5_K_M | 25.2 GB | fits in fast memory | ~9 tok/s | ollama run qwen2.5-coder:32b-instruct-q5_K_M |
| Mac Studio M4 Max (40c GPU) 48GB | Ollama | Q5_K_M | 25.2 GB | fits in fast memory | ~19 tok/s | ollama run qwen2.5-coder:32b-instruct-q5_K_M |
| Mac Studio M5 Max (40c GPU) 48GB | Ollama | Q5_K_M | 25.2 GB | fits in fast memory | ~21 tok/s | ollama run qwen2.5-coder:32b-instruct-q5_K_M |
| M1 Max 64GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~9 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| M2 Max 64GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~9 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| M3 Max 64GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~9 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| M4 Max 64GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~12 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| Mac Studio M4 Max (40c GPU) 64GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~12 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| Mac Studio M5 Max (40c GPU) 64GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~14 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| DGX Spark 64GB (ASUS, Dell, HP, MSI, Acer, Gigabyte) | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~6 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| Mac Studio M3 Ultra 96GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~19 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| Mac Studio M5 Ultra 96GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~27 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| M3 Max 128GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~9 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| M4 Max 128GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~12 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| Mac Studio M4 Max (40c GPU) 128GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~12 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| Mac Studio M5 Max (40c GPU) 128GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~14 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| DGX Spark 128GB (NVIDIA, ASUS, Dell, Lenovo, HP, MSI, Acer, Gigabyte) | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~6 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| M2 Ultra 192GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~18 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| 2x Mac Studio M5 Max (40c GPU) 128GB | llama.cpp | Q8_0 GGUF | 37.1 GB | fits in fast memory | ~12 tok/s | llama-server -hf unsloth/Qwen2.5-Coder-32B-Instruct-GGUF:Q8_0 --rpc node2:50052 -c 8192 # run rpc-server on each other node |
| Mac Studio M3 Ultra 256GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~19 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| Mac Studio M5 Ultra (64c GPU) 256GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~27 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| Mac Studio M5 Ultra (80c GPU) 256GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~27 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| 2x DGX Spark 128GB | llama.cpp | Q8_0 GGUF | 37.1 GB | fits in fast memory | ~5 tok/s | llama-server -hf unsloth/Qwen2.5-Coder-32B-Instruct-GGUF:Q8_0 --rpc node2:50052 -c 8192 # run rpc-server on each other node |
| 4x Mac Studio M5 Max (40c GPU) 128GB | llama.cpp | Q8_0 GGUF | 37.1 GB | fits in fast memory | ~11 tok/s | llama-server -hf unsloth/Qwen2.5-Coder-32B-Instruct-GGUF:Q8_0 --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node |
| 2x Mac Studio M5 Ultra (80c GPU) 256GB | llama.cpp | Q8_0 GGUF | 37.1 GB | fits in fast memory | ~23 tok/s | llama-server -hf unsloth/Qwen2.5-Coder-32B-Instruct-GGUF:Q8_0 --rpc node2:50052 -c 8192 # run rpc-server on each other node |
| Mac Studio M3 Ultra (80c GPU) 512GB | Ollama | Q8_0 | 37.2 GB | fits in fast memory | ~19 tok/s | ollama run qwen2.5-coder:32b-instruct-q8_0 |
| 4x DGX Spark 128GB | llama.cpp | Q8_0 GGUF | 37.1 GB | fits in fast memory | ~4 tok/s | llama-server -hf unsloth/Qwen2.5-Coder-32B-Instruct-GGUF:Q8_0 --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 | 37.1 GB | fits in fast memory | ~21 tok/s | llama-server -hf unsloth/Qwen2.5-Coder-32B-Instruct-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