readygpu

Magistral Small 2506

23.6B parameters · 40k context · Apache 2.0 · released 2025-06-04

reasoningchat

In short

The cheapest hardware in Canada that runs Magistral Small 2506 fully in memory at 128k context is the NVIDIA RTX PRO 6000 Blackwell, $3,133 at Amazon, at about 24 tok/s (estimated).

Updated . Speeds are estimates; how they are worked out.

What to run it on in Canada

Fully in memory at 128k context, priced at retailers today.

All hardware that runs it

Everything we price, best value first. Two cards split the model by layers.

HardwareQuantFitsDecodePricetok/s per 1,000
NVIDIA RTX PRO 6000 BlackwellQ8_0fits24 tok/s$3,1337.6
GMKtec EVO-X2 Ryzen AI Max+ 395 128GBQ8_0fits3 tok/s$5,0990.6
ASUS Ascent GX10 128GBQ8_0fits4 tok/s$8,2990.4
Mac Studio M4 Max 128GBQ8_0fits8 tok/s––
Mac mini M4 Pro 64GBQ8_0fits4 tok/s––
Framework Desktop Ryzen AI Max+ 395 128GBQ8_0fits3 tok/s––
Mac Studio M3 Ultra 512GBQ8_0fits11 tok/s––
NVIDIA DGX Spark 128GBQ8_0fits4 tok/s––
Mac Studio M3 Ultra 256GBQ8_0fits11 tok/s––
Mac mini M4 Pro 48GBQ4_K_Mfits5 tok/s––
Mac Studio M4 Max 64GBQ8_0fits8 tok/s––
Mac Studio M3 Ultra 96GBQ8_0fits11 tok/s––
Framework Desktop Ryzen AI Max+ 395 64GBQ8_0fits3 tok/s––
NVIDIA GeForce RTX 3060 12GBQ4_K_Moffload2 tok/s$4973.8
Intel Arc B580Q4_K_Moffload2 tok/s$5603.4
Intel Arc A770 16GBQ4_K_Moffload2 tok/s$6553.4
NVIDIA GeForce RTX 3080 10GBQ4_K_Moffload2 tok/s$7692.5
AMD Radeon RX 9060 XT 16GBQ4_K_Moffload2 tok/s$8002.6
AMD Radeon RX 9070 XTQ4_K_Moffload2 tok/s$1,0502.2
Intel Arc Pro B60Q4_K_Moffload3 tok/s$1,0602.8
NVIDIA GeForce RTX 5060 Ti 16GBQ4_K_Moffload2 tok/s$1,1202.1
NVIDIA GeForce RTX 5070Q4_K_Moffload2 tok/s$1,1811.7
AMD Radeon RX 7800 XTQ4_K_Moffload2 tok/s$1,3971.6
NVIDIA GeForce RTX 4070 SuperQ4_K_Moffload2 tok/s$1,4001.4
AMD Radeon RX 7600 XTQ4_K_Moffload2 tok/s$1,4401.5
NVIDIA GeForce RTX 5070 TiQ4_K_Moffload2 tok/s$1,6901.4
NVIDIA GeForce RTX 4070 Ti SuperQ4_K_Moffload2 tok/s$1,7901.3
NVIDIA GeForce RTX 4080 SuperQ4_K_Moffload2 tok/s$2,0001.2
AMD Radeon RX 7900 XTQ4_K_Moffload3 tok/s$2,0831.3
NVIDIA GeForce RTX 5080Q4_K_Moffload2 tok/s$2,3701.0
NVIDIA GeForce RTX 3090Q4_K_Moffload4 tok/s$2,8001.3
AMD Radeon AI PRO R9700Q4_K_Moffload6 tok/s$2,8002.2
NVIDIA GeForce RTX 3090 TiQ4_K_Moffload4 tok/s$3,9991.0
NVIDIA GeForce RTX 4090Q4_K_Moffload4 tok/s$7,1070.5
NVIDIA GeForce RTX 5090Q4_K_Moffload10 tok/s$10,3001.0
AMD Radeon RX 7900 XTXQ4_K_Moffload4 tok/s––
NVIDIA GeForce RTX 4060 Ti 16GBQ4_K_Moffload2 tok/s––
NVIDIA Tesla P40Q4_K_Moffload3 tok/s––

2 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).

QuantFile@8k@32k
Q4_K_Mgguf13.3 GB15.3 GB19.2 GB
Q5_K_Mgguf15.6 GB17.6 GB21.5 GB
Q6_Kgguf18.0 GB20.0 GB23.9 GB
Q8_0gguf23.3 GB25.3 GB29.2 GB

Run it

Copy a command for the runtime you use.

  • Q4_K_M
    • Ollamaollama run hf.co/unsloth/Magistral-Small-2506-GGUF:Q4_K_M
    • LM Studiolms get unsloth/Magistral-Small-2506-GGUF@q4_k_m
    • llama.cppllama-server -hf unsloth/Magistral-Small-2506-GGUF:Q4_K_M
  • Q5_K_M
    • Ollamaollama run hf.co/unsloth/Magistral-Small-2506-GGUF:Q5_K_M
    • LM Studiolms get unsloth/Magistral-Small-2506-GGUF@q5_k_m
    • llama.cppllama-server -hf unsloth/Magistral-Small-2506-GGUF:Q5_K_M
  • Q6_K
    • Ollamaollama run hf.co/unsloth/Magistral-Small-2506-GGUF:Q6_K
    • LM Studiolms get unsloth/Magistral-Small-2506-GGUF@q6_k
    • llama.cppllama-server -hf unsloth/Magistral-Small-2506-GGUF:Q6_K
  • Q8_0
    • Ollamaollama run hf.co/unsloth/Magistral-Small-2506-GGUF:Q8_0
    • LM Studiolms get unsloth/Magistral-Small-2506-GGUF@q8_0
    • llama.cppllama-server -hf unsloth/Magistral-Small-2506-GGUF:Q8_0

Notes

Performance may degrade past 40k tokens, so the card recommends setting max model length to 40k. Needs a recent vLLM with mistral_common >= 1.6.0.

Questions

What is the best GPU for running Magistral Small 2506 locally?
The cheapest hardware in Canada that runs Magistral Small 2506 fully in memory at 128k context is the NVIDIA RTX PRO 6000 Blackwell, $3,133 at Amazon, at about 24 tok/s (estimated).
Can I run Magistral Small 2506 on a 16 GB GPU?
Only partly: at 128k context it does not fit in 16 GB, so part of it runs from system RAM and it slows to about 2 tok/s on the Intel Arc A770 16GB (estimated).
Can I run Magistral Small 2506 on a 24 GB GPU?
Only partly: at 128k context it does not fit in 24 GB, so part of it runs from system RAM and it slows to about 3 tok/s on the Intel Arc Pro B60 (estimated).
How do I run Magistral Small 2506 locally?
With Ollama: ollama run hf.co/unsloth/Magistral-Small-2506-GGUF:Q4_K_M (the Q4_K_M build). LM Studio and llama.cpp commands are on this page.
What licence is Magistral Small 2506 released under?
Apache 2.0, which allows commercial use.