readygpu

DeepSeek R1 0528 Qwen3 8B

8.2B parameters · 128k context · mit · released 2025-05-29

chatreasoning

What to run it on in the US

Fully in memory at 8k 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.

HardwareQuantFitsDecodePricetok/s per 1,000
NVIDIA GeForce RTX 3090Q8_0fits59 tok/s––
NVIDIA GeForce RTX 5090Q8_0fits112 tok/s––
Mac Studio M4 Max 128GBQ8_0fits36 tok/s––
NVIDIA GeForce RTX 4090Q8_0fits63 tok/s––
NVIDIA GeForce RTX 5070 TiQ8_0fits56 tok/s––
Mac mini M4 Pro 64GBQ8_0fits18 tok/s––
NVIDIA GeForce RTX 5060 Ti 16GBQ8_0fits28 tok/s––
AMD Radeon RX 7900 XTXQ8_0fits53 tok/s––
Framework Desktop Ryzen AI Max+ 395 128GBQ8_0fits14 tok/s––
Mac Studio M3 Ultra 512GBQ8_0fits54 tok/s––
NVIDIA DGX Spark 128GBQ8_0fits17 tok/s––
NVIDIA GeForce RTX 3060 12GBQ8_0fits23 tok/s––
AMD Radeon RX 9070 XTQ8_0fits36 tok/s––
GMKtec EVO-X2 Ryzen AI Max+ 395 128GBQ8_0fits14 tok/s––
Mac Studio M3 Ultra 256GBQ8_0fits54 tok/s––
Mac mini M4 Pro 48GBQ8_0fits18 tok/s––
NVIDIA GeForce RTX 4060 Ti 16GBQ8_0fits18 tok/s––
NVIDIA GeForce RTX 5080Q8_0fits60 tok/s––
Mac Studio M4 Max 64GBQ8_0fits36 tok/s––
NVIDIA GeForce RTX 4070 Ti SuperQ8_0fits42 tok/s––
NVIDIA GeForce RTX 5070Q8_0fits42 tok/s––
AMD Radeon RX 9060 XT 16GBQ8_0fits18 tok/s––
Mac Studio M3 Ultra 96GBQ8_0fits54 tok/s––
NVIDIA GeForce RTX 4070 SuperQ8_0fits32 tok/s––
NVIDIA GeForce RTX 4080 SuperQ8_0fits46 tok/s––
NVIDIA RTX PRO 6000 BlackwellQ8_0fits112 tok/s––
AMD Radeon AI PRO R9700Q8_0fits36 tok/s––
AMD Radeon RX 7800 XTQ8_0fits35 tok/s––
AMD Radeon RX 7900 XTQ8_0fits44 tok/s––
ASUS Ascent GX10 128GBQ8_0fits17 tok/s––
Framework Desktop Ryzen AI Max+ 395 64GBQ8_0fits14 tok/s––
Intel Arc B580Q8_0fits23 tok/s––
Mac mini M4 32GBQ8_0fits8 tok/s––
NVIDIA GeForce RTX 3090 TiQ8_0fits63 tok/s––
AMD Radeon RX 7600 XTQ8_0fits16 tok/s––
Mac mini M4 16GBQ8_0fits8 tok/s––
NVIDIA GeForce RTX 3080 10GBQ6_Kfits59 tok/s––
Intel Arc Pro B60Q8_0fits23 tok/s––
NVIDIA Tesla P40Q8_0fits22 tok/s––
Intel Arc A770 16GBQ8_0fits28 tok/s––

Memory needed

Weights, KV cache and runtime buffers at each context length (KV cache at fp16).

QuantFile@8k@32k@64k@128k
Q4_K_Mgguf4.7 GB6.6 GB10.1 GB14.8 GB24.1 GB
mlx-4bitmlx4.3 GB6.2 GB9.7 GB14.4 GB23.7 GB
Q5_K_Mgguf5.4 GB7.3 GB10.8 GB15.5 GB24.9 GB
Q6_Kgguf6.3 GB8.1 GB11.7 GB16.3 GB25.7 GB
mlx-8bitmlx8.1 GB10.0 GB13.5 GB18.2 GB27.6 GB
Q8_0gguf8.1 GB10.0 GB13.5 GB18.2 GB27.6 GB

Run it

Copy a command for the runtime you use.

  • Q4_K_M
    • Ollamaollama run hf.co/unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q4_K_M
    • LM Studiolms get unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF@q4_k_m
    • llama.cppllama-server -hf unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q4_K_M
  • mlx-4bit
    • LM Studiolms get mlx-community/DeepSeek-R1-0528-Qwen3-8B-4bit
    • MLXmlx_lm.server --model mlx-community/DeepSeek-R1-0528-Qwen3-8B-4bit
  • Q5_K_M
    • Ollamaollama run hf.co/unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q5_K_M
    • LM Studiolms get unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF@q5_k_m
    • llama.cppllama-server -hf unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q5_K_M
  • Q6_K
    • Ollamaollama run hf.co/unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q6_K
    • LM Studiolms get unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF@q6_k
    • llama.cppllama-server -hf unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q6_K
  • mlx-8bit
    • LM Studiolms get mlx-community/DeepSeek-R1-0528-Qwen3-8B-8bit
    • MLXmlx_lm.server --model mlx-community/DeepSeek-R1-0528-Qwen3-8B-8bit
  • Q8_0
    • Ollamaollama run hf.co/unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q8_0
    • LM Studiolms get unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF@q8_0
    • llama.cppllama-server -hf unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q8_0