FitLocally by GPUwerk

Phi · 3.8B

Can you run Phi-4 Mini 3.8B locally?

Phi-4 Mini 3.8B is a dense model with 3.8B parameters. At 4-bit or better it runs on 120 of the 120 hardware configurations listed here; the smallest is the RTX 5060 (8 GB). On a 128 GB DGX Spark it runs at Q8_0, at roughly 40 tokens per second (our est.).

Data as of 2 November 2026

Fastest machines for Phi-4 Mini 3.8B

Every machine that runs it, fastest first. Filter by type; your machine is marked. Compare machines running it side by side

  1. 1RTX 5090Graphics card · 32 GB~239 tok/s ModelFit est.Q8_0 · 4.9 of 31 GB
  2. 2RTX PRO 6000 BlackwellGraphics card · 96 GB~239 tok/s ModelFit est.Q8_0 · 4.9 of 95 GB
  3. 32x RTX 5090Graphics card · 64 GB~223 tok/s ModelFit est.Q8_0 · 4.9 of 62 GB
  4. 42x RTX PRO 6000 BlackwellGraphics card · 192 GB~223 tok/s ModelFit est.Q8_0 · 4.9 of 190 GB
  5. 5M5 UltraMac · 96–512 GB · 5 variants~152 tok/s ModelFit est.Q8_0 · 4.9 GB
  6. 6RTX 4090Graphics card · 24 GB~152 tok/s ModelFit est.Q8_0 · 4.9 of 23 GB
  7. 7RTX 5080Graphics card · 16 GB~151 tok/s ModelFit est.Q8_0 · 4.9 of 15.0 GB
  8. 82x RTX 4090Graphics card · 48 GB~145 tok/s ModelFit est.Q8_0 · 4.9 of 46 GB
  9. 9RTX 5070 TiGraphics card · 16 GB~143 tok/s ModelFit est.Q8_0 · 4.9 of 15.0 GB
  10. 10RTX 3090 TiGraphics card · 24 GB~141 tok/s ModelFit est.Q8_0 · 4.9 of 23 GB
  11. 112x RTX 3090 TiGraphics card · 48 GB~136 tok/s ModelFit est.Q8_0 · 4.9 of 46 GB
  12. 12RTX 3090Graphics card · 24 GB~133 tok/s ModelFit est.Q8_0 · 4.9 of 23 GB
  13. 132x RTX 3090Graphics card · 48 GB~128 tok/s ModelFit est.Q8_0 · 4.9 of 46 GB
  14. 14RTX 4080 SUPERGraphics card · 16 GB~117 tok/s ModelFit est.Q8_0 · 4.9 of 15.0 GB
  15. 15RX 7900 XTXGraphics card · 24 GB~116 tok/s ModelFit est.Q8_0 · 4.9 of 23 GB
  16. 16RTX 4080Graphics card · 16 GB~115 tok/s ModelFit est.Q8_0 · 4.9 of 15.0 GB
  17. 17RTX 5070Graphics card · 12 GB~112 tok/s ModelFit est.Q8_0 · 4.9 of 11.0 GB
  18. 182x RX 7900 XTXGraphics card · 48 GB~112 tok/s ModelFit est.Q8_0 · 4.9 of 46 GB
  19. 19RTX 4070 Ti SUPERGraphics card · 16 GB~109 tok/s ModelFit est.Q8_0 · 4.9 of 15.0 GB
  20. 20M3 UltraMac · 96–512 GB · 5 variants~106 tok/s ModelFit est.Q8_0 · 4.9 GB
  21. 21RX 7900 XTGraphics card · 20 GB~102 tok/s ModelFit est.Q8_0 · 4.9 of 19.0 GB
  22. 22M5 MaxMac · 48–128 GB · 3 variants~97 tok/s ModelFit est.Q8_0 · 4.9 GB
  23. 23M2 UltraMac · 64–192 GB · 7 variants~94 tok/s ModelFit est.Q8_0 · 4.9 GB
  24. 24RX 9070 XTGraphics card · 16 GB~89 tok/s ModelFit est.Q8_0 · 4.9 of 15.0 GB
  25. 25RTX 4070 TiGraphics card · 12 GB~85 tok/s ModelFit est.Q8_0 · 4.9 of 11.0 GB
  26. 26RTX 4070 SUPERGraphics card · 12 GB~85 tok/s ModelFit est.Q8_0 · 4.9 of 11.0 GB
  27. 27RTX 4070Graphics card · 12 GB~85 tok/s ModelFit est.Q8_0 · 4.9 of 11.0 GB
  28. 28M1 UltraMac · 64–128 GB · 4 variants~84 tok/s ModelFit est.Q8_0 · 4.9 GB
  29. 29M4 MaxMac · 48–128 GB · 3 variants~81 tok/s ModelFit est.Q8_0 · 4.9 GB
  30. 30RTX 5060Graphics card · 8 GB~79 tok/s ModelFit est.Q8_0 · 4.9 of 7.0 GB
  31. 31RTX 5060 TiGraphics card · 16 GB~79 tok/s ModelFit est.Q8_0 · 4.9 of 15.0 GB
  32. 32M5 Max · 36 GB · 32cMac · 36 GB~75 tok/s ModelFit est.Q8_0 · 4.9 of 24 GB
  33. 33RTX 3070Graphics card · 8 GB~70 tok/s ModelFit est.Q8_0 · 4.9 of 7.0 GB
  34. 34RTX 3060 TiGraphics card · 8 GB~70 tok/s ModelFit est.Q8_0 · 4.9 of 7.0 GB
  35. 35M3 MaxMac · 48–128 GB · 3 variants~64 tok/s ModelFit est.Q8_0 · 4.9 GB
  36. 36M4 Max · 36 GB · 32cMac · 36 GB~63 tok/s ModelFit est.Q8_0 · 4.9 of 24 GB
  37. 37RTX 2070Graphics card · 8 GB~63 tok/s ModelFit est.Q8_0 · 4.9 of 7.0 GB
  38. 38M2 MaxMac · 32–96 GB · 5 variants~59 tok/s ModelFit est.Q8_0 · 4.9 GB
  39. 39RTX 3060Graphics card · 12 GB~58 tok/s ModelFit est.Q8_0 · 4.9 of 11.0 GB
  40. 40M5 ProMac · 24–64 GB · 6 variants~56 tok/s ModelFit est.Q8_0 · 4.9 GB
  41. 41M1 MaxMac · 32–64 GB · 4 variants~54 tok/s ModelFit est.Q8_0 · 4.9 GB
  42. 42RX 9060 XTGraphics card · 16 GB~52 tok/s ModelFit est.Q8_0 · 4.9 of 15.0 GB
  43. 43RTX 4060 TiGraphics card · 16 GB~50 tok/s ModelFit est.Q8_0 · 4.9 of 15.0 GB
  44. 44Tesla P100Graphics card · 16 GB~49 tok/s ModelFit est.Q8_0 · 4.9 of 15.0 GB
  45. 45M3 MaxMac · 36–96 GB · 2 variants~49 tok/s ModelFit est.Q8_0 · 4.9 GB
  46. 46RTX 4060Graphics card · 8 GB~47 tok/s ModelFit est.Q8_0 · 4.9 of 7.0 GB
  47. 47M4 ProMac · 24–64 GB · 6 variants~47 tok/s ModelFit est.Q8_0 · 4.9 GB
  48. 48Tesla T4Graphics card · 16 GB~43 tok/s ModelFit est.Q8_0 · 4.9 of 15.0 GB
  49. 49DGX Spark (GB10)AI mini PC · 128 GB~40 tok/s our est.Q8_0 · 4.9 of 122 GB
  50. 50Ryzen AI Max+ 395AI mini PC · 64–128 GB · 2 variants~37 tok/s our est.Q8_0 · 4.9 GB
  51. 51M2 ProMac · 16–32 GB · 2 variants~34 tok/s ModelFit est.Q8_0 · 4.9 GB
  52. 52M6Mac · 24–32 GB · 3 variants~33 tok/s ModelFit est.Q8_0 · 4.9 GB
  53. 53M1 ProMac · 16–32 GB · 2 variants~31 tok/s ModelFit est.Q8_0 · 4.9 GB
  54. 54M6Mac · 16 GB · 2 variants~30 tok/s ModelFit est.Q8_0 · 4.9 GB
  55. 55M5Mac · 16–32 GB · 6 variants~28 tok/s ModelFit est.Q8_0 · 4.9 GB
  56. 56M3 ProMac · 18–36 GB · 2 variants~27 tok/s ModelFit est.Q8_0 · 4.9 GB
  57. 57M4Mac · 16–32 GB · 3 variants~21 tok/s ModelFit est.Q8_0 · 4.9 GB
  58. 58M3Mac · 8–24 GB · 3 variants~18 tok/s ModelFit est.Q8_0 · 4.9 GB
  59. 59M2Mac · 8–24 GB · 3 variants~17 tok/s ModelFit est.Q8_0 · 4.9 GB
  60. 60M1Mac · 8–16 GB · 2 variants~11 tok/s ModelFit est.Q8_0 · 4.9 GB

Speeds are single-user decode estimates, not measurements. "ModelFit est." comes from ModelFit's model fitted on published runs (CC BY 4.0); "our est." is memory bandwidth divided by bytes read per token, times 0.55. Memory is what the best build needs, against what the machine can give a model. Method

About the model

Parameters
3.8B
Active per token
3.8B
Type
Dense
Good for
Coding, Chat

Builds and how to run them

QuantizationLoadMin. RAMCommand
Q4_K_M3.2 GB6 GBollama run phi4-mini:3.8b
Q8_03.8 GB8 GBollama run phi4-mini:3.8b-q8_0

Load, minimum RAM and Ollama commands: ModelFit, CC BY 4.0. Load is memory with the model loaded, before long contexts.

Questions

Can I run Phi-4 Mini 3.8B on a Mac?

Yes on a Mac with enough memory: the smallest Apple configuration listed here that runs it at 4-bit or better is the M3 · 8 GB.

How much memory does Phi-4 Mini 3.8B need?

About 4.9 GB at the smallest build in our data, including a margin for the runtime and an 8K-token context. Longer contexts need more.

Does Phi-4 Mini 3.8B run on a DGX Spark?

Yes, at Q8_0. The Spark has 128 GB of unified memory, about 121.6 GB usable.