FitLocally by GPUwerk

NVIDIA · 49B

Can you run Llama 3.3 Nemotron Super 49B locally?

Llama 3.3 Nemotron Super 49B is a dense model with 49B parameters. At 4-bit or better it runs on 54 of the 120 hardware configurations listed here; the smallest is the 2x RTX 4090 (48 GB). On a 128 GB DGX Spark it runs at Q4_K_M, at roughly 5.1 tokens per second (our est.).

Data as of 2 November 2026

Fastest machines for Llama 3.3 Nemotron Super 49B

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~41 tok/s our est.Q3_K_M · 25 of 31 GB 3-bit only
  2. 2RTX 4090Graphics card · 24 GB~30 tok/s our est.Q2_K · 19.8 of 23 GB 2-bit only
  3. 3RTX 3090 TiGraphics card · 24 GB~30 tok/s our est.Q2_K · 19.8 of 23 GB 2-bit only
  4. 4RX 7900 XTXGraphics card · 24 GB~29 tok/s our est.Q2_K · 19.8 of 23 GB 2-bit only
  5. 5RTX 3090Graphics card · 24 GB~28 tok/s our est.Q2_K · 19.8 of 23 GB 2-bit only
  6. 62x RTX 5090Graphics card · 64 GB~19 tok/s our est.Q8_0 · 54 of 62 GB
  7. 7RTX PRO 6000 BlackwellGraphics card · 96 GB~19 tok/s our est.Q8_0 · 54 of 95 GB
  8. 82x RTX 4090Graphics card · 48 GB~14 tok/s our est.Q6_K · 42 of 46 GB
  9. 92x RTX 3090 TiGraphics card · 48 GB~14 tok/s our est.Q6_K · 42 of 46 GB
  10. 102x RX 7900 XTXGraphics card · 48 GB~13 tok/s our est.Q6_K · 42 of 46 GB
  11. 112x RTX 3090Graphics card · 48 GB~13 tok/s our est.Q6_K · 42 of 46 GB
  12. 12M4 Max · 36 GB · 32cMac · 36 GB~12 tok/s our est.Q2_K · 19.8 of 24 GB 2-bit only
  13. 13M2 MaxMac · 32 GB · 2 variants~12 tok/s our est.Q2_K · 19.8 GB 2-bit only
  14. 14M1 MaxMac · 32 GB · 2 variants~12 tok/s our est.Q2_K · 19.8 GB 2-bit only
  15. 15M3 UltraMac · 96–512 GB · 5 variants~11 tok/s our est.Q6_K · 42 GB
  16. 16M2 UltraMac · 64–192 GB · 7 variants~11 tok/s our est.Q6_K · 42 GB
  17. 17M1 UltraMac · 64–128 GB · 4 variants~11 tok/s our est.Q6_K · 42 GB
  18. 18M4 MaxMac · 48–128 GB · 3 variants~10 tok/s our est.Q4_K_M · 31 GB
  19. 192x RTX PRO 6000 BlackwellGraphics card · 192 GB~10 tok/s our est.FP16 · 101 of 190 GB
  20. 20M3 Max · 36 GB · 30cMac · 36 GB~9.0 tok/s slow our est.Q2_K · 19.8 of 24 GB 2-bit only
  21. 21M3 MaxMac · 48–128 GB · 3 variants~7.4 tok/s slow our est.Q4_K_M · 31 GB
  22. 22M2 MaxMac · 64–96 GB · 3 variants~7.4 tok/s slow our est.Q4_K_M · 31 GB
  23. 23M1 MaxMac · 64 GB · 2 variants~7.4 tok/s slow our est.Q4_K_M · 31 GB
  24. 24M2 Pro · 32 GBMac · 32 GB~6.0 tok/s slow our est.Q2_K · 19.8 of 21 GB 2-bit only
  25. 25M1 Pro · 32 GBMac · 32 GB~6.0 tok/s slow our est.Q2_K · 19.8 of 21 GB 2-bit only
  26. 26M3 Max · 96 GB · 30cMac · 96 GB~5.6 tok/s slow our est.Q4_K_M · 31 of 72 GB
  27. 27M4 ProMac · 48–64 GB · 4 variants~5.1 tok/s slow our est.Q4_K_M · 31 GB
  28. 28DGX Spark (GB10)AI mini PC · 128 GB~5.1 tok/s slow our est.Q4_K_M · 31 of 122 GB
  29. 29Ryzen AI Max+ 395AI mini PC · 64–128 GB · 2 variants~4.7 tok/s slow our est.Q4_K_M · 31 GB
  30. 30M5Mac · 32 GB · 2 variants~4.6 tok/s slow our est.Q2_K · 19.8 GB 2-bit only
  31. 31M3 Pro · 36 GBMac · 36 GB~4.5 tok/s slow our est.Q2_K · 19.8 of 24 GB 2-bit only
  32. 32M4 · 32 GBMac · 32 GB~3.6 tok/s slow our est.Q2_K · 19.8 of 21 GB 2-bit only
  33. 33M6 · 32 GB · 12cMac · 32 GBno estimateQ2_K · 19.8 of 21 GB 2-bit only
  34. 34M5 Max · 36 GB · 32cMac · 36 GBno estimateQ2_K · 19.8 of 24 GB 2-bit only
  35. 35M5 MaxMac · 48–128 GB · 3 variantsno estimateQ4_K_M · 31 GB
  36. 36M5 ProMac · 48–64 GB · 4 variantsno estimateQ4_K_M · 31 GB
  37. 37M5 UltraMac · 96–512 GB · 5 variantsno estimateQ4_K_M · 31 GB
Too big for 47 machines

RX 7900 XT, M5 Pro · 24 GB · 20c, M5 Pro · 24 GB · 16c, M4 Pro · 24 GB · 20c, M4 Pro · 24 GB · 16c, M6 · 24 GB · 12c, M6 · 24 GB · 10c, M5 · 24 GB · 10c, M5 · 24 GB · 8c, M4 · 24 GB, M3 · 24 GB, M2 · 24 GB, RTX 5080, RTX 5070 Ti, RTX 5060 Ti, RTX 4080 SUPER, RTX 4080, RTX 4070 Ti SUPER, RTX 4060 Ti, Tesla T4, Tesla P100, RX 9070 XT, RX 9060 XT, M3 Pro · 18 GB, RTX 5070, RTX 4070 Ti, RTX 4070 SUPER, RTX 4070, RTX 3060, M2 Pro · 16 GB, M1 Pro · 16 GB, M6 · 16 GB · 12c, M6 · 16 GB · 10c, M5 · 16 GB · 10c, M5 · 16 GB · 8c, M4 · 16 GB, M3 · 16 GB, M2 · 16 GB, M1 · 16 GB, RTX 5060, RTX 4060, RTX 3070, RTX 3060 Ti, RTX 2070, M3 · 8 GB, M2 · 8 GB, M1 · 8 GB. Llama 3.3 Nemotron Super 49B needs about 31 GB at its smallest build.

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

DGX Spark setup notes: GPUwerk guide for Llama 3.3 Nemotron Super 49B.

About the model

Parameters
49B
Active per token
49B
Type
Dense
Good for
Dense model for batch work
Model card
Hugging Face
Hugging Face count
49.87B (agrees)

Questions

Can I run Llama 3.3 Nemotron Super 49B 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 M5 Max · 48 GB · 40c.

How much memory does Llama 3.3 Nemotron Super 49B need?

About 19.8 GB at the smallest build in our data, including a margin for the runtime and an 8K-token context. Longer contexts need more.

Does Llama 3.3 Nemotron Super 49B run on a DGX Spark?

Yes, at Q4_K_M. The Spark has 128 GB of unified memory, about 121.6 GB usable.