nvidia · text · mixture of experts
NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-BF16
nvidia/NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-BF16NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-BF16 at Q4_K_M is exactly 51,601,807,488 bytes (48.06 GiB / 51.60 GB) — an effective 5.478 bits per weight, not the nominal 4.
From the file· summed from 1 file(s)From the file· KV per layer
Parameters
75.4B
total, not active
Architecture
nemotron_h_moe
null layers
Context
262,144
native (config.json)
License
other
Shipped quantizations
● exact bytes, summed from published files
| Quant | Size● | Exact bytes● | Effective bpw● | Tensors● | Publisher |
|---|---|---|---|---|---|
| Q2_K | 29.30 GiB | 31,461,152,896 | 3.340 | — | RemySkye |
| Q3_K_S | 33.22 GiB | 35,670,087,808 | 3.787 | — | RemySkye |
| Q3_K_M | 37.78 GiB | 40,567,117,952 | 4.307 | — | RemySkye |
| IQ4_XS | 40.00 GiB | 42,954,905,728 | 4.560 | — | RemySkye |
| Q3_K_L | 41.18 GiB | 44,217,473,152 | 4.694 | — | RemySkye |
| Q4_0 | 41.43 GiB | 44,488,284,288 | 4.723 | — | RemySkye |
| Q4_K_S | 43.15 GiB | 46,331,287,680 | 4.919 | — | RemySkye |
| Q4_1 | 45.95 GiB | 49,342,666,880 | 5.239 | — | RemySkye |
| Q4_K_M | 48.06 GiB | 51,601,807,488 | 5.478 | — | RemySkye |
| Q5_0 | 50.47 GiB | 54,197,049,472 | 5.754 | — | RemySkye |
| Q5_K_S | 51.13 GiB | 54,901,692,544 | 5.829 | — | RemySkye |
| Q5_K_M | 54.65 GiB | 58,681,006,208 | 6.230 | — | RemySkye |
| Q5_1 | 55.00 GiB | 59,051,432,064 | 6.269 | — | RemySkye |
| Q6_K | 62.63 GiB | 67,243,104,384 | 7.139 | — | RemySkye |
| Q8_0 | 77.72 GiB | 83,453,368,448 | 8.860 | — | RemySkye |
| BF16 | 146.01 GiB | 156,772,423,808 | 16.644 | — | RemySkye |
Compare with
same modality, comparable size
Will it run on your card?
full quant x context sweep
Radeon RX 6500 XT 4GBGeForce RTX 3050 6GBGeForce RTX 5050 8GBGeForce RTX 3080 10GBGeForce RTX 2080 Ti 11GBGeForce RTX 5070 12GBGeForce RTX 5060 Ti 16GBApple M3 Pro 18GBGeForce RTX 3080 Ti 20GBGeForce RTX 5090 D V2 24GBGeForce RTX 5090 D 32GBApple M5 Max 36GBApple M5 Pro 48GBApple M5 Max 64GBApple M3 Ultra 96GBApple M5 Max 128GBApple M2 Ultra 192GBApple M3 Ultra 256GBApple M3 Ultra 512GB
Why other calculators give a different number
A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 39.48 GiB. The real file is 48.06 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.
Architecture
from config.json
Layers
—
Attention heads
32
KV heads
2
Head dim
128
Hidden size
4096
Vocab
131,072
Sliding window
none
SWA period
—
MLA
no
Experts
512
Experts per token
1
use_sliding_window
—
Questions people ask
- How much VRAM does NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-BF16 need?
- Q4_K_M is exactly 51,601,807,488 bytes (48.06 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
- Is NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-BF16 a mixture-of-experts model?
- Yes — 512 experts, 1 routed per token. Every expert must be resident, but only the routed ones are read per token, which is why its memory requirement and its speed behave very differently.
- Which quantization of NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-BF16 should I use?
- Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.