nvidia · text · mixture of experts

NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16

nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16

NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 at UD-IQ1_M is exactly 188,031,516,192 bytes (175.12 GiB / 188.03 GB) — an effective 2.684 bits per weight, not the nominal 1.

From the file· summed from 5 file(s)From the file· KV per layer
Parameters
561B
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
QuantSize●Exact bytes●Effective bpw●Tensors●Publisher
UD-IQ1_M5 shards175.12 GiB188,031,516,1922.684—unsloth
UD-IQ2_XXS5 shards180.51 GiB193,819,655,6802.766—unsloth
UD-IQ2_M5 shards180.64 GiB193,960,558,0802.768—unsloth
UD-IQ3_XXS6 shards207.99 GiB223,323,569,8243.187—unsloth
UD-IQ3_S7 shards233.30 GiB250,502,659,8403.575—unsloth
UD-Q3_K_M7 shards255.21 GiB274,030,739,1683.911—unsloth
UD-IQ4_XS7 shards266.47 GiB286,118,723,2964.084—unsloth
UD-IQ4_NL8 shards273.97 GiB294,171,787,1684.199—unsloth
UD-Q4_K_S8 shards305.18 GiB327,684,276,0964.677—unsloth
UD-Q4_K_M9 shards334.55 GiB359,225,442,2405.127—unsloth
UD-Q5_K_S10 shards364.80 GiB391,697,744,0005.590—unsloth
UD-Q5_K_M10 shards397.22 GiB426,510,467,1366.087—unsloth
UD-Q6_K11 shards429.09 GiB460,735,987,9686.576—unsloth
Q8_014 shards544.13 GiB584,258,241,0568.339—unsloth
BF1624 shards1023.56 GiB1,099,040,900,57615.686—unsloth

Compare with

same modality, comparable size
vs DeepSeek-V4-Flashvs DeepSeek-V4-Flash-0731vs Laguna-S-2.1vs Hy3vs GLM-5.2vs Qwen3-Coder-NextAll alternatives →Will it fit on my card? →

Will it run on your card?

full quant x context sweep
By VRAM:8GB12GB16GB24GB32GB48GB
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 UD-IQ1_M at roughly 293.64 GiB. The real file is 175.12 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json
Layers
—
Attention heads
64
KV heads
2
Head dim
128
Hidden size
8192
Vocab
131,072
Sliding window
none
SWA period
—
MLA
no
Experts
512
Experts per token
22
use_sliding_window
—

Questions people ask

How much VRAM does NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 need?
UD-IQ1_M is exactly 188,031,516,192 bytes (175.12 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
Is NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 a mixture-of-experts model?
Yes — 512 experts, 22 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-3-Ultra-550B-A55B-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.