nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16
nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16
Parameters
5.4B
Context
256K
Downloads
4.7M
Likes
126
Architecture
nemotron_h
42 layers · 8 KV heads · head 128 · hidden 3136
GPU Requirements
VRAM / RAM / disk estimates by quantization and usage scenario
| Quantization | Minimum | Recommended | Production |
|---|---|---|---|
| FP1616bit | 17.8 GB RAM 32.0 · Disk 54.6 | 36.8 GB RAM 55.3 · Disk 80.6 | 155.6 GB RAM 233.4 · Disk 119.6 |
| FP88bit | 11.5 GB RAM 32.0 · Disk 46.8 | 30.2 GB RAM 45.4 · Disk 72.8 | 148.7 GB RAM 223.0 · Disk 111.8 |
| INT88bit | 11.5 GB RAM 32.0 · Disk 46.8 | 30.2 GB RAM 45.4 · Disk 72.8 | 148.7 GB RAM 223.0 · Disk 111.8 |
| AWQ4bit | 8.4 GB RAM 32.0 · Disk 43.1 | 27.1 GB RAM 40.6 · Disk 69.1 | 145.4 GB RAM 218.1 · Disk 108.1 |
| GPTQ4bit | 8.5 GB RAM 32.0 · Disk 43.1 | 27.1 GB RAM 40.6 · Disk 69.1 | 145.4 GB RAM 218.1 · Disk 108.1 |
| GGUF4bit | 8.5 GB RAM 32.0 · Disk 43.2 | 27.2 GB RAM 40.8 · Disk 69.2 | 145.5 GB RAM 218.2 · Disk 108.2 |
GPU Recommendations
1× H100 SXM
80 GB VRAM
≈ 1293.4 tok/s
Effective production throughput ≈ 905.4 tok/s
score 0.5777
2× Tesla V100
32 GB VRAM
≈ 0 tok/s (N/A — benchmark unavailable)
Effective production throughput ≈ 0 tok/s
score 0.6206
8× H200 SXM
1128 GB VRAM
≈ 5930.5 tok/s
Effective production throughput ≈ 4151.3 tok/s
score 0.2197
1× RTX 5090
32 GB VRAM
≈ 691.9 tok/s
Effective production throughput ≈ 484.3 tok/s
score 0.6741
💬 Community — real deployment experience
0 Articles · 0 Benchmarks
GPUs that fit (single card)
RTX 5090
32 GB VRAM
$155
/mo · vast
A100 SXM4
80 GB VRAM
$225
/mo · vast
RTX A6000
48 GB VRAM
$293
/mo · vast
L40S
48 GB VRAM
$314
/mo · vast
A100 PCIE
80 GB VRAM
$355
/mo · vast