LLMMIT
SmolLM3 3B
HuggingFaceTB/SmolLM3-3B
Parameters
3B
Context
64K
Downloads
821.6K
Likes
1K
Architecture
transformer
GPU Requirements
VRAM / RAM / disk estimates by quantization and usage scenario
| Quantization | Minimum | Recommended | Production |
|---|---|---|---|
| FP1616bit | 12.0 GB RAM 32.0 · Disk 47.7 | 28.1 GB RAM 42.1 · Disk 73.7 | 125.3 GB RAM 188.0 · Disk 112.7 |
| FP88bit | 8.5 GB RAM 32.0 · Disk 43.4 | 24.4 GB RAM 36.6 · Disk 69.4 | 121.5 GB RAM 182.2 · Disk 108.4 |
| INT88bit | 8.5 GB RAM 32.0 · Disk 43.4 | 24.4 GB RAM 36.6 · Disk 69.4 | 121.5 GB RAM 182.2 · Disk 108.4 |
| AWQ4bit | 6.8 GB RAM 32.0 · Disk 41.3 | 22.6 GB RAM 33.9 · Disk 67.3 | 119.6 GB RAM 179.4 · Disk 106.3 |
| GPTQ4bit | 6.8 GB RAM 32.0 · Disk 41.3 | 22.6 GB RAM 34.0 · Disk 67.3 | 119.6 GB RAM 179.4 · Disk 106.3 |
| GGUF4bit | 6.9 GB RAM 32.0 · Disk 41.3 | 22.7 GB RAM 34.0 · Disk 67.3 | 119.7 GB RAM 179.5 · Disk 106.3 |
GPU Recommendations
Best ValueBEST
1× RTX 5090
32 GB VRAM
$303/mo
≈ 1235.9 tok/s
Effective production throughput ≈ 865.1 tok/s
score 0.685
Cheapest
1× L4
24 GB VRAM
$76.2/mo
≈ 206.9 tok/s
Effective production throughput ≈ 144.8 tok/s
score 0.6392
Performance
8× H200 SXM
1128 GB VRAM
$20,440/mo
≈ 10593.1 tok/s
Effective production throughput ≈ 7415.2 tok/s
score 0.2197
Min Complexity
1× RTX 3090
24 GB VRAM
$81.4/mo
≈ 645.5 tok/s
Effective production throughput ≈ 451.8 tok/s
score 0.6437
💬 Community — real deployment experience
0 Articles · 0 Benchmarks
GPUs that fit (single card)
L4
24 GB VRAM
$76.2
/mo · vast
RTX 3090
24 GB VRAM
$81.4
/mo · vast
RTX 4090
24 GB VRAM
$190
/mo · tensordock
RTX A6000
48 GB VRAM
$245
/mo · vast
A100 SXM4
80 GB VRAM
$276
/mo · vast