Gemma
google/gemma-4-31B-it
google/gemma-4-31B-it
Parameters
22.6B
Context
256K
Downloads
9.9M
Likes
3.5K
Architecture
gemma4
60 layers · 16 KV heads · head 256 · hidden 5376
GPU Requirements
VRAM / RAM / disk estimates by quantization and usage scenario
| Quantization | Minimum | Recommended | Production |
|---|---|---|---|
| FP1616bit | 65.1 GB RAM 97.7 · Disk 104.8 | 149.7 GB RAM 224.6 · Disk 130.8 | 731.1 GB RAM 1096.7 · Disk 169.8 |
| FP88bit | 38.4 GB RAM 57.6 · Disk 71.9 | 121.8 GB RAM 182.8 · Disk 97.9 | 702.0 GB RAM 1053.0 · Disk 136.9 |
| INT88bit | 38.4 GB RAM 57.6 · Disk 71.9 | 121.8 GB RAM 182.8 · Disk 97.9 | 702.0 GB RAM 1053.0 · Disk 136.9 |
| AWQ4bit | 25.6 GB RAM 38.4 · Disk 56.1 | 108.4 GB RAM 162.6 · Disk 82.1 | 688.0 GB RAM 1032.0 · Disk 121.1 |
| GPTQ4bit | 25.8 GB RAM 38.6 · Disk 56.3 | 108.6 GB RAM 162.9 · Disk 82.3 | 688.2 GB RAM 1032.3 · Disk 121.3 |
| GGUF4bit | 26.1 GB RAM 39.1 · Disk 56.7 | 108.9 GB RAM 163.4 · Disk 82.7 | 688.6 GB RAM 1032.8 · Disk 121.7 |
GPU Recommendations
Best ValueBEST
4× H200 SXM
564 GB VRAM
$10,220/mo
≈ 1053 tok/s
Effective production throughput ≈ 737.1 tok/s
score 0.3206
Cheapest
8× RTX 4070S Ti
128 GB VRAM
$396/mo
≈ 0 tok/s (N/A — benchmark unavailable)
Effective production throughput ≈ 0 tok/s
score 0.5259
Performance
8× H200 SXM
1128 GB VRAM
$20,440/mo
≈ 1404 tok/s
Effective production throughput ≈ 982.8 tok/s
score 0.2197
Min Complexity
1× H200 SXM
141 GB VRAM
$2,555/mo
≈ 438.8 tok/s
Effective production throughput ≈ 307.2 tok/s
score 0.666
💬 Community — real deployment experience
0 Articles · 0 Benchmarks
GPUs that fit (single card)
H200 SXM
141 GB VRAM
$2,555
/mo · lambda
B200
180 GB VRAM
$3,013
/mo · vast
B200 SXM
180 GB VRAM
—