Qwen/Qwen3-8B
Qwen/Qwen3-8B
Lightweight 8B model ideal for edge deployment and cheap cloud GPUs.
Parameters
7.9B
Context
40K
Downloads
16.3M
Likes
1.3K
Architecture
qwen3
36 layers · 8 KV heads · head 128 · hidden 4096
GPU Requirements
VRAM / RAM / disk estimates by quantization and usage scenario
| Quantization | Minimum | Recommended | Production |
|---|---|---|---|
| FP1616bit | 24.3 GB RAM 36.4 · Disk 62.8 | 40.9 GB RAM 61.4 · Disk 88.8 | 138.7 GB RAM 208.0 · Disk 127.8 |
| FP88bit | 14.6 GB RAM 32.0 · Disk 50.9 | 30.8 GB RAM 46.2 · Disk 76.9 | 128.1 GB RAM 192.2 · Disk 115.9 |
| INT88bit | 14.6 GB RAM 32.0 · Disk 50.9 | 30.8 GB RAM 46.2 · Disk 76.9 | 128.1 GB RAM 192.2 · Disk 115.9 |
| AWQ4bit | 10.0 GB RAM 32.0 · Disk 45.2 | 25.9 GB RAM 38.9 · Disk 71.2 | 123.1 GB RAM 184.6 · Disk 110.2 |
| GPTQ4bit | 10.0 GB RAM 32.0 · Disk 45.3 | 26.0 GB RAM 39.0 · Disk 71.3 | 123.1 GB RAM 184.7 · Disk 110.3 |
| GGUF4bit | 10.1 GB RAM 32.0 · Disk 45.4 | 26.1 GB RAM 39.2 · Disk 71.4 | 123.3 GB RAM 184.9 · Disk 110.4 |
GPU Recommendations
1× RTX 5090
32 GB VRAM
≈ 452.5 tok/s
Effective production throughput ≈ 316.8 tok/s
score 0.6817
2× Tesla V100
32 GB VRAM
≈ 0 tok/s (N/A — benchmark unavailable)
Effective production throughput ≈ 0 tok/s
score 0.6305
8× H200 SXM
1128 GB VRAM
≈ 3878.8 tok/s
Effective production throughput ≈ 2715.2 tok/s
score 0.2197
1× RTX A6000
48 GB VRAM
≈ 193.9 tok/s
Effective production throughput ≈ 135.7 tok/s
score 0.6352
💬 Community — real deployment experience
0 Articles · 0 Benchmarks
GPUs that fit (single card)
RTX A6000
48 GB VRAM
$210
/mo · vast
RTX 5090
32 GB VRAM
$236
/mo · vast
L40S
48 GB VRAM
$341
/mo · vast
RTX PRO 6000 WS
48 GB VRAM
$482
/mo · vast
RTX PRO 6000 S
48 GB VRAM
$633
/mo · vast