Qwen/Qwen3.8-27B-FP8
Qwen/Qwen3.8-27B-FP8
Parameters
21.8B
Context
256K
Downloads
5.2M
Likes
869
Architecture
qwen3_5
64 layers · 4 KV heads · head 256 · hidden 5120
GPU Requirements
VRAM / RAM / disk estimates by quantization and usage scenario
| Quantization | Minimum | Recommended | Production |
|---|---|---|---|
| FP1616bit | 74.3 GB RAM 111.4 · Disk 119.7 | 112.1 GB RAM 168.2 · Disk 145.7 | 323.4 GB RAM 485.1 · Disk 184.7 |
| FP88bit | 41.5 GB RAM 62.3 · Disk 79.4 | 77.9 GB RAM 116.9 · Disk 105.4 | 287.7 GB RAM 431.6 · Disk 144.4 |
| INT88bit | 41.5 GB RAM 62.3 · Disk 79.4 | 77.9 GB RAM 116.9 · Disk 105.4 | 287.7 GB RAM 431.6 · Disk 144.4 |
| AWQ4bit | 25.8 GB RAM 38.7 · Disk 59.9 | 61.5 GB RAM 92.2 · Disk 85.9 | 270.5 GB RAM 405.8 · Disk 124.9 |
| GPTQ4bit | 26.0 GB RAM 39.0 · Disk 60.2 | 61.7 GB RAM 92.5 · Disk 86.2 | 270.8 GB RAM 406.1 · Disk 125.2 |
| GGUF4bit | 26.4 GB RAM 39.6 · Disk 60.7 | 62.1 GB RAM 93.1 · Disk 86.7 | 271.2 GB RAM 406.8 · Disk 125.7 |
GPU Recommendations
1× H100 SXM
80 GB VRAM
≈ 249.6 tok/s
Effective production throughput ≈ 174.7 tok/s
score 0.6798
4× Tesla V100
64 GB VRAM
≈ 0 tok/s (N/A — benchmark unavailable)
Effective production throughput ≈ 0 tok/s
score 0.5425
8× H200 SXM
1128 GB VRAM
≈ 1144.6 tok/s
Effective production throughput ≈ 801.2 tok/s
score 0.2197
1× A100 SXM4
80 GB VRAM
≈ 0 tok/s (N/A — benchmark unavailable)
Effective production throughput ≈ 0 tok/s
score 0.678
💬 Community — real deployment experience
0 Articles · 0 Benchmarks
GPUs that fit (single card)
A100 SXM4
80 GB VRAM
$344
/mo · vast
A100 PCIE
80 GB VRAM
$499
/mo · vast
A100 80GB
80 GB VRAM
$876
/mo · vast
H100 SXM
80 GB VRAM
$1,404
/mo · vast
H200 NVL
141 GB VRAM
$1,608
/mo · vast