deepseek-ai/DeepSeek-V4-Flash-0731
deepseek-ai/DeepSeek-V4-Flash-0731
Parameters
304.2B
Context
1024K
Downloads
1M
Likes
3.2K
Architecture
deepseek_v4
43 layers · 1 KV heads · head 512 · hidden 4096
GPU Requirements
VRAM / RAM / disk estimates by quantization and usage scenario
| Quantization | Minimum | Recommended | Production |
|---|---|---|---|
| FP1616bit | 722.3 GB RAM 1083.4 · Disk 922.9 | 770.2 GB RAM 1155.2 · Disk 948.9 | 887.5 GB RAM 1331.2 · Disk 987.9 |
| FP88bit | 363.9 GB RAM 545.8 · Disk 480.9 | 395.5 GB RAM 593.2 · Disk 506.9 | 496.5 GB RAM 744.8 · Disk 545.9 |
| INT88bit | 363.9 GB RAM 545.8 · Disk 480.9 | 395.5 GB RAM 593.2 · Disk 506.9 | 496.5 GB RAM 744.8 · Disk 545.9 |
| AWQ4bit | 191.4 GB RAM 287.1 · Disk 268.3 | 215.2 GB RAM 322.8 · Disk 294.3 | 308.4 GB RAM 462.6 · Disk 333.3 |
| GPTQ4bit | 193.7 GB RAM 290.5 · Disk 271.0 | 217.5 GB RAM 326.3 · Disk 297.0 | 310.8 GB RAM 466.3 · Disk 336.0 |
| GGUF4bit | 198.2 GB RAM 297.2 · Disk 276.5 | 222.2 GB RAM 333.3 · Disk 302.5 | 315.7 GB RAM 473.6 · Disk 341.5 |
GPU Recommendations
Best ValueBEST
8× H200 SXM
1128 GB VRAM
$20,440/mo
≈ 104.5 tok/s
Effective production throughput ≈ 73.1 tok/s
score 0.2706
Cheapest
8× RTX A6000
384 GB VRAM
$1,678/mo
≈ 16.7 tok/s
Effective production throughput ≈ 11.7 tok/s
score 0.492
Performance
4× H200 SXM
564 GB VRAM
$10,220/mo
≈ 78.4 tok/s
Effective production throughput ≈ 54.9 tok/s
score 0.3711
Min Complexity
2× H200 SXM
282 GB VRAM
$5,110/mo
≈ 52.3 tok/s
Effective production throughput ≈ 36.6 tok/s
score 0.5924
💬 Community — real deployment experience
0 Articles · 0 Benchmarks