DeepSeek-Coder-V2 236B
deepseek-ai/DeepSeek-Coder-V2-Instruct
236B MoE coding model.
Parameters
236B
Context
128K
Downloads
890K
Likes
2.1K
Architecture
moe_transformer
GPU Requirements
VRAM / RAM / disk estimates by quantization and usage scenario
| Quantization | Minimum | Recommended | Production |
|---|---|---|---|
| FP1616bit | 561.0 GB RAM 841.5 · Disk 724.8 | 602.1 GB RAM 903.1 · Disk 750.8 | 724.2 GB RAM 1086.3 · Disk 789.8 |
| FP88bit | 283.0 GB RAM 424.5 · Disk 381.9 | 311.4 GB RAM 467.1 · Disk 407.9 | 420.9 GB RAM 631.4 · Disk 446.9 |
| INT88bit | 283.0 GB RAM 424.5 · Disk 381.9 | 311.4 GB RAM 467.1 · Disk 407.9 | 420.9 GB RAM 631.4 · Disk 446.9 |
| AWQ4bit | 149.2 GB RAM 223.8 · Disk 216.9 | 171.5 GB RAM 257.2 · Disk 242.9 | 274.9 GB RAM 412.4 · Disk 281.9 |
| GPTQ4bit | 150.9 GB RAM 226.4 · Disk 219.0 | 173.3 GB RAM 260.0 · Disk 245.0 | 276.8 GB RAM 415.3 · Disk 284.0 |
| GGUF4bit | 154.4 GB RAM 231.6 · Disk 223.3 | 176.9 GB RAM 265.4 · Disk 249.3 | 280.6 GB RAM 421.0 · Disk 288.3 |
GPU Recommendations
Best ValueBEST
8× H200 SXM
1128 GB VRAM
$20,440/mo
≈ 134.7 tok/s
Effective production throughput ≈ 94.3 tok/s
score 0.2603
Cheapest
8× RTX 3090
192 GB VRAM
$630/mo
≈ 26.3 tok/s
Effective production throughput ≈ 18.4 tok/s
score 0.5301
Performance
4× H200 SXM
564 GB VRAM
$10,220/mo
≈ 101 tok/s
Effective production throughput ≈ 70.7 tok/s
score 0.3505
Min Complexity
1× B200
180 GB VRAM
$3,013/mo
≈ 0 tok/s (N/A — benchmark unavailable)
Effective production throughput ≈ 0 tok/s
score 0.5739
💬 Community — real deployment experience
0 Articles · 0 Benchmarks
GPUs that fit (single card)
B200
180 GB VRAM
$3,013
/mo · vast
B200 SXM
180 GB VRAM
—