DeepSeekDeepSeek V3 on NVIDIA Grace Blackwell Ultra GB300
Yes — Grace Blackwell Ultra GB300 handles DeepSeek V3 well at Q4_K_M — 22 tok/s. Solid daily-driver performance on 288 GB VRAM.
Model Size
671B
Device VRAM
288 GB
Bandwidth
8000 GB/s
Quantization
Q4_K_M
Performance by Quantization
OwnRig currently has one published compatibility entry for DeepSeek V3 on NVIDIA Grace Blackwell Ultra GB300 at Q4_K_M. This is the best supported pairing we can stand behind today.
| Quantization | Speed | TTFT | Fits in VRAM | Rating | Confidence |
|---|---|---|---|---|---|
| Q4_K_M | 22 tok/s | 320ms | ✓ Yes | Good | estimated |
Notes
Q4_K_M
671B MoE model at Q4_K_M (180GB). 108GB headroom. Same architecture as R1.
About DeepSeek V3
DeepSeek V3 (671B) is a chat, coding, ai coding, reasoning, multi-purpose model. Massive MoE model rivaling GPT-4 class. Only ~37B parameters active per token despite 671B total. Requires multi-GPU or very large unified memory (128GB+ Apple Silicon at Q2/Q3). Not for casual home use. Included for completeness and to show what the high end looks like.
View all DeepSeek V3 hardware options →About NVIDIA Grace Blackwell Ultra GB300
NVIDIA Grace Blackwell Ultra GB300 has 288 GB at 8000 GB/s. Street price: $30,000.
See all models NVIDIA Grace Blackwell Ultra GB300 can run →Estimate method: Estimated from GB300 bandwidth ratio vs RTX 4090. Reference hardware source: dell.com (2026-03-23)
Performance varies by driver version, inference engine, quantization method, context length, and system configuration. Figures shown are estimates based on community benchmarks and may not reflect your exact setup. Product names are trademarks of their respective owners. OwnRig is independent and not affiliated with any hardware or AI model provider.