| name | dgx-spark-llm |
| description | DGX Spark LLM optimization guidance and best practices |
DGX Spark LLM Optimization
Guidance for running LLMs optimally on NVIDIA DGX Spark (Grace-Blackwell architecture).
Hardware Overview
- CPU: Grace ARM64 (72 cores)
- GPU: Blackwell GPU
- Memory: 119GB unified memory (GPU can access system RAM)
- Architecture: GB10 (ARM + Blackwell unified)
Optimal llama.cpp Settings
GPU Offloading
-ngl 99 # Full GPU offload (all layers)
With 119GB unified memory, most models fit entirely in GPU memory.
CPU Threads
--threads 8
Grace CPU benefits from moderate thread count. Higher isn't always better due to memory bandwidth.
Flash Attention
--flash-attn
Blackwell supports flash attention natively. Always enable for better memory efficiency.
Context Size
For large models (30B+):
- Start with 8192 context
- Increase to 16384 or 32768 if needed
- Monitor memory usage with
/llama:status
Quantization Recommendations
With 119GB VRAM, prefer higher quality quantizations:
| Model Size | Recommended Quant | Reasoning |
|---|
| 7B | Q8_0 or F16 | Fits easily, maximize quality |
| 13B | Q8_0 | Still fits with room to spare |
| 30B | Q8_0 | Fits in 119GB |
| 70B | Q6_K or Q5_K_M | May need lower quant |
Model Selection
For the DGX Spark's capabilities:
Coding Tasks
- Qwen2.5-Coder-32B (Q8_0)
- DeepSeek-Coder-33B (Q8_0)
- CodeLlama-34B (Q8_0)
General Reasoning
- Nemotron-3-Nano-30B (MoE, excellent for reasoning)
- Qwen2.5-32B-Instruct
- Llama-3.1-70B (Q5_K_M)
Fast Inference
- GLM-4.7-Flash
- Llama-3.2-3B (for quick tasks)
- Phi-3-mini (very fast)
Performance Tuning
Batch Size
For interactive use:
--batch-size 512
For throughput:
--batch-size 2048
Continuous Batching
Enable for multiple concurrent users:
--cont-batching
Memory Mapping
For models close to memory limit:
--mlock # Lock model in memory
Troubleshooting
Out of Memory
- Reduce context size:
--ctx-size 4096
- Use lower quantization: Q4_K_M instead of Q8_0
- Reduce batch size:
--batch-size 256
Slow Inference
- Ensure full GPU offload:
-ngl 99
- Enable flash attention:
--flash-attn
- Check no CPU fallback in logs
Model Loading Slow
- Use mmap (default): fast initial load
- Consider
--mlock for consistent performance