Can the DGX Spark (GB10) run Phi-4 Mini?
Yes. Phi-4 Mini fits the DGX Spark (GB10)'s 128 GB at Q4_K_M (about 4.3 GB needed), running at an estimated 63 tok/s. There's room for up to FP16 if you want higher quality.
Every quantization
Phi-4 Mini (3B) on 128 GB
| Quant | Weights | Needs | Fit | Est. speed |
|---|---|---|---|---|
| Q4_K_M | 2.5 GB | 4.3 GB | Fits easily | ~63 tok/s |
| Q5_K_M | 3 GB | 4.8 GB | Fits easily | ~52 tok/s |
| Q8_0 | 4.5 GB | 6.3 GB | Fits easily | ~35 tok/s |
| FP16 | 6 GB | 7.8 GB | Fits easily | ~26 tok/s |
Estimates. Needs = weights + 1.8 GB for KV cache and runtime at modest context. Speed is single-stream decode scaled from the DGX Spark (GB10)'s 273.2 GB/s bandwidth (Estimated from memory bandwidth (llama.cpp decode)). Long contexts need more memory and run slower.
Run it
Copy, paste, go
ollama run phi-4-mini:3bllama-server -hf unsloth/Phi-4-mini-instruct-GGUF -ngl 99Ollama pulls a Q4 build by default. Pick a larger quant only if the table above says it fits. Full Phi-4 Mini details →
Compare
Same model, other hardware - and other models on this card
Phi-4 Mini on other hardware
- RTX 5090~411 tok/s
- RTX 5090 D~411 tok/s
- RTX 6000D~360 tok/s
- RTX 4090~250 tok/s
- Radeon RX 7900 XTX~220 tok/s
- RTX 5080~220 tok/s
- RTX 3090~219 tok/s
- RTX 3080 Ti~209 tok/s
Also runs on the DGX Spark (GB10)
- gpt-oss-120b~35 tok/s
- Qwen2.5 72B~3.9 tok/s
- Llama 3.3 70B~3.9 tok/s
- Qwen3.6 35B A3B~60 tok/s
- Qwen3.5 35B A3B~60 tok/s
- dolphin-2.9.1-yi-1.5-34b~7.8 tok/s
- Gemma 4 31B~8.3 tok/s
- Qwen3 32B~8.3 tok/s
Frequently asked
Quick answers
Can the NVIDIA DGX Spark (GB10) run Phi-4 Mini?
Yes. Phi-4 Mini fits the DGX Spark (GB10)'s 128 GB at Q4_K_M (about 4.3 GB needed), running at an estimated 63 tok/s. There's room for up to FP16 if you want higher quality.
How much VRAM does Phi-4 Mini need?
About 2.5 GB of weights at Q4_K_M, 4.5 GB at Q8_0 and 6 GB at FP16. Add roughly 1.8 GB for the KV cache and runtime at modest context lengths.
How fast is Phi-4 Mini on the DGX Spark (GB10)?
Roughly 63 tokens per second at Q4_K_M (single user, llama.cpp-style decode). This is an estimate from the DGX Spark (GB10)'s 273.2 GB/s memory bandwidth.