Can the DGX Spark (GB10) run Granite 4.2 3B?
Yes. Granite 4.2 3B fits the DGX Spark (GB10)'s 128 GB at Q4_K_M (about 3.8 GB needed), running at an estimated 78 tok/s. There's room for up to FP16 if you want higher quality.
Every quantization
Granite 4.2 3B (3.7B) on 128 GB
| Quant | Weights | Needs | Fit | Est. speed |
|---|---|---|---|---|
| Q4_K_M | 2 GB | 3.8 GB | Fits easily | ~78 tok/s |
| Q5_K_M | 3 GB | 4.8 GB | Fits easily | ~52 tok/s |
| Q8_0 | 4 GB | 5.8 GB | Fits easily | ~39 tok/s |
| FP16 | 7 GB | 8.8 GB | Fits easily | ~22 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
llama-server -hf ibm-granite/granite-4.2-3b-GGUF -ngl 99Ollama pulls a Q4 build by default. Pick a larger quant only if the table above says it fits. Full Granite 4.2 3B details →
Compare
Same model, other hardware - and other models on this card
Granite 4.2 3B on other hardware
- RTX 5090~514 tok/s
- RTX 5090 D~513 tok/s
- RTX 6000D~450 tok/s
- RTX 4090~313 tok/s
- Radeon RX 7900 XTX~275 tok/s
- RTX 5080~275 tok/s
- RTX 3090~274 tok/s
- RTX 3080 Ti~261 tok/s
Also runs on the DGX Spark (GB10)
- gpt-oss-120b~35 tok/s
- Llama 3.3 70B~3.9 tok/s
- Nemotron 3.5 Super VL~4.1 tok/s
- Qwen3.6 35B A3B~60 tok/s
- Laguna-XS-2.1~8.3 tok/s
- Gemma 4 31B~8.3 tok/s
- Qwen3 32B~8.3 tok/s
- Qwen3-Coder-30B-A3B-Instruct~59 tok/s
Frequently asked
Quick answers
Can the NVIDIA DGX Spark (GB10) run Granite 4.2 3B?
Yes. Granite 4.2 3B fits the DGX Spark (GB10)'s 128 GB at Q4_K_M (about 3.8 GB needed), running at an estimated 78 tok/s. There's room for up to FP16 if you want higher quality.
How much VRAM does Granite 4.2 3B need?
About 2 GB of weights at Q4_K_M, 4 GB at Q8_0 and 7 GB at FP16. Add roughly 1.8 GB for the KV cache and runtime at modest context lengths.
How fast is Granite 4.2 3B on the DGX Spark (GB10)?
Roughly 78 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.