OllamaFrom the builder
NVIDIA DGX Spark performance
Ollama tested a range of models on NVIDIA DGX Spark using release-day firmware 580.95.05 and Ollama v0.12.6, and published its test script. Each test ran 10 times at temperature 0 with output capped at 500 tokens and caching off, covering models such as gpt-oss, gemma3, llama3.1, deepseek-r1 and qwen3. Ollama also notes that the 120GB of VRAM on the GB10 Grace Blackwell Superchip fits the full gpt-oss-120b model.
The test script and its readme are made available and can be customized for your own testing.
Saved . A shortened excerpt from the official blog; the original publisher retains authorship.