Károly Zsolnai-Fehér describes testing NVIDIA Nemotron 3 Ultra and finding that it struggled on his graphics and game coding prompts while working well on terminal tasks and file organization. His observations are a personal evaluationEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria. rather than a comprehensive benchmark.
He discusses the released weights, research and licensing, and notes that the large model needs substantial hardware to run locallyA local model runs on computing infrastructure controlled directly by the user rather than only through a remote model service.. He also explains its mixture-of-experts architectureA mixture-of-experts model contains multiple specialized subnetworks and activates a selected subset for each input instead of using every parameter every time., memory layers and other techniques intended to improve inference efficiency.
Watch on YouTube
