Nvidia introduces Diffusion Nemotron, a 60-billion parameter two-tower AI model that uses parallel block processing to speed up text generation.