NVIDIA Adds Day-1 Support for DiffusionGemma AI

NVIDIA provides day-1 support for Google DeepMind’s DiffusionGemma text generation model across RTX GPUs and DGX systems.

NVIDIA has announced immediate support for Google DeepMind’s DiffusionGemma, an open model designed for text generation, across its entire hardware ecosystem. This day-1 integration spans GeForce RTX GPUs, RTX PRO Platforms, and DGX systems ranging from Spark Mini PCs to high-end workstations. Leveraging its tensor core architecture and the CUDA software stack, NVIDIA provides a full-stack framework for the model. Performance benchmarks show NVIDIA H100 Tensor Core GPUs on DGX Stations achieving 1000 tokens per second, while DGX Spark systems reach 150 tokens per second. These solutions, compatible with hardware like the RTX 5090, deliver approximately four times the performance of equivalent autoregressive models.