NVIDIA (NVDA) Hardware Powers Google DeepMind’s Parallel DiffusionGemma Model

NVIDIA Corporation (NASDAQ:NVDA) is one of the 12 Best Growth Stocks Trading in Oversold Territory. On June 10, Google DeepMind released DiffusionGemma, an experimental open-weight model that uses NVIDIA hardware to achieve breakthrough text generation speeds. By abandoning the traditional token-by-token autoregressive approach in favor of parallel block generation, the model shifts the processing bottleneck from memory bandwidth to compute, allowing NVIDIA’s Tensor Cores to process up to 256 tokens simultaneously.

This parallel architecture is specifically optimized for the NVIDIA ecosystem, including GeForce RTX GPUs, the NVIDIA RTX PRO platform, and DGX Spark systems. Through the synergy between Google’s diffusion head and NVIDIA’s CUDA software stack, the model achieves performance gains of up to 4x over equivalent autoregressive models, hitting speeds of 2,000 tokens per second on NVIDIA DGX Station and 1,000 tokens per second on an H100 GPU.

By enabling this compute-heavy, low-latency performance, NVIDIA Corporation (NASDAQ:NVDA) makes advanced, local-first AI generation accessible to developers and researchers without the need for cloud-based inference. With day-zero support across key tools like Hugging Face, vLLM, and the NVIDIA NeMo framework, users can immediately deploy DiffusionGemma for rapid iteration, agentic loops, and interactive workflows on their own hardware.

NVIDIA Corporation (NASDAQ:NVDA) is a fabless semiconductor and AI computing company that designs GPUs, AI accelerators, Application Programming Interfaces/APIs, and system-on-a-chip units. Through its CUDA ecosystem, the company enables industries ranging from autonomous vehicles to scientific research by advancing AI, accelerated computing, and data center infrastructure.

READ NEXT: 33 Stocks That Should Double in 3 Years and Cathie Wood 2026 Portfolio: 10 Best Stocks to Buy. 

Disclosure: None. Follow Insider Monkey on Google News.