NVIDIA Shows AI Supercomputer Hopper, Price-Performance Ratio Ahead of Intel

Recently, NVIDIA showcased its new supercomputer Hopper in the latest results of the machine learning performance benchmark MLPerf. Named "NVIDIA Eos," this system is equipped with over 10,000 H100 GPUs interconnected via InfiniBand. The test results indicate that Hopper's computational power significantly surpasses that of competitors like Intel. Analysis suggests that Hopper delivers four times the performance per dollar compared to Intel's Gaudi2. As manufacturers other than NVIDIA gradually withdraw from the MLPerf training tests, the benchmark's significance is diminishing. However, Hopper's debut still underscores NVIDIA's leading position in the field of AI computational power.

NVIDIA Launches Revolutionary AI Data Center Design to Enhance High-Performance Computing

At the 2025 GTC conference, NVIDIA introduced the 'Omniverse DSX Blueprint' design, specifically tailored for gigawatt-scale AI data centers, known as the 'AI Factory.' This solution is based on the Omniverse framework and supports various scales from 100 million watts to 1 billion watts. It aims to efficiently train and run large AI models, meeting the growing demand for AI computing, and represents a significant advancement in artificial intelligence infrastructure.

NVIDIA Open Sources OmniVinci All-Modal Understanding Model with Only 1/6 of the Training Data

NVIDIA released the OmniVinci all-modal understanding model, leading top models by 19.05 points in multiple benchmark tests. The model uses only 0.2 trillion training tokens, achieving six times the data efficiency of competitors. It aims to achieve unified understanding of vision, audio, and text, advancing machine multimodal cognitive capabilities.

NVIDIA Launches OmniVinci, a Multimodal Understanding Model That Sets a New SOTA with 19.05 Points Higher

NVIDIA released the multimodal understanding model OmniVinci, which outperformed top models by 19.05 points in benchmark tests. The model achieves excellent performance with only 1/6 of the training data. It aims to enable AI systems to simultaneously understand vision, audio, and text, simulating human multisensory perception of the world.

NVIDIA Will Sell the DGX Spark Personal AI Supercomputer for $3999

NVIDIA will start selling the DGX Spark Personal AI Supercomputer on October 15th, priced at $3999. This compact device can be placed on a desk and offers powerful computing capabilities, suitable for AI model training and research. Users can purchase it through the official website and US partners, targeting data scientists and AI researchers.

NVIDIA's Personal AI Supercomputer DGX Spark Will Be Released This Wednesday, Bringing AI Power to the Desktop Era

NVIDIA's first personal AI supercomputer, DGX Spark, will be released this week. It offers powerful performance for handling complex AI models and has a compact design suitable for desktop placement. Orders can be placed online on the official website starting October 15th, or purchased through designated channels in the United States. The official price has been adjusted from the previously announced 3000 USD to 3999 USD.