NVIDIA's GTC 2025 unveiled groundbreaking advancements in AI technology. The new Blackwell Ultra GPU, designed for large-scale AI inference, delivers a 50% performance boost over its predecessor, reaching approximately 15 PetaFLOPS using FP4 precision. It features advanced HBM3E memory, increasing capacity from 192GB to 288GB. NVIDIA also previewed the upcoming Rubin architecture, set to launch in 2026, capable of 50 PetaFLOPS (FP4), with the Rubin Ultra GPU reaching 100 PetaFLOPS and utilizing HBM4 memory. These innovations mark a significant shift towards AI inference, emphasizing the growing demand for powerful computing solutions in the evolving AI landscape.
https://www.patreon.com/posts/nvidia-unveils-124754224