Nvidia launches personal AI computers and Blackwell Ultra chip at GTC 2025

One of the most significant announcements at the GTC 2025 event (USA) was the Blackwell Ultra platform, an upgraded version of the previous Blackwell architecture. Blackwell Ultra is designed to deliver significantly higher performance than the Hopper-generation GPUs (graphics cards), enhancing AI inference processing capabilities. According to Nvidia, this platform can accelerate token processing per second by up to 25 times compared to the previous generation, enabling AI models to operate faster and more efficiently.

Additionally, Nvidia has integrated the next-generation NVLink, allowing multiple Blackwell Ultra GPUs to be interconnected to form high-performance computing systems. This is particularly crucial as AI computing demands continue to rise, especially in scientific research and enterprise applications.

Beyond hardware, Nvidia introduced Dynamo, a new software platform described as the "operating system of the AI factory." Dynamo supports the entire AI operational cycle, from training and deployment to model optimization. This platform enables enterprises to easily deploy large-scale AI, optimize performance on Blackwell Ultra GPUs, and reduce model training time.

One of the most notable announcements at GTC 2025 was the DGX Station and DGX Spark personal AI computer series. These compact supercomputers offer powerful AI processing capabilities, allowing researchers and businesses to develop AI directly in their offices. DGX Station features the GB10 Grace Blackwell Superchip, capable of executing up to one quadrillion AI computations per second. Meanwhile, DGX Spark is a smaller version targeting small and medium-sized enterprises at a more affordable price point, making AI more accessible to a broader audience.

At GTC 2025, Nvidia also emphasized the role of CUDA-X, an accelerated computing platform applied across various fields, including physics, data science, computer engineering, deep learning, quantum computing, healthcare, and telecommunications. CUDA-X includes specialized libraries such as cuDNN, cuBLAS, and cuOPT, optimizing processing performance on Nvidia GPUs.

One notable highlight is CUDA-Q, a new toolkit for quantum computing that facilitates easier access to this technology for researchers and enterprises. By continuously expanding the CUDA-X ecosystem, Nvidia is strengthening its position not only in AI but also across multiple advanced technology domains.

The GTC 2025 conference demonstrated Nvidia's ongoing expansion in AI and accelerated computing. With products like Blackwell Ultra, Dynamo, DGX Station, DGX Spark, and the CUDA-X ecosystem, Nvidia is not only focusing on hardware but also building a robust software and services platform, driving AI development globally.

The event will continue with in-depth discussions on AI trends, deep learning, supercomputing, and cloud computing, promising to deliver valuable insights for the tech community.

As always, for the latest news on hardware launches and industry developments, be sure to follow our dedicated hardware coverage.