Nvidia Unveils New Blackwell GPU Architecture and GB200 Superchip
In a significant advancement in the computer and artificial intelligence industry, Nvidia has unveiled its new GPU architecture, Blackwell, and the GB200 superchip, promising to transform the future of AI and computing with impressive performance and efficiency gains. Jensen Huang, CEO of Nvidia, has been the one to reveal these developments, positioning Nvidia at the forefront of accelerated computing.
The GPU B200, part of the Blackwell architecture, boasts 208 billion transistors and has been dubbed the “most powerful chip in the world.” This superchip is capable of delivering 30 times the performance of its predecessor, the H100 GPU, while consuming 25 times less energy. Additionally, the innovations of Blackwell will make training models with up to 10 trillion parameters much more accessible, marking a milestone in the industry.
The significance of Blackwell lies in its potential to lead a new era in computational power, consolidating Nvidia’s leadership in accelerated computing and AI. Its widespread adoption in the industry is already underway, with companies like Amazon Web Services, Dell Technologies, Google, Meta, Microsoft, OpenAI, Oracle, Tesla, and xAI aligning to integrate these chips into their operations.
Blackwell introduces six revolutionary technologies that will enable significant advancements in data processing, engineering simulation, computer-aided electronic design, computer-aided drug design, quantum computing, and generative AI. These technologies include the world’s most powerful chip, a second-generation transformer engine, fifth-generation NVLink, an RAS engine, secure AI capabilities, and a decompression engine.
The GB200 Grace Blackwell Superchip connects two B200 Tensor Core GPUs to Nvidia’s Grace CPU via a low-power NVLink interconnect, offering exceptional performance for the most compute-intensive workloads. Systems powered by GB200 can connect to NVIDIA’s Quantum-X800 InfiniBand and Spectrum™-X800 Ethernet platforms for advanced networking at speeds of up to 800Gb/s.
The launch of Blackwell and the GB200 represents a generational leap in power and efficiency, reaffirming Nvidia’s leadership in accelerated computing and artificial intelligence. With the widespread adoption of these chips in the industry, they are poised to be the driving force behind the next wave of advances in AI and computing.