Nvidia has introduced the Blackwell B200 GPU, touting it as the most potent chip for AI applications. The new GPU promises substantial performance enhancements and energy efficiency gains compared to its predecessors. Nvidia claims that the GB200 superchip, which combines two B200 GPUs with a Grace CPU, can deliver 30 times the performance for LLM inference workloads while being significantly more efficient. The B200 GPU features a second-gen transformer engine that doubles the compute, bandwidth, and model size by utilizing four bits for each neuron instead of eight. Additionally, it introduces a next-gen NVLink switch that enables 576 GPUs to communicate with each other, boasting 1.8 terabytes per second of bidirectional bandwidth. Nvidia is also offering larger designs, such as the GB200 NVL72, which integrates 36 CPUs and 72 GPUs into a single liquid-cooled rack, providing a total of 720 petaflops of AI training performance or 1,440 petaflops of inference. The company asserts that major cloud service providers, including Amazon, Google, Microsoft, and Oracle, are planning to incorporate the NVL72 racks into their offerings. Furthermore, Nvidia's systems can scale to tens of thousands of the GB200 superchips, connected together with high-speed networking.