Nvidia reveals Blackwell B200 GPU, the “world's most powerful chip” for AI

Nvidia's H100 AI chip propelled it to a multitrillion-dollar valuation, outpacing tech giants like Alphabet and Amazon. Now, with the introduction of the Blackwell B200 GPU and GB200 "superchip," Nvidia aims to widen its lead.

The B200 GPU boasts 20 petaflops of FP4 horsepower, while the GB200 combines two GPUs with a Grace CPU, offering 30 times the performance for LLM inference workloads. Nvidia claims significant efficiency gains, reducing costs and energy consumption by up to 25 times over the H100.

The GB200 enables faster training, with 2,000 Blackwell GPUs performing tasks previously requiring 8,000 Hopper GPUs and consuming only four megawatts. Key enhancements include a second-gen transformer engine doubling compute and bandwidth, and a next-gen NVLink switch allowing 576 GPUs to communicate bidirectionally at 1.8 terabytes per second.

Nvidia packages these GPUs into larger designs like the NVL72 rack, supporting 27-trillion parameter models. Cloud service providers, including Amazon, Google, Microsoft, and Oracle, plan to offer NVL72 racks. Nvidia also introduces the DGX Superpod for DGX GB200, combining eight systems into one for high computing power. The company's systems can scale to tens of thousands of GB200 superchips, connected via high-speed networking.

While focused on AI and GPU computing, the Blackwell GPU architecture hints at future RTX 50-series desktop graphics cards.

Courtesy of - ‪@NVIDIA‬


#Nvidia, #AI, #GPU, #Blackwell, #Superchip, #Technology, #Innovation, #Performance, #Efficiency, #Training, #Cloud, #Data, #Compute, #Networking, #Scaling, #CloudComputing, #DataCenter, #Supercomputing, #FutureTech, #RTX