Nvidia Developing B30 AI Chip for China with Multi-GPU Scaling Capabilities Amid Export Controls

Nvidia is reportedly developing a new AI chip for China, called the B30, that complies with U.S. export controls, according to a recent report by The Information. This move comes after the U.S. government banned the export of Nvidia’s H20 chips to China due to concerns over their potential use in building supercomputers for military applications. The B30 is part of a broader Blackwell-based family of accelerators, which various leaks have referred to as RTX Pro 6000D, B40, and now B30, suggesting there may be different versions within the BXX series. The B30 solutions are reportedly designed with multi-GPU scaling capabilities, using GDDR7 memory and GB20X silicon—the same technology found in consumer-grade RTX 50 GPUs. Although many assume that this 'multi-GPU scaling' feature involves NVLink, Nvidia has not included NVLink support in its recent consumer-grade GPU designs. However, it is possible that Nvidia has modified the GB202 chips from the RTX 5090 to enable NVLink functionality. At Computex, Nvidia unveiled the RTX Pro Blackwell servers, capable of housing up to eight RTX Pro 6000 GPUs. These are interconnected via ConnectX-8 SuperNICs featuring built-in PCIe 6.0 switches, facilitating efficient GPU-to-GPU communication. A similar configuration is likely to be used in the B30, allowing for seamless connection and performance scaling between multiple GPUs. When questioned about the H20 export ban, Nvidia CEO Jensen Huang firmly stated that the company would not develop any additional alternatives based on the Hopper architecture, signaling a strategic shift to the Blackwell architecture for future China-bound products. The U.S. government's restrictions specifically target the high memory bandwidth and interconnect bandwidth of the H20, deeming them too powerful for unrestricted export to China. Nvidia has been grappling with these export control policies for years, a situation that has also affected competitors like AMD, which recently projected potential losses of up to $800 million due to the ban on its MI308 chips. Huang criticized these policies, describing them as "a failure" and warning that they could push China to accelerate its own innovation in AI hardware. He expressed concern that if companies like Huawei are compelled to design and produce competitive hardware, the United States might lose its technological edge and potentially even set future global standards. The development of the B30 and other Blackwell-based accelerators for the Chinese market represents Nvidia’s adaptation to these stringent export controls while maintaining its presence and competitiveness in one of the world’s largest tech markets. By adhering to the imposed regulations, Nvidia aims to continue serving Chinese customers without violating U.S. policy, a delicate balancing act that underscores the complex dynamics of international technology trade. To stay updated on the latest developments in this space, follow Tom’s Hardware on Google News and make sure to hit the Follow button.