NVIDIA is working on a new chip for artificial intelligence (AI) for the Chinese market, which will surpass the current model H20. The chip, conditionally called the B30A, is created on the basis of the Blackwell architecture, but its sales depend on the solution of American regulators.
According to sources, the B30A will use a single -chip design, which will provide approximately half the performance of the flagship accelerator NVIDIA B300. However, the chip will receive high -speed memory and NVLINK technology for quick data transfer between processors – as in H20, which is based on the older Hopper architecture.
Although the exact characteristics have not yet been approved, NVIDIA plans to send the first samples to Chinese clients for testing the next month. The company said that all their products are developed as part of the authorities and are intended exclusively for commercial use.
The US authorities have not yet commented on the situation.
