Details about NVIDIA’s upcoming chip for the Chinese AI market have emerged, signaling a significant move by the tech giant to capture attention with its Blackwell chips. This follows the recent export restrictions on the H20, prompting NVIDIA to innovate for the Chinese consumer.
With the US limitations in place, NVIDIA is exploring the Blackwell architecture as a fresh solution. Reports suggest that this new AI chip for China will incorporate GDDR7 memory and aims to ship an impressive one million units by the end of 2025, highlighting substantial demand in the region.
Rumors indicate that this special edition Blackwell GPU is set for mass production in early July and might be called the 6000D or B40. It is expected to boast GDDR7 memory with a bandwidth of approximately 1.7TB/s. The NVLink speed is estimated at 550GB/s per direction.
In response to US restrictions, NVIDIA’s upcoming chips will forgo HBM and embrace GDDR7 technology. The anticipated memory bandwidth is akin to the RTX 5090 GPU. Additionally, the chip is likely to capitalize on CUDA capabilities to enhance performance and efficiency, ensuring NVIDIA remains competitive in China.
Despite differences, NVIDIA’s “China-specific” Blackwell chip is anticipated to outperform Huawei’s current solutions due to superior architectural designs and a robust software ecosystem. However, the dependency concerns have pushed Chinese firms to consider domestic alternatives like Huawei’s Ascend chips, which seem to be gathering momentum.
NVIDIA’s swift action in the Chinese market suggests a new product launch as early as July, promising to re-establish its presence. CEO Jensen Huang has been vocal about opposing the US’s tough stance on China’s AI progression, hinting that NVIDIA remains committed to serving Chinese consumers.
