Flagship Nvidia GPU to get VRAM downgrade in one region, according to manufacturer
Table of Contents
Nvidia continues to battle with the U.S. government restrictions on sales of high-end chips to the Chinese market. Team Green is not willing to give up on the lucrative Chinese market, and the U.S. government shows no sign of budging when it comes to its trade restrictions on AI tech, so it has come down to a peculiar sort of arms race.
Earlier this month, we reported that Nvidia had been forced to suspend sales of the RTX 5090D in China to comply with U.S. import restrictions on AI and high-performance computing technology. The ‘D' variant of its flagship GPU is a cut-down version designed specifically for the Chinese market, though even the restricted version of this GPU was not able to satisfy new regulations.
RTX 5090D reportedly limited to 24GB after July
Previously, Nvidia sought to meet requirements by implementing hardware-level limits on its AI capabilities, but kept the general specs of the RTX 5090D the same as the international versions. Up until now, the GB202 Blackwell GPU powering the RTX 5090D had the same core count, memory configuration, and power limits as the standard RTX 5090.
According to a report from Korean hardware blogger Harukaze5719 on Twitter/X, this is about to change. A new version of the RTX 5090D will have only 24GB of VRAM, down from the 32GB available in the international model and the existing 5090D. This reportedly comes from Manli, an Nvidia-partnered GPU manufacturer.
Harukaze specifies that the price of the card will remain unchanged, which means that Chinese customers will be paying the same price for less. This new version of the card will be available from July, according to the report.
Previous restrictions on the H20 GPUs have cost Nvidia an estimated $5.5 billion this quarter alone, so it is not surprising that the company wants to keep this market open, but the continued uncertainty might drive customers in China towards AMD or Intel.
About the Author