For the second generation in a row, Nvidia's flagship 90-series GPU
isn't officially available in China. Like the RTX 4090 gaming GPU before it, the RTX 5090 was blocked under U.S. export controls as it exceeds the control thresholds for high-performance AI-chips. These rules restrict high-performance Nvidia graphics cards that exceed thresholds in memory bandwidth, power draw, and overall AI throughput from entering the Chinese market.This began with the RTX 4090D, which Nvidia launched as a China-only
model. The GPU had intentionally lower Tensor cores, CUDA cores, and power draw to comply with the export rules, which limited AI capabilities while keeping gaming performance mostly intact. Now, with the 50-series, the company initially launched an RTX 5090D for the Chinese market with lower Tensor Core AI TOPs than the 5090s available in the global market. The model launched in early 2025, but as restrictions tightened further, it too was reportedly pulled from sale.
Reports now point to a new workaround in development: the RTX 5090D V2. While we await official confirmation from Nvidia, insiders suggest this version will offer further cutbacks to stay compliant -- and still deliver power to the Chinese market.
Read more: The Most Expensive Game Consoles Of All Time, Ranked By Launch Price
China May Be Getting An Even Weaker RTX 5090D This Time
The original RTX 5090D (V1) featured the same 21760 CUDA cores and 32 GB DDR7 memory as the global RTX 5090, but to meet official U.S. government regulations, Nvidia cut its AI performance from 3352 to 2375 TOPs. However, even those reductions weren't enough. As a report by Tom's Hardware suggests, Nvidia reportedly instructed brand partners to cancel any outstanding orders and halt shipment of the RTX 5090D in China. While Nvidia has yet to confirm the ban, the abrupt halt in distribution lines, combined with recent supply chain leaks, has fueled speculation about an upcoming China-exclusive 5090D V2 (also being referred to as the 5090DD).
The "leaked" RTX 5090D V2, reportedly, will swap out the GB202-250 die for a new GB202-240 variant. It also cuts the VRAM from 32GB to 24GB and reduces the memory bus from 512-bit to 384-bit, which drops bandwidth from 1.79 TB/s to 1.34 TB/s to match the newly established 1.4 TB/s limit for the Chinese market. Gaming performance reportedly will stay on par with the U.S. RTX 5090 because nothing other than these two metrics seems to have changed -- not even the power pull. The tradeoff, however, is in the creator workloads and AI acceleration -- areas where reduced VRAM and bandwidth could make a measurable difference.
Importantly, Nvidia hasn't publicly confirmed the existence or launch of the RTX 5090D V2, but rumors and leaks report an announcement by August 5, with sales expected to begin the weeks that follow.
Want the latest in tech and auto trends? Subscribe to our free newsletter for the latest headlines, expert guides, and how-to tips, one email at a time.
Read the original article on SlashGear.
