Nvidia planning new RTX 5090 'DD' variant for China — 24GB card with tweaked GPU latest attempt to comply with strict export restrictions
Reported specifications suggest a 25% cutback in GPU bandwidth.

It appears the recent U.S. export restrictions, initially targeting H20 accelerators, have now been extended to Nvidia's consumer-grade RTX 5090D for China. As such, the company is reportedly developing a further cut-down option, the RTX 5090DD, as an alternative. Reported details from MEGAsizeGPU and specifications by Kopite indicate a significant reduction in GPU bandwidth, which is a trend that might extend to other finer aspects as well, like ROP counts and AI TOPs.
Last month, reports alleged that Nvidia is halting deliveries of its RTX 5090D GPUs to China, following the recent ban on the export of its H20 chips. Washington's primary concern stemmed from the potential of these GPUs, particularly their high GPU bandwidth, to be used in China's supercomputing efforts. On the server side, Nvidia is reportedly pivoting to a Blackwell-based design with GDDR7 memory as the successor to the H20, though specific details remain under wraps.
RTX 5090 DD?PG145-SKU40GB202-240-K*-A121760FP32384-bit GDDR7 24G 28Gbps575Wand there's a surprise.June 18, 2025
The RTX 5090D launched just before the Chinese New Year, with 'D' referring to last year's 'Dragon' Chinese Zodiac. However, with the current year being the Year of the Snake, the RTX 5090DD's naming feels a bit behind the times, perhaps an RTX 5090DS would've been a better fit, but I digress. Kopite claims the RTX 5090DD will feature the PG145 board as a reference design for board partners, as opposed to Nvidia's compact PCB (PG144A) for its FE (Founders Edition) class models.
Specifications-wise, according to the leak, we're looking at the GB202-240 chip, a tier below GB202-250 that was found on the original RTX 5090D. The core count remains similar, at 21,760 CUDA cores, or 170 Streaming Multiprocessors. The memory-bus width takes a significant hit, shrinking from 512 bits to a 384-bit interface. This reduction also slashes the available memory from 32GB to 24GB, and consequently also lowers the GPU bandwidth to 1344 GB/s, 25% slower than the original RTX 5090 and RTX 5090D. The TDP remains similar at 575W, with power likely to be supplied by the same 12V-2x6 connector.
GPU | Die | CUDA Cores | SMs | Bus-Width | VRAM | Bandwidth |
---|---|---|---|---|---|---|
RTX 5090 | GB202-300 | 21760 | 170/192 | 512-bit | 32GB | 1792 GB/s |
RTX 5090D | GB202-250 | 21760 | 170/192 | 512-bit | 32GB | 1792 GB/s |
RTX 5090DD (Leaked) | GB202-240 | 21760 | 170/192 | 384-bit | 24GB | 1344 GB/s |
RTX 5080 | GB203-400 | 10752 | 84/84 | 256-bit | 16GB | 960 GB/s |
RTX 5070 Ti | GB203-200 | 8960 | 70/84 | 256-bit | 16GB | 896 GB/s |
RTX 5070 | GB205-300 | 6144 | 48/50 | 192-bit | 12GB | 672 GB/s |
RTX 5060 Ti | GB206-300 | 4352 | 34/36 | 128-bit | 8GB | 448 GB/s |
RTX 5060 Ti | GB206-300 | 4352 | 34/36 | 128-bit | 8GB | 448 GB/s |
RTX 5060 | GB206-250 | 3840 | 30/36 | 128-bit | 8GB | 448 GB/s |
The leaker suggests a 'surprise', likely hinting at further cutbacks on other hardware specifications, such as ROP units or AI TOPs (probably via firmware). We'll wait for the official announcement or reviews for further details on pricing and performance.
Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.

Hassam Nasir is a die-hard hardware enthusiast with years of experience as a tech editor and writer, focusing on detailed CPU comparisons and general hardware news. When he’s not working, you’ll find him bending tubes for his ever-evolving custom water-loop gaming rig or benchmarking the latest CPUs and GPUs just for fun.