NVIDIA Introduces RTX 4090 D: China‑Specific GPU with Reduced CUDA and Tensor Cores

Because of export control rules set by the U.S. Department of Commerce, the consumer‑grade flagship graphics card RTX 4090 cannot be sold in China, so NVIDIA created a special version for the Chinese market called the NVIDIA RTX 4090 D.

This card complies with the TPP (Total Processing Power) metric required by the export regulations; to achieve this NVIDIA trimmed the number of CUDA cores and Tensor cores while leaving the rest of the configuration identical to the standard RTX 4090.

The performance gap in the TPP metric is only about 10 % (the limit is 4800, while the RTX 4090 scores 5286), so NVIDIA simply reduced that 10 % to bring the RTX 4090 D under the limit.

Key differences between RTX 4090 D and RTX 4090:

GPU core changed from Ada Lovelace AD102‑300 to AD102‑250

CUDA cores reduced from 16,384 to 14,592

Tensor/RT cores reduced from 512/128 to 456/128

Base clock increased from 2230 MHz to 2280 MHz

Power consumption lowered from 450 W to 425 W

FP32 performance: pending

RT TFLOPS: pending

Tensor‑TOPS: pending

The special‑edition card is priced the same as the regular RTX 4090, with a starting price of ¥12,999, meaning there is no price reduction despite the hardware downgrade.

Other key specifications (identical to RTX 4090):

Chip process: TSMC 4N

Transistor count: 76 billion

L2 cache: 72 MB

Memory: 24 GB GDDR6X

Memory bus width: 384‑bit

Memory speed: 21.0 Gbps

Bandwidth: 1008 GB/s