Nvidia’s RTX 4090D Dragon GPU Features
Nvidia unveiled the China-exclusive GeForce RTX 4090D, which meets US export laws. A 384-bit wide memory interface, 14,592 CUDA cores, 24GB of GDDR6X memory, and 425W power consumption characterize the new GPU.
The RTX 4090D has less CUDA cores and power draw than the 4090. CUDA cores dropped 12.8% to 14,592 (128 SMs to 114 SMs) and power draw down 5.9% to 425W from 450W in the RTX 4090D. No changes were made to the 384-bit wide bus, 24GB of GDDR6X memory, or 2.52 GHz boost clock. Only the base clock has been raised to 2.28 GHz from 2.23 GHz.
Due to geopolitics, Nvidia stopped selling the conventional RTX 4090 and other AI/HPC-focused GPUs to China, prompting the new consumer RTX 40 series GPU. The new regulations limit the semiconductor processors that chip manufacturers, including Nvidia, can export to those that meet certain performance standards established by the US government.
Total Processing Power (TPP) is the performance metric that is employed; it is computed by multiplying the number of bits by the maximum compute using TFLOPs or TOPS for a certain bit-length. The U.S. export restrictions allow a maximum threshold of 4,800. The RTX 4090 landed up on the ban list since it just so occurred that its performance level in this benchmark is 10% greater (5,286) than the regulation limit.
Nvidia and its AIB partners have not released direct performance comparisons between the 4090D and 4090, but it is a given that the new RTX 4090D has a TPP rating of precisely 4,800 or lower, allowing it to be marketed in China. This new model will only be available in China; it will not be available in other nations . It is possible that the 4090D will surface in other sanctioned states, but that is the only way it will probably leave China. The RTX 4090D is designed only to get around US export laws and provide China with the fastest consumer GPU that is permitted by law.