And now, Jensen Huang is finally lifted the veil The RTX 5090, RTX 5080, and RTX 5070 family were all taken offline last night, with the first two due to arrive on January 30th and the RTX 5070 series scheduled for February. In compliance with US export restrictions, the company also announced the China-exclusive RTX 5090D, which surprisingly retains most of the specs of its global sibling – unless you’re an AI enthusiast.
The RTX 5090 is based on Nvidia Blackwell architecture with the flagship GB202 chip and 170 streaming multiprocessors supporting 21,760 CUDA cores. Nvidia has increased the video memory from 24GB to 32GB, and is also using faster and newer GDDR7 memory with 1,792TB/s bandwidth. US trade sanctions restrict the sale in China of any GPU with a TPP or total compute power greater than 4800. The Ada-based RTX 4090 exceeded this limit by only 10%, forcing Nvidia to introduce the RTX 4090D to the Chinese market.
According to Nvidia, the RTX 5090D has specs that are virtually identical to the global RTX 5090, right down to base and boost clocks. The only discrepancy exists in AI performance, which has been reduced from 3352 AI TOPS on the RTX 5090 to 2375 AI TOPS on the RTX 5090D; decrease by 29%. The RTX 5090D operates at the very edge of US export regulations with reported TPP reached 4750, just 50 points below the limit. Previous Nvidia sanctions are lifted narrowly caught the attention of US Commerce Secretary Gina Raimondo. It will be interesting to see if the RTX 5090D makes it to retail in China without further silicon cuts.
Blackwell provides hardware support for the FP4 data type while delivering significant performance gains per tensor core. WITH Generating multiple frames in DLSS 4 a feature exclusive to the RTX 50 series, completely disabling support for certain data types at the SM level seems unlikely. Despite the apparent setback, the RTX 5090D can deliver 31% more AI TOPS than the RTX 5080. Therefore, we doubt that the performance difference between the RTX 5090D and RTX 5090 when using AI-accelerated technologies such as frame generation will be a major issue . .
In general, the performance of a pure raster is unlikely to suffer. Nvidia hasn’t clarified the details of how the RTX 5090D complies with export regulations, so we may have to wait for the official document to find out more. The RTX 5090D is scheduled to retail in China on January 30, starting at RMB 16,499 ($2,250).