NVIDIA RTX 6000D made-for-China GPU: 84GB GDDR7, lower clocks, slower than RTX PRO 6000

NVIDIA is cooking up a new China-exclusive GPU with the new RTX 6000D Blackwell GPU, recently tested on Geekbench and teases what is happening under the hood.

2

The company has sliced the CUDA core count, VRAM capacity, and GPU clock speeds of its new RTX 6000D "Blackwell Pro for China" GPU. NVIDIA created the RTX 6000D after ever-increasing US export restrictions for AI hardware were imposed, with the RTX 6000D Blackwell GPU being a different SKU than the one available globally.

Inside, NVIDIA's new RTX 6000D GPU for China features 156 SMs or 19,968 CUDA cores, representing 17% less CUDA cores than the RTX PRO 6000, as well as less VRAM with 84GB of GDDR7 memory compared to 96GB GDDR7 on the full RTX PRO 6000. We should expect to see NVIDIA using 3GB GDDR7 memory modules on the RTX 6000D for China, indicating a 448-bit memory bus compared to the 512-bit memory bus on the RTX PRO 6000.

• Read more: AMD gets export licenses approval for China-bound Instinct MI308 GPUs
• Read more: Chinese companies prepared to pay $24,000 for NVIDIA's next-gen B30A AI GPU
• Read more: China to ban foreign chips being used for AI inference, but need NVIDIA for now
• Read more: NVIDIA B30A being prepped for China: faster than H20, half the performance of B300

The cut-down continues for the RTX 6000D in China, with GPU clocks at 2430MHz compared to 2600MHz on the full RTX PRO 6000, with an unknown TDP at the moment. On the performance side of things, the RTX 6000D reaches 390,656 points on the OpenCL benchmark test on Geekbench 6, considerably lower than the 450,000 to 500,000 points on the full RTX PRO 6000.