Nvidia has revealed the PCB design of the RTX 6000D, a Blackwell-based graphics card created exclusively for the workstation and server market in China.
This model appears as a response to current trade restrictions, featuring an unusual but still powerful memory configuration.
Unusual Memory Configuration With 84 GB Of VRAM
The most interesting detail from the leak, which comes from a teardown of the GPU, is the layout of its memory modules. Unlike the standard RTX PRO 6000 Blackwell, which uses a full configuration of 32 memory chips, the RTX 6000D uses 28 GDDR7 memory chips, each with 3 GB capacity.
This leaves four empty slots on the PCB, resulting in a total of 84 GB of VRAM.
The reduction in memory is not random. By removing these modules, the memory interface is reduced to 448-bit, instead of the 512-bit interface found on the PRO version. Even with this cut, 84 GB of VRAM is still more than enough for demanding tasks such as artificial intelligence workloads and high-end rendering.
In terms of core count, the RTX 6000D features 19,968 CUDA cores, slightly lower than the 24,064 CUDA cores available on the RTX PRO version.
The official TDP for the server edition is rated at 600W, but early tests show actual power consumption closer to 419W under load. This suggests either improved power efficiency or firmware limits that reduce real-world consumption.
Another important detail is cooling. The card comes with a passive cooling system designed for data centers. However, some manufacturers are replacing this passive block with custom liquid cooling solutions to improve thermal performance.
The RTX 6000D shows Nvidia’s effort to provide high-performance solutions while complying with restrictions imposed by the United States. Even with some limitations compared to the PRO model, it remains a powerful option for AI and professional workloads in the Chinese market.
