Initially, NVIDIA launched the RTX PRO 5000 Blackwell with 48GB of GDDR7 memory, which means it used a total of 24 modules at 2GB capacity. The new version uses the same design, but switches to 3GB modules. This makes the card much more interesting to professionals in data science, AI, HPC, and other markets.

Based on the GB202 GPU, the RTX PRO 5000 Blackwell packs 14,080 CUDA cores, 176 ROPs, and 440 TMUs, 5th generation Tensor and 4th generation ray tracing cores, three 9th gen NVENC encoders, three 6th gen NVDEC decoders, and the same 300W maximum power consumption. There appears to be an error in NVIDIA's own spec sheet, as the 72GB version should have a 384-bit memory interface.

As said, the design is the same, which uses a blower-style dual-slot cooler.