NVIDIA´s upcoming RTX PRO 6000 ´Blackwell´ workstation-class GPU has emerged in fresh benchmark results, revealing underwhelming early performance relative to the gaming-focused GeForce RTX 5090. Powered by the new GB202 GPU, the professional card is equipped with 24,064 CUDA cores across 188 streaming multiprocessors and achieves boost clocks up to 2,617 MHz. The board features 96 GB of next-generation GDDR7 memory with full error-correcting code, implemented through innovative dual-sided 3 GB modules, targeting high-end visualization and compute applications.
Initial Geekbench 6.4.0 OpenCL testing placed the RTX PRO 6000 Blackwell at a score of 368,219, falling short of the GeForce RTX 5090´s 376,858, despite the workstation card´s superior core count (24,064 vs. 21,760) and higher boost frequency (2,617 MHz vs. 2,410 MHz). Subtest analysis showed the RTX PRO lagging notably in background blur and face detection tasks but leading slightly in horizon detection and Gaussian blur, indicating a mixed performance profile across varied workloads. These early results have been linked to pre-release driver limitations, a temporary cap on addressable memory (currently set at 23.8 GB), and conservative power settings restricting the hardware´s full potential.
The RTX PRO 6000 family introduces distinct models optimized for differing deployment scenarios. The Max‑Q Workstation Edition is designed with a 300 W thermal design point for relatively compact work environments, maintaining the full core and memory specifications but dialing back clocks and voltages for quieter, more efficient operation. In contrast, the standard Workstation and Server editions accommodate a higher thermal limit of up to 600 W, unlocking increased sustained performance for compute-intensive and persistent workloads in desktop and rack-mounted systems. As the RTX PRO 6000 approaches its official release, expectations are that new drivers will better leverage its hardware capabilities, especially for demanding OpenCL and Artificial Intelligence software in professional contexts.
