NVIDIA unveiled two new professional graphics cards at SIGGRAPH 2025, expanding its Blackwell architecture family with the ´RTX PRO 4000 SFF Edition´ and the ´RTX PRO 2000´. Both cards are built to accelerate Artificial Intelligence workloads and ray tracing while keeping power draw and physical size low for compact workstations. They include fourth-generation RT cores and fifth-generation Tensor cores, which underpin the company´s quoted gains in rendering and model throughput.
The larger of the two, the ´RTX PRO 4000 SFF Edition´, packs 24 GB of ECC GDDR7 memory and delivers 770 Artificial Intelligence TOPS of throughput. NVIDIA positions it as a direct uplift over the previous RTX A4000, claiming up to 2.5 times faster performance on Artificial Intelligence tasks and 1.7 times better ray tracing. Notably, the new SFF edition retains a 70 W total graphics power rating, matching the card it replaces while offering a substantial step up in compute density for constrained power envelopes.
The ´RTX PRO 2000´ aims at mainstream design and Artificial Intelligence workflows with 16 GB of ECC GDDR7 memory and 545 Artificial Intelligence TOPS. NVIDIA´s comparisons to the RTX A2000 show up to 1.6 times faster 3D modeling, 1.4 times faster CAD workloads, improvements in rendering, 1.4 times quicker image generation, and 2.3 times faster text generation. Both cards are targeted at small form-factor workstations where space and thermal budgets are limited. Distribution plans list the RTX PRO 2000 through PNY and TD SYNNEX, while both cards will appear in complete systems from manufacturers including Dell, HP, and Lenovo. NVIDIA expects the cards to ship later this year, but specific pricing details have not been disclosed.
