Nvidia has presented two new professional graphics cards (RTX PRO 4000 SFF y RTX PRO 2000) at the Siggraph 2025 conference. The company thus expands its offer of solutions under the latest generation architecture ‘Blackwell’.
These GPUS are below in the performance of the RTX Pro 6000 series, since they are designed for use in Compact work stations and seek to accelerate AI for professional tasks, while maintaining energy efficiency. These graphics cards arrive with fourth generation RT nuclei and fifth -generation tensioning centers and incorporate significant performance improvements in smaller formats.
RTX PRO 4000 SFF
It is a small version of the 4000 Blackwell RTX card. The company states that it offers more than double the performance than the RTX A4000 SFF of the previous generationwith improved ray layout capabilities and 50% more bandwidth. And all this maintaining the same maximum TDP of 70 watts of the previous model.
With 24 GB of ECCDR7 memory and a yield of 770 Ia tops, this GPU could be an ideal option for professionals they are looking for Power in a more compact format. These characteristics make it especially useful for work stations with limited space, reduced in energy and temperature consumption, and adjusted in budgets.
Specs:
- GPU Architecture: Nvidia Blackwell
- NÚCLEOS NVIDIA CUDA: 8960
- Tensioner nuclei: 5th generation
- Lightning Center: 4th Generation
- GPU memory: 24 GB GDDDR7 with etc.
- Memory interface: 192 bits
- Memory bandwidth: 432 GB/s
- System interface: pcie 5.0 x8
- Screen connectors: 4x mini displayport 2.1b
- Maximum simultaneous screens: 4x 3840 × 2160 at 165 Hz / 2x 7680 × 4320 to 100 Hz
- Video engines: 2x Nvent (9th generation), 2x NVDEC (6th generation)
- Energy consumption: Total plate power: 70 w
- Form factor: 2.7 ″ X 6.6 ″ L, double slot, medium height
- GRAPHIC API: Directx 12, Shader Model 6.7, OpenGL 4.6, Vulkan 1.4
- PEPER API: CUDA 12.8, OPENCL 3.0
RTX PRO 2000
The second card presented includes 16 GB of dedicated memory ECC GDDR7 and offers 545 AI power tops With the same TGP of 70 W. Nvidia states that this card is designed for conventional design and IA workflows, promising that it is approximately 1.5 times faster than the RTX A2000 in 3D modeling, a computer -assisted design and rendered. The company also states that it is useful for the generation of images and text with AI, which improves its performance so that work flows based on IA are faster and fluid.
Specs:
- GPU Architecture: Blackwell
- NVIDIA CUDA: 4352 NÚCLEOS
- Tensioner nuclei: 5th generation
- Lightning Center: 4th Generation
- GPU memory: 16 GB GDDDR7 with etc.
- Memory interface: 128 bits
- Memory bandwidth: 288 GB/s
- System interface: pcie 5.0 x 8
- Screen connectors: 4x mini displayport 2.1b
- Energy consumption: maximum energy consumption: 70 w
- Form factor: 2.7 ″ high x 6.6 ″ long, double slot
The new NVIDIA cards support the main graphics APIs, such as Directx 12, Shader Model 6.7, OpenGL 4.6, Vulkan 1.4, Cuda 12.8 and OpenCl 3.0. They are expected to be available At the end of 2025 through suppliers such as PNY and TD Synnex, and also in complete manufacturers such as Dell, HP and Lenovo.
More information | Nvidia blog
