NVIDIA launched two new professional computing accelerator cards at SIGGRAPH 2025. The cards expand its Blackwell architecture range with the launch of the RTX PRO 4000 SFF Edition and RTX PRO 2000. These compact GPUs are designed to accelerate AI operations for professional tasks while remaining energy efficient.
The graphics card comes with fourth-generation RT cores and fifth-generation Tensor cores. They pack significant performance gains into a smaller form factor. The RTX PRO 4000 SFF Edition has 24 GB of ECC GDDR7 memory. It has a throughput of 770 AI TOPS. This means it handles AI tasks 2.5x faster than the older model (i.e. the RTX A4000) and has 1.7x better ray tracing performance. While bringing updated specs, NVIDIA managed to keep the TGP at the same 70 W as its predecessor.
RTX PRO 2000 is equipped with 16 GB ECC GDDR7 video memory, and its 545 AI TOPS computing power is targeted at mainstream design and AI workflows. Compared to NVIDIA RTX A2000, it delivers up to 1.6x faster 3D modeling, up to 1.4x faster CAD performance, and significantly faster rendering. AI performance has also been significantly improved, with image generation 1.4 times faster and text generation 2.3 times faster. Both GPUs are ideal for compact workstations with limited space or tight power budgets.
The RTX PRO 2000 will be available through PNY and TD SYNNEX, while the two graphics cards will be integrated into complete systems from manufacturers such as Dell, HP and Lenovo. The cards are expected to be available later this year, but exact pricing has not yet been announced.
RTX PRO 4000 Blackwell SFF Edition Specifications
GPU architecture: NVIDIA Blackwell
NVIDIA CUDA Core count: 8,960
Tensor Cores: Fifth Generation
Ray Tracing Core: Fourth Generation
GPU memory: 24 GB GDDR7 with ECC
Memory interface: 192 bits
Memory bandwidth: 432 GB/s
System interface: PCIe 5.0 x8
Display interface: 4x Mini DisplayPort 2.1b
Maximum simultaneous display:: 4x 3840x2160 @ 165 Hz / 2x 7680x4320 @ 100 Hz
Video Engine: 2x NVENC (9th gen), 2x NVDEC (6th gen)
Power consumption: Total board power: 70W
Cooling Solution: Active
Dimensions: 2.7" x 6.6" long, dual slot, low profile
Graphics API: DirectX 12, Shader Model 6.7, OpenGL 4.6, Vulkan 1.4
Computing API: CUDA 12.8, OpenCL 3.0
NVIDIA RTX PRO 2000 Blackwell Specifications
GPU architecture: Blackwell
NVIDIA CUDA: Core count 4,352
Tensor Cores: Fifth Generation
Ray Tracing Core: Fourth Generation
AI TOPS 545: AI TOPS
Single precision performance: 17 TFLOPS
RT core performance: 52 TFLOPS
GPU memory: 16 GB GDDR7 with ECC
Memory interface: 128 bit
Memory bandwidth: 288 GB/s
System interface: PCIe 5.0 x 8
Display interface: 4x mini DisplayPort 2.1b
Maximum number of simultaneous displays: up to four displays simultaneously
Video Engine: 1x NVENC (9th gen), 1x NVDEC (6th gen)
Power consumption: Maximum power consumption: 70W
Cooling Solution: Active
Dimensions: 2.7 inches high x 6.6 inches long, dual slots
Graphics API: DirectX 12, Shader Model 6.7, OpenGL 4.6, Vulkan 1.4
Computing API: CUDA 12.8, OpenCL 3.0