Nvidia unveils compact Blackwell professional graphics cards at SIGGRAPH 2025: RTX Pro 4000 SFF and Pro 2000 GPUs make an appearance
At SIGGRAPH 2025, NVIDIA has launched two new GPUs designed for professionals: the RTX PRO 4000 SFF and RTX PRO 2000. These GPUs, based on the Blackwell architecture, are geared towards compact workstations with AI and ray tracing acceleration.
NVIDIA RTX PRO 4000 SFF: Key Specifications and Performance
- Architecture: NVIDIA Blackwell
- CUDA Cores: 8,960
- Tensor Cores: 5th generation (280 Gen 5 Tensor Cores)
- RT Cores: 4th generation (70 Gen 4 RT Cores)
- Memory: 24 GB ECC GDDR7
- Memory Bandwidth: 432 GB/s (50% more than previous gen)
- AI Throughput: About 770 AI TOPS
- Power Consumption: 70 W TDP
- Form Factor: Low-profile, small form factor (dual-slot, compact size)
- Connectivity: Four Mini DisplayPort 2.1b connectors
- PCIe Interface: PCIe Gen 5
- Additional Features: 9th Gen NVENC, 6th Gen NVDEC supporting 4:2:2 codec; AI Management Processor for optimized AI workloads
The RTX PRO 4000 SFF is the most powerful low-profile professional graphics card, offering more than twice the AI performance of the previous RTX A4000 SFF. It boasts improved ray tracing and efficiency, making it suitable for design studios, engineering firms, financial and industrial applications requiring AI inference, neural rendering, and photorealistic real-time visualization in space-constrained environments.
NVIDIA RTX PRO 2000: Key Specifications and Performance
- Memory: 16 GB ECC GDDR7
- AI Throughput: 545 AI TOPS
- Power Consumption: 70 W TDP
- Designed for mainstream design, AI image and text generation, 3D modeling, CAD, and rendering.
- Compared to the previous RTX A2000, it offers approximately 1.5x faster performance in relevant workloads.
The RTX PRO 2000 is tailored for mainstream design and AI workflows, offering a balance of power efficiency and performance.
Expected Availability
Both GPUs were announced and unveiled at SIGGRAPH 2025 (August 12, 2025). Detailed availability dates have not been explicitly published, but given the announcement timing at a major industry event, availability is likely imminent or within the latter half of 2025.
Summary Table
| Feature | RTX PRO 4000 SFF | RTX PRO 2000 | |--------------------------|--------------------------------|--------------------------------| | Architecture | NVIDIA Blackwell | NVIDIA Blackwell | | CUDA Cores | 8,960 | Not specified | | Tensor Cores | 280 (5th Gen) | Not specified | | RT Cores | 70 (4th Gen) | Not specified | | Memory | 24 GB ECC GDDR7 | 16 GB ECC GDDR7 | | Memory Bandwidth | 432 GB/s | Not specified | | AI Throughput (TOPS) | 770 | 545 | | Power Consumption (TDP) | 70 W | 70 W | | Form Factor | Low-profile, small form factor | Compact workstation form | | Target Workloads | High-end AI, ray tracing, neural rendering, complex datasets | Mainstream design, AI generation, 3D modeling, CAD, rendering | | Launch Event | SIGGRAPH 2025 (Aug 12, 2025) | SIGGRAPH 2025 (Aug 12, 2025) |
These GPUs, with their balanced power efficiency and performance, are ideal for professionals needing powerful AI and graphics acceleration in compact and power-limited workstation environments. Pricing information for the RTX PRO 4000 SFF and RTX PRO 2000 is not yet available. The RTX PRO 2000 and RTX PRO 4000 SFF will be available through PNY and TD Synnex for individual card upgrades, as well as from system integrators including Boxx, Dell, HP, and Lenovo for complete workstation solutions. The RTX Pro 4000 SFF will also be available in systems from Nvidia partners like Dell, HP, and Lenovo.
These new GPUs from NVIDIA, the RTX PRO 4000 SFF and RTX PRO 2000, are not only designed for professionals but also integrate artificial-intelligence capabilities, making them a blend of technology and gadgets. With the launch at SIGGRAPH 2025, these GPUs, based on the Blackwell architecture, offer enhanced AI performance and ray tracing acceleration, catering to a variety of fields like design studios, engineering firms, and even financial and industrial applications.