NVIDIA has announced that its new GPU, the RTX PRO™ 6000 Blackwell Server Edition, will be available in the world’s most popular enterprise servers through agreements with Cisco, Dell Technologies, HPE, Lenovo, and Supermicro. This move marks a significant step toward replacing architecture based solely on CPUs with GPU-accelerated computing platforms in corporate data centers.
The new 2U RTX PRO servers provide universal acceleration for workloads ranging from agent-based AI and content creation to data analysis, scientific simulations, graphics rendering, and physical and industrial AI applications.
“AI is reinventing computing for the first time in 60 years. What started in the cloud is now transforming the architecture of on-premises data centers,” said Jensen Huang, founder and CEO of NVIDIA.
Up to 45 Times More Performance Than CPU-Only Systems
According to NVIDIA, these systems offer:
- Up to 45x more performance and 18x greater energy efficiency compared to 2U CPU-only servers.
- Lower total cost of ownership (TCO) due to significant reductions in power consumption and optimized space and cooling.
- Direct integration with the NVIDIA AI Data Platform, a customizable reference design for building storage systems optimized for enterprise AI.
At the event, Dell showcased an update to its Dell AI Data Platform, integrated with NVIDIA’s reference design, along with the Dell PowerEdge R7725 (2U) configured with two RTX PRO 6000 GPUs, NVIDIA AI Enterprise, and NVIDIA Networking connectivity.
Blackwell Architecture for Critical Workloads
The new GPUs incorporate the latest innovations:
- Fifth-generation Tensor Cores and second-generation Transformer Engine supporting FP4, achieving up to six times more inference performance compared to the previous generation (NVIDIA L40S).
- Fourth-generation RTX for photorealistic rendering and visualization, with up to four times the performance of L40S.
- Advanced virtualization with NVIDIA Multi-Instance GPU (MIG), enabling up to four isolated instances per GPU.
- Increased energy efficiency per watt for sustainable operations.
Boosting Physical AI and Industrial Robotics
RTX PRO Servers power the development of physical AI through integration with NVIDIA Omniverse™ libraries and foundational models like NVIDIA Cosmos™, enabling:
- Digital twins for factories and robotic simulation.
- Massively parallel generation of synthetic data for training models.
- Simulation workflows up to four times faster than with L40S.
They also support advanced blueprints such as NVIDIA Metropolis’s new video search and summarization model, vision-language models, and synthetic generation extensions to enhance safety and productivity in industrial environments.
Optimization for Agent-Based AI and Reasoning Models
These servers are certified for NVIDIA AI Enterprise, a software layer that accelerates and secures AI development and deployment within enterprises. They are ideal for running AI agents with reasoning models like Llama Nemotron Super, delivering up to three times better performance per dollar on NVFP4 with a single RTX PRO 6000, compared to FP8 on NVIDIA H100. This results in more accurate reasoning with lower operational costs.
Availability and Partners
In addition to the initial five major manufacturers (Cisco, Dell, HPE, Lenovo, and Supermicro), other providers such as Advantech, ASUS, Foxconn, GIGABYTE, MSI, QCT, Wistron, and Wiwynn will also offer certified configurations.
- The 4U models with eight RTX PRO 6000 GPUs are already available.
- The mainstream 2U versions are expected to reach the market later this year.
FAQs
1. What distinguishes a traditional CPU server from an RTX PRO 6000 Blackwell server?
Blackwell GPU servers multiply performance and energy efficiency, enabling massive workloads in AI, rendering, and analysis that would be much slower and more expensive with CPUs alone.
2. What is physical AI?
Physical AI refers to artificial intelligence applied to robotic systems and real-world environments, capable of perceiving, reasoning, and acting in physical settings like factories, autonomous vehicles, or smart cities.
3. What is FP4 in AI?
FP4 is a four-bit precision format that accelerates AI inference by reducing power consumption and increasing speed without sacrificing accuracy in most cases.
4. Are RTX PRO Servers only for large corporations?
While designed for enterprise and data centers, they can also be implemented in research environments, universities, or startups with advanced AI needs.