Built on the NVIDIA Blackwell architecture, NVIDIA RTX PRO 6000 Blackwell Server Edition combines AI acceleration and professional visual computing. It is designed for workloads ranging from generative and agentic AI to physical AI, scientific computing, rendering, 3D graphics, and video.
The NVIDIA RTX PRO 6000 Blackwell Server Edition is available as self-service GPU instances on FPT AI Factory. Try Now
Cutting-Edge Blackwell Architecture
The NVIDIA RTX PRO™ 6000 Blackwell Server Edition leverages NVIDIA’s latest Blackwell architecture, featuring 5th-gen Tensor Cores and 4th-gen Ray Tracing Cores. With 96 GB of GDDR7 memory and up to 1.597 TB/s of memory bandwidth, it delivers top-tier performance for AI, rendering, and compute workloads in a single GPU.
The larger memory capacity gives AI and visual computing workloads more room to work with large models, datasets, and complex 3D content. The GPU also delivers 120 TFLOPS of FP32 performance and 4 PFLOPS of FP4 Tensor Core performance.
For comparison, the previous-generation NVIDIA L40S comes with 48 GB of GPU memory and 864 GB/s of memory bandwidth. The RTX PRO 6000 Blackwell doubles the memory capacity while significantly increasing memory bandwidth.
| Specification | NVIDIA L40S | NVIDIA RTX PRO 6000 Blackwell |
| GPU memory | 48 GB | 96 GB |
| Memory type | GDDR6 | GDDR7 |
| Memory bandwidth | 864 GB/s | 1.597 TB/s |
| FP32 performance | 91.6 TFLOPS | 120 TFLOPS |
Combined with the new Transformer Engine and native FP4 support, this translates into up to 6x faster AI inference and up to 4x faster photorealistic rendering than the L40S — plus up to 8x higher inference throughput than the previous Hopper architecture.
The GPU also supports Multi-Instance GPU (MIG), letting multiple workloads or users share it safely, with PCIe Gen5 enabling fast data transfer for continuous, always-on cloud use.
Built for AI Inference and Physical AI
NVIDIA RTX PRO 6000 Blackwell brings together the capabilities needed for several demanding workloads on a single GPU platform. For AI inference, the NVIDIA RTX PRO 6000 Blackwell supports production-grade AI solutions through NVIDIA AI Enterprise software, including LLM inference, AI agents, generative AI, computer vision, and speech AI. With fifth-gen Tensor Cores and a fully integrated media pipeline, NVIDIA reports a large performance leap over the previous-generation L40S for multimodal agentic and generative AI applications.
For physical AI, the GPU delivers strong performance for OpenUSD-based 3D content, synthetic data generation, and robotics workflows, built on NVIDIA Omniverse, which enables generative physical AI applications for industrial digitalization. Its 4th-gen RT Cores also support RTX Mega Geometry, enabling up to 100x more ray-traced triangles for photoreal, physically accurate scenes.
NVIDIA RTX PRO 6000 Blackwell on FPT AI Factory
The NVIDIA RTX PRO 6000 Blackwell Server Edition is now available on FPT AI Factory through GPU Virtual Machine. Configurations are available from one GPU to eight GPUs. The configurations can support different stages of AI or visual computing workloads, from development and inference on a single GPU to larger multi-GPU deployments.
With the NVIDIA RTX PRO 6000 Blackwell GPU Cloud on FPT AI Factory, developers and enterprises can provision the GPU as a Virtual Machine and use it for AI inference, physical AI, simulation, rendering, and other GPU-intensive workloads.
Explore GPU Virtual Machine on FPT AI Factory: factory.fpt.ai/gpu-virtual-machine
