The NVIDIA RTX PRO™ 6000 Blackwell Max-Q 96GB Workstation Edition represents NVIDIA's top-tier professional desktop graphics architecture. Built on the NVIDIA Blackwell microarchitecture, the Max-Q variant is engineered specifically for high-density, multi-GPU workstation setups, delivering enterprise-class performance in a power-efficient 300 W dual-slot envelope.
Technical Specifications Overview
| Feature |
Specification |
| GPU Architecture |
NVIDIA Blackwell |
| CUDA Cores |
24,064 |
| Tensor Cores |
752 (5th Generation) |
| Ray Tracing Cores |
188 (4th Generation) |
| VRAM Capacity |
96 GB GDDR7 with ECC support |
| Memory Bus & Bandwidth |
512-bit / Up to 1,792 GB/s (1.8 TB/s) |
| Peak FP32 Performance |
110 TFLOPS |
| Peak FP4 AI Performance |
3.51 PFLOPS (with Sparsity) |
| Power Consumption (TDP) |
300 W |
| Form Factor |
Dual-slot, Full-Height / Full-Length (Active Blower Fan) |
| Host Interface |
PCIe 5.0 x16 |
| Display Connectors |
4x DisplayPort 2.1b |
Architectural Highlights
1. 5th-Generation Tensor Cores & FP4 Precision
Equipped with 752 5th-Gen Tensor Cores, the card introduces native hardware support for FP4 precision alongside FP8, FP16, BF16, and TF32. This provides up to 3x the AI processing throughput over the previous generation, making it possible to prototype and run large language models (LLMs) and agentic AI models locally without immediately relying on cloud infrastructure.
2. Neural Shading & 4th-Gen RT Cores
The 188 4th-Gen Ray Tracing Cores feature RTX Mega Geometry acceleration, doubling ray-triangle intersection throughput over previous Ada Lovelace architectures. Coupled with new neural shaders embedded directly into the Streaming Multiprocessors (SM), it enables DLSS 4 Multi-Frame Generation for real-time 3D simulation and photorealistic rendering.
3. High-Density Scalability & Multi-Instance GPU (MIG)
-
Max-Q Power Envelope: By capping maximum power at 300 W, workstations can host up to four RTX PRO 6000 Max-Q GPUs in a single chassis, combining up to 384 GB of total GDDR7 memory.
-
Universal MIG: Allows a single 96GB GPU to be partitioned into up to 4 fully isolated hardware instances (24 GB per instance), complete with dedicated compute, cache, and memory for multi-tenant or concurrent task execution.
4. Next-Gen Video Engines
Features Ninth-Gen NVENC and Sixth-Gen NVDEC engines with full hardware acceleration for 4:2:2 H.264/HEVC as well as AV1 encoding and decoding, catering to broadcast-grade virtual production and real-time video processing.
Target Use Cases
-
Agentic & Generative AI: Local fine-tuning and deployment of multi-billion parameter LLMs, diffusion models, and autonomous AI agents.
-
Enterprise Data Science: Large-scale dataset analytics using CUDA-X acceleration (RAPIDS) with 96 GB VRAM headroom per card.
-
3D Design & Simulation: Complex CAD, BIM (AECO), and real-time ray-traced rendering on platforms like NVIDIA Omniverse.
-
High-Density Workstation Compute: Scalable multi-GPU setups where thermal and power efficiency are essential.
