NVIDIA RTX PRO 6000 96GB Blackwell Workstation Edition

NVIDIA RTX PRO 6000 96GB Blackwell Workstation Edition

NVIDIA logo

NVIDIA RTX PRO 6000 96GB Blackwell Workstation Edition graphics card

Fine-tuning trillion-parameter models, hyper-realistic film rendering, and computational fluid dynamics simulations – these are workloads that demand a GPU without compromise. The NVIDIA RTX PRO 6000 Blackwell Workstation Edition is engineered for exactly that. The NVIDIA Blackwell architecture, with 24064 CUDA cores, 5th-generation Tensor Cores with FP4 support, and 96GB of GDDR7 ECC memory with 1792GB/s bandwidth, delivers 4000 AI TOPS, 125 TFLOPS in FP32, and 380 TFLOPS of RT core performance. An innovative cooler with dual axial air paths dissipates the 600W TBP of heat without performance drops, so you can enjoy its full power without any limitations. For professionals who need the very best a workstation GPU has to offer today, the NVIDIA RTX PRO 6000 Blackwell Workstation Edition is the only answer.

Nvidia GeForce RTX 40

GDDR7 / 96GB

Operating memory
NVIDIA GPU

PCIe 5.0 x16

Bus interface
NVIDIA GPU

1× PCIe CEM5 16-pin

Auxiliary power
Nvidia GeForce RTX 40

Blackwell (GB202)

Chip architecture
Nvidia GeForce RTX 40

1200W

Recommended PSU power
NVIDIA RTX PRO 6000 96GB Blackwell Workstation Edition graphics card

Key features of the NVIDIA RTX PRO 6000 96GB Blackwell Workstation Edition graphics card

  • Features 96GB of GDDR7 ECC memory with 1792GB/s bandwidth on a 512-bit bus for fine-tuning trillion-parameter models, extensive scientific datasets, hyper-detailed 3D scenes, and concurrent multi-application workflows without memory constraints
  • NVIDIA Blackwell architecture with 24064 CUDA cores and a boost frequency of approximately 2595MHz achieves 125 TFLOPS in FP32 and 380 TFLOPS from its RT cores
  • 5th-generation Tensor Cores with FP4 support and DLSS 4 Multi Frame Generation achieve 4000 AI TOPS for local LLM fine-tuning, training domain AI models, and generative AI workflows
  • A powerful cooler with dual axial air paths dissipates 600W TBP of heat without thermal performance throttling
  • 4th-generation RT Cores with 380 TFLOPS performance and RTX Mega Geometry
  • Multi-Instance GPU enables you to partition the card into up to four isolated 24GB instances, two 48GB instances, or a single full 96GB instance
  • Four 9th-generation NVENC video encoders and four 6th-generation NVDEC decoders with support for 4:2:2 10-bit H.264 and HEVC ensure real-time production video workflows, streaming, and VFX pipelines
  • Powered by a 1× PCIe CEM5 16-pin connector with a 600W TBP in an extended-height dual-slot design measuring 5.4" × 12" (137 × 304.8mm); the recommended PSU power for a workstation build is at least 1200W
NVIDIA GPU

Memory that defines a new class of performance

A total of 96GB of GDDR7 ECC memory with 1792GB/s bandwidth on a 512-bit bus elevates the GPU into a class previously reserved for server infrastructure. Entire billion-parameter models remain in memory without distribution across multiple cards, film scenes are processed without overflowing into system RAM, and concurrent workflows run smoothly even at the full 600W power profile. ECC protection actively eliminates computational errors. This makes every fine-tuning, simulation, and rendering result precise and trustworthy – whether it's a domain-specific LLM, computational fluid dynamics, or a real-time film sequence.

NVIDIA GPU

A new standard for local AI infrastructure

Performance of 4000 TOPS combined with FP4 precision support in the 5th-generation Tensor Cores elevates the NVIDIA RTX PRO 6000 96GB Blackwell Workstation Edition into a category once exclusive to data centres. The card handles LLM fine-tuning, generative AI, training of domain-specific models, and inference of natively large networks locally, with full data control and without recurring cloud costs. DLSS 4 Multi Frame Generation then adds up to three times higher frame rates in supported rendering applications and simulations, while the AI Management Processor ensures intelligent distribution of computing resources across workloads.

NVIDIA GPU

RTX Mega Geometry and neural shaders

Until now, ray tracing has been limited by the number of polygons a GPU could process in real time. The 4th-generation RT Cores with RTX Mega Geometry technology raise this limit by an order of magnitude. The NVIDIA RTX PRO 6000 96GB Blackwell Workstation Edition can handle up to 100× more ray-traced triangles than the previous generation, paving the way for physically accurate scenes with billions of polygons without offline rendering. New neural shaders integrate neural networks directly into programmable shaders, so AI is no longer a post-processing layer but a natural part of the graphics pipeline. For architects, VFX artists, product designers, and researchers, this means real-time visualisations that faithfully reflect physical reality.

Nvidia GeForce RTX 40

NVIDIA Blackwell architecture

A breakthrough generation for AI, neural rendering, and scientific computing
Nvidia GeForce RTX 40

4000 TOPS of AI performance

Local LLM fine-tuning and generative AI at a data centre level
Nvidia GeForce RTX 40

96GB GDDR7 ECC

Double the capacity and throughput compared to the previous professional generation
Nvidia GeForce RTX 40

5th generation Tensor cores

FP4 precision and DLSS 4 Multi-Frame Generation for maximum AI acceleration
Nvidia GeForce RTX 40

4th generation RT cores

RTX Mega Geometry for physically accurate scenes with billions of polygons
Nvidia GeForce RTX 40

4× DisplayPort 2.1b

Up to 8K at 240Hz or 16K at 60Hz for precise visualisation workflows
Nvidia GeForce RTX 40

Multi-Instance GPU

Share the power of a single card securely between four users or workloads
Nvidia GeForce RTX 40

Max-Q for 4GPU configurations

300W in a dense design for scalable AI infrastructure on a workstation

Specifications

Chip

Chip manufacturer NVIDIA
Model designation RTX PRO 6000
Graphics processor GB202
Chip architecture Blackwell
Number of stream processors 24,064×
Production technology 4 nm

Performance & usage

Performance Maximum
Use Professional

Memory

VRAM capacity 96 GB
Memory type GDDR7
Bus width 512-Bit

Frequency

Core frequency 1,590 MHz
Core clock speed (Boost clock) 2,288 MHz
Memory frequency 28,000 MHz (28 GHz)
Effective memory clock (bandwidth) 28 Gbps

Interface

PC connection PCI Express x16 5.0
Bus interface PCIe 5.0 x16
Monitor connection DisplayPort 2.1b

Technology

Special graphics card features Tensor Cores
DirectX version 12 Ultimate

Cooling system

Cooling type Active
Cooling size Dual-slot
number of fans 1 ×

Dimensions

Width 304.8 mm
Depth 137 mm

Consumption

TDP 600 W

AI

Total number of TOPS 4,000
Software platform NVIDIA CUDA
On-device AI processing GPU-accelerated AI
Supported precision FP32

Tops

Total number of TOPS 4,000

AI specifications

Primary AI use LLM Inference
CUDA version 12.x
Code:  NVrpr01
Warranty: 36 months
Links: Producer's Website: