96GB ECC Memory
The 96GB framebuffer is the strongest reason to choose RTX PRO 6000 Blackwell Workstation Edition over lower-tier workstation and consumer GPUs.
$7,950.00
RTX PRO 6000 Blackwell Workstation Edition is NVIDIA’s high-memory Blackwell professional workstation GPU with 96GB GDDR7 ECC, a 512-bit memory interface, 1,792 GB/s bandwidth, 24,064 CUDA cores and up to 600W total board power. It is designed for advanced AI development, large-model inference, rendering, simulation, visualization and other workstation workloads that need very large GPU memory.
Use the live price shown on this page for current RTX PRO 6000 price information. For B2B purchasing, confirm the exact NVIDIA board, condition, warranty coverage, included scope, destination and delivery terms in the written quotation before ordering.
Related products, categories and research.
Flagship 96GB Blackwell Professional GPU
The NVIDIA RTX PRO 6000 Blackwell Workstation Edition is NVIDIA’s full-power flagship professional desktop GPU for advanced AI development, local large-language-model workloads, rendering, simulation, scientific visualization, CAD/CAE, virtual production and high-end content creation.
For buyers searching for an RTX PRO 6000 for sale, the exact edition matters. This listing is for the NVIDIA RTX PRO 6000 Blackwell Workstation Edition with 96GB GDDR7 ECC memory and a 600W maximum board-power rating. It is not the 300W Max-Q Workstation Edition and it is not the passively cooled Server Edition.
NVIDIA specifies 24,064 CUDA cores, fifth-generation Tensor Cores, fourth-generation RT Cores, up to 4,000 AI TOPS, 96GB of GDDR7 memory with ECC and approximately 1,792 GB/s memory bandwidth.
The live WooCommerce price above is the current store price. Request a written offer for quantity purchases, the exact unit condition and destination-specific delivery terms.

Product overview
The RTX PRO 6000 Blackwell Workstation Edition is designed for professional users who need very large GPU memory, enterprise workstation software support and maximum single-GPU workstation performance.
Its defining hardware feature is the 96GB GDDR7 ECC framebuffer. That memory capacity allows unusually large AI models, 3D scenes, engineering datasets, scientific visualizations and production assets to remain resident on the GPU instead of being divided across smaller cards or repeatedly transferred to system memory.
The Workstation Edition uses NVIDIA’s Blackwell architecture with 24,064 CUDA cores, fifth-generation Tensor Cores and fourth-generation RT Cores. NVIDIA rates AI performance at up to approximately 4,000 AI TOPS.
The card connects through PCI Express 5.0 x16 and uses an extended-height, dual-slot double-flow-through cooler designed for a 600W maximum board-power envelope.
1
Confirm the Workstation EditionVerify that the quotation states the 600W Workstation Edition rather than Max-Q or Server Edition.
2
Confirm workstation compatibilityCheck physical clearance, PCIe layout, power delivery, 16-pin cabling and chassis airflow before ordering.
3
Confirm condition and warrantyThe final offer should state new, open-box or other condition, exact part identification, warranty and supplied accessories.
Buyer value
The 96GB framebuffer is the strongest reason to choose RTX PRO 6000 Blackwell Workstation Edition over lower-tier workstation and consumer GPUs.
Fifth-generation Tensor Cores provide Blackwell-generation AI acceleration including FP4-capable workflows where supported.
RTX professional drivers and workstation certification are designed for production environments where application stability and software support matter.
The 512-bit GDDR7 interface provides approximately 1,792 GB/s bandwidth for high-throughput AI, rendering and data workloads.
Four ninth-generation encoders and four sixth-generation decoders provide substantial media-processing capability.
Supported environments can partition the GPU into isolated 24GB, 48GB or full 96GB instances.
Manufacturer specifications
These specifications refer specifically to the NVIDIA RTX PRO 6000 Blackwell Workstation Edition.
| Manufacturer | NVIDIA |
|---|---|
| Product | RTX PRO 6000 Blackwell Workstation Edition |
| Architecture | NVIDIA Blackwell |
| CUDA cores | 24,064 |
| Tensor Cores | 5th Generation |
| RT Cores | 4th Generation |
| AI performance | Up to 4,000 AI TOPS |
| FP32 performance | Approx. 125 TFLOPS manufacturer rating |
| RT performance | Approx. 380 TFLOPS |
| GPU memory | 96GB GDDR7 with ECC |
| Memory interface | 512-bit |
| Memory bandwidth | 1,792 GB/s |
| System interface | PCI Express 5.0 x16 |
| Display outputs | 4 × DisplayPort 2.1b |
| NVENC | 4 × 9th Generation |
| NVDEC | 4 × 6th Generation |
| MIG | Up to 4 × 24GB, 2 × 48GB or 1 × 96GB instances where supported |
| Maximum board power | 600W |
| Power connector | 1 × PCIe CEM5 16-pin |
| Dimensions | 5.4 in H × 12.0 in L |
| Form factor | Dual-slot, extended height |
| Cooling | Double-flow-through |
| DirectX | DirectX 12 |
| OpenGL | OpenGL 4.6 |
| Vulkan | Version depends on supported driver; verify intended software stack |
| CUDA | CUDA-compatible; public launch datasheet referenced CUDA 12.8 |
| OpenCL | OpenCL 3.0 |
Check the supported NVIDIA professional driver branch, operating system and application version before deployment. API support can change with the driver release. Match any Vulkan, CUDA or other API requirement to the documentation for your intended software stack rather than assuming a launch datasheet describes every later release.
Peak TOPS and TFLOPS are theoretical ratings, not an application performance guarantee. NVIDIA defines the 4,000 AI TOPS reference as theoretical FP4 performance using sparsity. Compare your actual framework, precision and model before using that figure in a purchase decision.
Check the renderer version, scene memory and support for the current Blackwell architecture. A scene that exceeds available GPU memory can behave very differently from a small benchmark. Use representative assets and the intended quality settings.
Check the application’s support for the required codec, bit depth and media engines. The GPU’s encoding and decoding capabilities do not imply that every software edition or workflow exposes every acceleration feature.
Estimate model weights, context-related memory and runtime overhead together. Confirm framework and driver support before commissioning the workstation; memory capacity alone does not establish model speed or compatibility.
Local AI workloads
The 96GB ECC framebuffer makes the RTX PRO 6000 particularly relevant to local AI workloads that exceed the capacity of 24GB, 32GB or 48GB GPUs.
Potential use cases include large-language-model inference, fine-tuning, multimodal AI, embedding pipelines, generative image/video workflows, agentic AI development and local research environments.
Actual model compatibility depends on parameter count, quantization, precision, framework, KV-cache requirements, context length and software optimization.
Because Blackwell support evolved rapidly after launch, professional buyers should confirm that the intended CUDA, PyTorch, TensorRT, llama.cpp or other framework version fully supports the GPU before commissioning a production workstation.
GPU partitioning
NVIDIA lists Multi-Instance GPU support for the RTX PRO 6000 Blackwell Workstation Edition.
Where supported by the operating system and software stack, one physical 96GB GPU can be divided into:
MIG can be valuable for universities, AI development teams, shared workstations and other environments where several isolated workloads need access to one high-memory accelerator.
Compatibility
The card uses PCI Express 5.0 x16. Confirm motherboard slot spacing, lane allocation and physical clearance.
The workstation power supply must support the 600W GPU plus CPU, memory, storage, fans and other system components.
NVIDIA specifies one PCIe CEM5 16-pin connector. Use the correct cable and power-supply configuration.
The card is approximately 12 inches long and 5.4 inches high, so ordinary dual-slot compatibility alone is not enough.
Chassis airflow must allow both flow-through cooling paths to operate without severe obstruction.
Validate motherboard firmware, operating-system support, NVIDIA professional drivers and application certification.
Edition comparison
Compare the cooling designs visually as well as checking the written edition name. The following are alternative-edition reference images, not items included with this Workstation Edition offer.


| Specification | Workstation Edition | Max-Q Workstation | Server Edition |
|---|---|---|---|
| GPU memory | 96GB GDDR7 ECC | 96GB GDDR7 ECC | 96GB GDDR7 ECC |
| Maximum power | 600W | 300W | 400–600W configurable |
| Form factor | 5.4 × 12 in, dual slot | 4.4 × 10.5 in, dual slot | 4.4 × 10.5 in, dual slot |
| Cooling | Double flow through | Active | Passive |
| Deployment | Maximum single-GPU desktop performance | Dense / multi-GPU workstations | Validated rack servers |
| Best fit | High-performance workstation | Power-efficient workstation density | Enterprise server infrastructure |
Lower-power alternative

| Specification | RTX PRO 6000 | RTX PRO 5000 48GB |
|---|---|---|
| Memory | 96GB GDDR7 ECC | 48GB GDDR7 ECC |
| Memory bandwidth | 1,792 GB/s | 1,344 GB/s |
| AI performance | 4,000 AI TOPS | Up to 2,064 AI TOPS |
| Maximum power | 600W | 300W |
| Length | 12.0 in | 10.5 in |
| Best reason to choose | 96GB capacity and flagship throughput | Lower power and cost when 48GB is enough |
If your applications reliably fit inside 48GB, the RTX PRO 5000 can be the more rational workstation purchase because it substantially reduces power and chassis demands.
Professional workloads
Large local inference, fine-tuning and agentic AI workflows that benefit from 96GB VRAM.
Image, video, multimodal and other Blackwell-accelerated generative workflows.
GPU rendering in Blender, V-Ray and other CUDA/OptiX-enabled professional applications.
Large assemblies, engineering visualization, simulation and digital-twin workflows.
High-resolution real-time rendering, Unreal Engine and complex production environments.
Four NVENC and four NVDEC engines accelerate demanding ingest, encode, decode and finishing pipelines.
Buyer limitations
The live WooCommerce price above is the current store reference. Request a written quotation for the exact edition, quantity, unit condition, supplied accessories and destination. Compare offers on an equivalent basis: OEM identity, support arrangements, tax treatment and freight scope can materially change the total cost.
For a volume purchase, provide your workstation configuration and required delivery country. Confirm inventory and the commercial terms before approving the order. Historical market prices and another retailer’s advertised offer do not establish the scope or availability of this listing.
High-value GPU verification
For unusually discounted units, also see the
RTX PRO 6000 used buying and verification guide
.
Package contents
NVIDIA does not publish one universal accessory bundle for every OEM, regional or distribution channel.
Treat only the graphics card and accessories explicitly listed in the quotation or packing scope as included.
Never assume accessories shown in retailer photographs are part of every commercial package.
Warranty & support
Warranty eligibility depends on the manufacturer or partner, sales channel, product condition and purchase terms. Do not assume that a general manufacturer policy establishes coverage for a particular reseller, OEM or previously owned card. Request the exact supplied warranty, responsible service provider and claim procedure in the written quotation.
International purchasing
Confirm current inventory, tax treatment, warranty channel and delivery method.
Confirm VAT, customs, importer responsibility, warranty handling and destination delivery terms.
Confirm import VAT, duties, commercial invoice requirements and warranty return procedures.
Request a destination-specific freight and insurance quotation for high-value GPU shipments.
Confirm customs, insurance, declared value and delivery responsibility before payment.
Advanced GPUs can be subject to export or re-export controls depending on destination, end user and intended use.
Technical sources
Official NVIDIA specifications, features and workload information.
Manufacturer technical specification reference.
Procurement checklist
Buyer questions
Commercial and technical answers for buyers evaluating the NVIDIA RTX PRO 6000 Blackwell Workstation Edition.
Use the live store price above as the current reference. Request a written quotation for the required quantity, exact unit condition, included accessories, warranty and destination-specific delivery terms.
AI Robot Supplier lists the RTX PRO 6000 Blackwell Workstation Edition for commercial procurement. High-value or multi-unit buyers can request a written quotation covering exact configuration, warranty and destination shipping.
The Blackwell Workstation Edition provides 96GB GDDR7 memory with ECC.
NVIDIA lists approximately 1,792 GB/s.
NVIDIA specifies 24,064 CUDA cores.
NVIDIA rates the Workstation Edition at up to approximately 4,000 AI TOPS under its defined Blackwell AI-performance methodology.
Maximum board power is 600W. The complete workstation power supply must also account for the CPU and all other system components.
NVIDIA specifies one PCIe CEM5 16-pin connector.
The Workstation Edition is approximately 5.4 inches high and 12 inches long in an extended-height dual-slot design.
Yes. It uses PCI Express 5.0 x16.
Choose a supported professional driver branch appropriate to the operating system and application. Check the release documentation for the API versions and framework support required by your workstation.
Its 96GB memory capacity makes it highly relevant to large local AI workloads. Model fit depends on precision, quantization, context length, KV cache and software framework.
Yes. NVIDIA lists configurations of up to four 24GB instances, two 48GB instances or one full 96GB instance where supported.
Both provide 96GB GDDR7 ECC. The standard Workstation Edition is a 600W model designed for maximum single-GPU desktop performance, while Max-Q is a 300W model designed for greater power efficiency and denser workstation deployments.
No. Server Edition uses passive cooling and is designed for validated rackmount server airflow. The desktop Workstation Edition uses double-flow-through active cooling.
Choose RTX PRO 6000 Blackwell Workstation Edition when your workflow genuinely needs 96GB memory or flagship Blackwell throughput. RTX PRO 5000 can be more cost- and power-efficient when 48GB is sufficient.
It can run graphics workloads and games, but it is primarily a professional workstation GPU. Buyers focused only on gaming generally do not need its 96GB ECC memory, professional drivers or workstation features.
Confirm the exact warranty supplied with this unit in the written quotation. Manufacturer, partner, OEM and previously owned hardware can have different eligibility and service arrangements.
International purchasing depends on destination, end user, stock, carrier requirements and any applicable export or re-export rules. Confirm destination eligibility before payment.
96GB professional Blackwell GPU
Request a destination-specific quotation with your quantity, country, workstation configuration and purchasing deadline. Exact edition, condition, stock, warranty, price and delivery terms can be confirmed before order approval.
Add AI Robot Supplier as a preferred source on Google.
Tell us what you need — we respond within one business day.