NVIDIA RTX PRO 6000 Blackwell Server Edition (96GB)
$15,339.00
Skip to product information
NVIDIA RTX PRO 6000 Blackwell Server Edition (96GB)
$15,339.00
The NVIDIA RTX PRO 6000 Blackwell Server Edition is a 96GB GDDR7 ECC graphics card for server deployments running generative AI, scientific computing, rendering and video workloads. It combines large ECC-protected graphics memory with dedicated AI and ray-tracing hardware. It suits teams that already have a compatible server chassis and need a lot of GPU memory in a single card.
Who it's for
- Running large models, datasets and complex scenes that need 96GB of GPU memory
- FP4 inference on fifth-generation Tensor cores
- Sharing one GPU between up to four isolated workloads with MIG
- Demanding video pipelines that use four encode and four decode engines
Key features
- 96GB GDDR7 ECC: Capacity for large models, datasets and complex scenes, with 1,597 GB/s memory bandwidth.
- Blackwell acceleration: Fifth-generation Tensor cores support FP4 inference, while fourth-generation RT cores accelerate ray tracing.
- Shared GPU resources: MIG partitions one GPU into up to four isolated instances with dedicated resources.
- Professional media processing: Four encode and four decode engines support demanding video pipelines.
- Built-in security: Confidential computing and secure boot with root of trust.
- Passive cooling: The heatsink relies on airflow provided by the server.
Specifications
| Spec | Detail |
|---|---|
| Architecture | NVIDIA Blackwell |
| GPU memory | 96 GB GDDR7 with ECC |
| Memory interface | 512-bit |
| Memory bandwidth | 1,597 GB/s |
| CUDA cores | 24,064 |
| Tensor cores | 752, fifth generation |
| RT cores | 188, fourth generation |
| FP32 performance | 120 TFLOPS |
| Peak FP4 Tensor performance | 4 PFLOPS |
| Peak FP8 Tensor performance | 2 PFLOPS |
| Peak FP16 / BF16 Tensor performance | 1 PFLOP |
| Peak TF32 Tensor performance | 234 TFLOPS |
| Peak RT core performance | 355 TFLOPS |
| Host interface | PCI Express 5.0 x16 |
| Multi-Instance GPU | Up to 4 isolated instances; up to 24 GB per instance |
| Video engines | 4 ninth-generation NVENC; 4 sixth-generation NVDEC |
| Cooling | Passive heatsink; requires server-provided airflow |
| Maximum board power | Up to 600 W, configurable |
| Power connector | 1 × PCIe CEM5 16-pin |
| Form factor | Full-height, full-length, dual-slot |
| Card dimensions | 4.4 inches high × 10.5 inches long |
| Display connectors | 4 × DisplayPort 2.1 |
| NVLink | Not supported |
| Security | Confidential computing; secure boot with root of trust |
Good to know
- Performance figures are manufacturer-rated peaks; application throughput depends on precision, software and system configuration.
- This listing is for the passive, air-cooled Server Edition GPU card. Its heatsink depends on forced airflow from a compatible server chassis.
- Confirm chassis clearance, PCIe slot support, power cabling and cooling capacity before installation.
- It is not a complete server.
- Optional enterprise software licensing is separate unless expressly included in your order.
- Manufacturer documentation referenced by the original listing: NVIDIA product specifications, PNY specifications, and a product brief with integration requirements.
Choose this card if your server can feed a passive, dual-slot card with forced airflow and you want 96GB of ECC memory per GPU. If you need a complete machine rather than a component, look at the full systems in the ComputaHardware range instead.