Lenovo 4X67B09287 Thinksystem NVIDIA RTX PRO 6000 Blackwell Server Edition 96GB PCIE GEN5 Passive

Lenovo 4X67B09287 NVIDIA RTX PRO 6000 Blackwell Server GPU 96GB GDDR7 PCIe Gen 5

Overview

The Lenovo 4X67B09287 is a data-center-class GPU built around NVIDIA's RTX PRO 6000 Blackwell architecture, provisioned specifically for server and workstation rack environments where sustained AI inference, simulation, and compute-intensive workloads demand both massive memory capacity and generation-5 interconnect bandwidth. With 96 GB of GDDR7 on a 512-bit bus, this card handles model sizes and datasets that saturate 24 GB or 48 GB alternatives before the job even starts. If you're sizing a server for large-scale AI inference, deep learning training, or GPU-accelerated analytics pipelines, the 4X67B09287 is engineered for exactly that environment.

The passive cooling design signals this card's purpose clearly: it is built to live inside a rack server chassis with forced-air cooling from system fans, not an open-air workstation. Plan accordingly — passive GPU plus inadequate chassis airflow is a thermal failure waiting to happen.

Key Features

• 96 GB GDDR7 Memory on a 512-Bit Bus: This is the headline specification that separates the RTX PRO 6000 from consumer and mid-tier professional GPUs. 96 GB means you can run large language models, multi-frame rendering queues, or multi-tenant inference workloads entirely in GPU memory without offloading to system RAM. The 512-bit bus width paired with GDDR7 delivers up to 1,597 GB/s of memory bandwidth — critical when your workload is memory-bound, not compute-bound.
• 120 TFLOPS Peak FP32 Single-Precision Matrix Performance: 120 TFLOPS of single-precision throughput gives this card meaningful headroom for real-time AI inference and mixed-precision training. For comparison, this positions the RTX PRO 6000 Blackwell at the high end of professional server GPU performance available via standard PCIe form factor — relevant when you're evaluating whether this card can sustain your batch inference SLA without additional nodes.
• 24,064 CUDA Cores: The sheer core count supports highly parallel workloads — scientific simulations, rendering farms, large-scale vision analytics — where thread-level parallelism directly maps to throughput. More CUDA cores mean more simultaneous operations per clock cycle.
• PCIe Gen 5 x16 Interface: PCIe 5.0 x16 doubles the theoretical host-to-GPU bandwidth over Gen 4. This matters when your pipeline moves large tensors or inference results back to the CPU frequently — bottlenecks at the bus become bottlenecks in your pipeline. Confirm your server platform supports PCIe 5.0 x16 slots before purchasing; installing into a Gen 4 slot will work but will cap host transfer rates.
• Passive Cooling (600W TDP): At 600W maximum power draw, this is a high-TDP card. Passive cooling means zero fan noise and no fan-failure risk from the GPU itself, but it absolutely requires a server chassis designed for high-airflow forced cooling. Verify chassis thermal specifications against 600W sustained before deploying. Undersized airflow will trigger thermal throttling or shutdown.
• 4x DisplayPort 2.1b Outputs: Four DP 2.1b ports support high-resolution multi-display workstation use cases — visualization suites, control room operators, or hybrid server/workstation deployments. DisplayPort 2.1b handles 8K and high-refresh-rate 4K configurations where needed.
• DirectX 12.0, OpenGL 4.6, Shader Model 6.6: Full API coverage means this card supports the full range of GPU-accelerated visualization, simulation, and compute frameworks without driver-level workarounds. OpenGL 4.6 is the current stable release — relevant for scientific and engineering applications that haven't migrated to Vulkan.
• Multi-Certification (FCC, CE, UL/cUL, BSMI, KCC, VCCI, RCM, ICES): Certified across North American, European, Australian, Japanese, Korean, and Taiwanese markets. If you're deploying into regulated environments or multi-region data centers, certification coverage is already in place — no additional compliance testing needed for these markets.

Integration and Compatibility

The 4X67B09287 connects via PCIe x16 Gen 5 and supports Ethernet alongside the PCIe interface for out-of-band management configurations. With 96 GB of on-board GDDR7, it is suited for integration into AI inference servers, HPC nodes, and GPU-accelerated analytics platforms running CUDA-based frameworks such as TensorRT, PyTorch, and RAPIDS. CUDA support is confirmed. DirectX 12.0 and OpenGL 4.6 support ensures compatibility with visualization and simulation software stacks. The passive thermal design requires a server chassis with validated forced-air cooling capable of handling the 600W sustained TDP — check your chassis vendor's GPU compatibility list before deploying this card. The card ships from Lenovo's Lenovo ThinkSystem server GPU line, targeting rack-mount server platforms that support full-length, full-height PCIe add-in cards. If you're building or expanding a server and storage infrastructure, verify slot mechanical clearance and PCIe slot power delivery (auxiliary power connectors required for 600W) before finalizing the build. For teams evaluating GPU options for AI-accelerated video analytics or NVR platforms, the 96 GB GDDR7 capacity supports running multiple concurrent deep learning inference models that would exhaust lower-capacity cards. Buyers planning GPU-accelerated workloads should also review GPU server selection considerations to match chassis airflow, power delivery, and PCIe slot configuration to this card's requirements.

Frequently Asked Questions

Q: What is the maximum power consumption of the Lenovo 4X67B09287?

A: The RTX PRO 6000 Blackwell Server Edition draws up to 600W at peak load. Ensure your server chassis and power supply can sustain this draw continuously, and that chassis airflow is rated for passive GPU cooling at this TDP.

Q: Does the 4X67B09287 require a PCIe Gen 5 slot to function?

A: The card is designed for PCIe Gen 5 x16, but it is backward compatible with PCIe Gen 4 slots. Installing in a Gen 4 slot will reduce host-to-GPU bandwidth but will not prevent the card from operating. For full bandwidth (critical for data-intensive workloads), a PCIe Gen 5 platform is required.

Q: Is the 4X67B09287 compatible with CUDA-based AI frameworks?

A: Yes. The card is CUDA-enabled with 24,064 CUDA cores, supporting CUDA-based frameworks including TensorRT, PyTorch, and RAPIDS on compatible server platforms.

Q: What certifications does the 4X67B09287 carry?

A: The card holds RCM, BSMI, CE, FCC, ICES, KCC, cUL/UL, and VCCI certifications, covering North American, European, Australian, Korean, Taiwanese, and Japanese regulatory markets.

Q: Can the 4X67B09287 be used in a workstation with standard air cooling?

A: This is a passive-cooled server GPU. It does not include a fan and relies entirely on chassis-forced airflow. It is not suitable for open-air desktop workstations or chassis without validated forced-air cooling for high-TDP passive GPUs.

Q: How many displays does the 4X67B09287 support?

A: The card has four DisplayPort 2.1b outputs, supporting up to four simultaneous high-resolution displays — capable of driving 8K or high-refresh-rate 4K monitors for visualization and control room use cases.