Exxact Validates 4x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition GPUs

3/12/2026: Update with Stress Test Graph
8/18/2025: Initial Blog Upload

A Problem Worth Solving

The NVIDIA® RTX PRO™ 6000 Blackwell Max-Q Workstation Edition GPU with 96GB of VRAM has made waves across all industries that value GPU acceleration and high VRAM: AI, Engineering, 3D Content Creation, Life Science, and more. With multi-GPU workstations, the deployment flexibility and performance capabilities are second to none.

With multi-GPUs workstations, we have a long-standing validation process that every Exxact solution has to clear to ensure proper cooling of the high heat components like multiple GPUs, memory modules, processors, and other components.

Validating four NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition GPUs in a workstation has been challenging due to a higher heat output than the previous generation. Three NVIDIA RTX PRO 6000 Max-Q were considered the thermal limit for air-cooled configurations, with other integrators resorting to liquid cooling. As of today, March 11, 2026, we still have yet to see another solution integrator offer an air-cooled 4x Max-Q configuration.

However, at Exxact, we thought otherwise. NVIDIA had designed the NVIDIA RTX PRO 6000 Blackwell Max-Q to scale to 4x GPUs in a workstation. We saw the opportunity as a challenge to offer an air-cooled 4x NVIDIA RTX PRO 6000 Blackwell Max-Q workstation. We are the first to proudly say, “We did it.”

The First 4x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation by Exxact

Exxact Corporation is the first to deliver a 4x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition in a workstation without liquid cooling the GPUs. With careful chassis selection, airflow design, and thermal monitoring adjustments, we observed GPU stability peaking 85°C over a 3 hour OCCT stress test.

Introducing the Exxact Valence 4x Max-Q Workstation. This testing was performed on an AMD Threadripper PRO 7000WX platform with all 8 channels of memory populated. Here are the specifications of the system tested in a fixed 24°C/75°F environment with the side panels on:

In a stock configuration without the Exxact cooling solution, we also observed four NVIDIA RTX PRO 6000 Blackwell Max-Qs thermal throttling, undervolting, and leaving valuable performance (and money) on the table. However, with our custom cooling solution, each of our GPUs pulled the maximum 300W TDP with a temperature below 90°C on all four cards. This goes without saying; deploying your system in a hotter environment or restricting front, side, or rear panel airflow can adversely affect performance.

We are extremely proud of our engineering team for their ability to deliver the first validated 4x-GPU workstation featuring the extremely popular NVIDIA RTX PRO 6000 Blackwell Max-Q GPU. We ran these same tests on with great success:

• AMD Threadripper PRO 9000WX
• Intel Xeon W-3500
• Dual Intel Xeon Scalable 5th Gen

We are currently running tests on the new Intel Xeon 600 workstation processor platform.

Why Air-Cooled versus Liquid-Cooled?

Air cooling is still the most practical choice for many workstation users because:

• It is simpler to deploy, with fewer components and less setup.
• It is easier to service, which matters for IT teams supporting multi-GPU fleets.
• It is more predictable over time, with no pumps to fail, no coolant to maintain, and no risk of leaks.
• It reduces operational variables, which helps teams deploy, and transport the system faster.

With today’s high-power GPUs, the tradeoff is that air-cooled systems must be engineered correctly. A quad-GPU build needs a chassis that supports high static pressure airflow, a fan and ducting strategy that feeds each card with cool intake air, and a thermal budget that keeps hotspots under control during sustained load.

Liquid cooling can be a good fit when you are constrained by acoustic targets and need lower fan speeds, when you need maximum thermal headroom in warmer ambient environments, or when you are comfortable with additional maintenance and component complexity.

For our validation, the goal was clear: prove 4x RTX PRO 6000 Blackwell Max-Q stability in a real workstation, with the side panels on, using air cooling. By combining chassis selection, airflow optimization, and continuous thermal monitoring, we achieved stable performance at full 300W TDP per GPU without relying on liquid cooling.

Why the NVIDIA RTX 6000 PRO Blackwell Max-Q is Your Next GPU

The NVIDIA RTX PRO 6000 Blackwell Max-Q is designed with deployment flexibility and scalability in mind. It is a dual-slot workstation-class GPU design for multi-GPU deployments and can be fitted in a typical workstation and a multi-GPU server.

It features the largest VRAM capacity in a workstation-class GPU with 96GB of GDDR7 per card; when deployed in a 4x GPU workstation configuration, 384GB of GPU memory can power almost any workload. Coupled with the new GDDR7, the memory bandwidth was increased, resulting in significantly better GPU-to-GPU and GPU-to-component communication.

At Exxact, our customers are extremely excited to get their hands on this GPU. Here are some popular use cases we see that can leverage our Tensor Workstation featuring four NVIDIA RTX PRO 6000 Blackwell Max-Q:

Multi-GPU for AI Development and Local Inference

A 4x Max-Q GPU setup is ideal when you need high VRAM per desk for development, testing, and secure on-prem deployment. Example 4x Max-Q GPUs use cases:

• Local LLM serving (RAG): 96–192GB aggregate VRAM serves 70B–180B parameter models (e.g., Llama 3.1 70B, Mixtral 8x22B) fully on-prem without quantization compromises.
• Fine-tuning: Enables full-parameter fine-tuning of 7B–13B models or LoRA runs on 70B models, cutting per-epoch time by 3–4x versus single-GPU.
• Throughput-focused inference: Multi-GPU parallelism delivers 2–4x the token throughput of a single GPU for sustained multi-user workloads.
• Multi-modal workloads: Run vision-language models like LLaVA-34B or InternVL2-40B locally without sending data offsite.

Our Valence 4x Max-Q Workstation handles previously enterprise-only AI tasks. It can't train foundational LLMs (those require hundreds to thousands of GPUs), but it's a strong fit for serving 5–50 concurrent internal users or iterating on sub-100B parameter models without server infrastructure.

Multi-GPU for Engineering Simulation & MPD

4x Max-Q GPUs help when models are too large for a single GPU, or when teams want to run multiple design iterations in parallel from a desk-side workstation. Example 4x GPU workstation use cases:

• CFD and thermal: Handle mesh sizes of 50M–500M+ cells in solvers like Ansys Fluent — cases that would otherwise require cluster time.
• FEA for large assemblies: Faster nonlinear and contact-heavy runs on assemblies exceeding 1M+ DOF, where GPU memory is the primary bottleneck.
• Parameter sweeps and digital twins: Run 4 independent design variants simultaneously, compressing a day of sequential iteration into hours.
• Pre/post visualization: Interactively review 10–100GB datasets locally without streaming to a remote node.

High-performance GPUs increase productivity and iteration count, with typical simulation time reductions of 40–70% versus CPU-only workflows.

Multi-GPU for Engineering Simulation & MPD

4x Max-Q GPUs help when models are too large for a single GPU, or when teams want to run multiple design iterations in parallel from a desk-side workstation. Example 4x GPU workstation use cases:

• CFD and thermal: Handle mesh sizes of 50M–500M+ cells in solvers like Ansys Fluent — cases that would otherwise require cluster time.
• FEA for large assemblies: Faster nonlinear and contact-heavy runs on assemblies exceeding 1M+ DOF, where GPU memory is the primary bottleneck.
• Parameter sweeps and digital twins: Run 4 independent design variants simultaneously, compressing a day of sequential iteration into hours.
• Pre/post visualization: Interactively review 10–100GB datasets locally without streaming to a remote node.

High-performance GPUs increase productivity and iteration count, with typical simulation time reductions of 40–70% versus CPU-only workflows.

Multi-GPU for 3D Design, Rendering, and Visualization

4x Max-Q GPUs boost throughput for creators who need faster iteration and bigger scenes. Example 4x GPU workstation use cases:

• GPU rendering: Near-linear scaling in Octane, Redshift, and V-Ray GPU — 4 GPUs deliver roughly 3–3.8x the throughput of one card on complex production scenes.
• Virtual production and real-time 3D: Maintain 120+ FPS headroom in Unreal Engine 5 with Lumen and Nanite enabled.
• Large scene assembly: Handle geometry exceeding 500M–1B polygons and texture sets in the 50–200GB range without constant cache eviction.
• AI-assisted workflows: Sub-second latency on upscaling, denoising, and generative passes that would stall on a single GPU.

For studios and creators, multi-GPU shortens render queues and speeds up heavy GPU-accelerated workflows while keeping interactive work local for faster reviews and iteration.

Conclusion

At Exxact, we validate every configuration to ensure optimal performance in the most strenuous use case. Our goal is to deliver the most capable system that works right out of the box. Each system is designed to support a wide range of computing environments and workloads without compromise.

The AMD Threadripper PRO 9995WX paired with 4x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition is the most performant workstation configuration for handling all types of workloads in AI, Engineering, Life Science, and 3D content. Want to configure this Exxact platform or any other 4x Max-Q workstations? Contact us today!