I'm temporarily handling the company's IT and systems work while we look for vendors, and I need to gather quotes for a new deployment at a facility in Germany. We have three or four demanding engineering applications that require strong single-core CPU performance and dedicated graphics. One application recommends an NVIDIA A6000, and the stated memory requirements range from 64 GB to 128 GB per application. There are currently only three users—roughly one per application—but that number is expected to grow quickly. A centralized VDI setup on one or two powerful servers seems attractive from a management, security, and compliance perspective. However, I'm unsure about the network, GPU, storage, licensing, and server requirements, as well as whether the user experience would be good enough for engineering workloads. I'm also concerned that a VDI environment could introduce more complexity than makes sense for a small company. Would it be more practical and cost-effective to provide high-end workstations for each engineer, possibly with remote access and centralized management, or should we plan for VDI from the start?
5 Answers
A possible middle ground is to keep interactive engineering work on dedicated GPU workstations while centralizing file storage, license services, backups, patching, and identity management. If simulations are separate from the interactive design work, those workloads can be sent to a Linux-based HPC or cloud environment. This gives the engineers responsive local graphics without forcing every workload into a complex VDI platform.
Engineering users tend to notice even small amounts of display latency, so they may have a poor experience with remote graphics even when ordinary office applications perform well. In many manufacturing environments, VDI is used for most employees while engineers receive secured local workstations because they provide better CAD performance and are easier to justify financially. Put the systems in a controlled configuration, use endpoint management and security policies, and keep a spare or reference workstation available rather than immediately investing in a GPU server farm.
With only three users, I’d start with dedicated workstations rather than building a shared GPU-accelerated VDI platform. Configure each system around the application vendor’s recommendations, using a high-clock CPU, the required professional GPU, and enough memory for the largest real-world projects. You can lock them down and manage them centrally with tools such as Intune and Autopilot, while keeping the engineering software close to the hardware. Also verify whether each application is licensed per machine, per user, or through a floating-license server—licensing can make pooled desktops impractical very quickly.
Cloud desktops such as Azure Virtual Desktop or similar hosted GPU workstations could be useful for a limited proof of concept, particularly if the user count will grow. However, for three users, the recurring compute, GPU, storage, bandwidth, and software licensing costs may exceed the cost of a few capable workstations. A cloud pilot can still provide useful performance and cost data before purchasing hardware, especially if future users will be distributed across locations.
GPU-backed VDI can work, but it is not a simple way to save money. You need compatible server GPUs or vGPU licensing, substantial RAM and CPU capacity, fast storage, low-latency networking, remote-display technology, and a plan for GPU and host failures. The advertised 64–128 GB requirements also do not automatically combine efficiently on a shared host, especially when several users are opening large models at once. I’d run a short pilot with one dedicated workstation or hosted desktop, using the heaviest real project and the actual network path before committing to a larger architecture.
Also test licensing for situations where managers or executives need to view CAD files. Some products count those viewers or machines differently, and that can complicate an otherwise workable remote-desktop design.