NVIDIA Workstation GPU Buying Guide

NVIDIA RTX PRO is a range of graphics cards designed for professional applications such as CAD & CAE, AEC & BIM, rendering, AI, video editing and VFX. There are multiple NVIDIA RTX PRO graphics cards, each optimised for different applications, allowing you pick the perfect card for your project and budget.

NVIDIA RTX PRO 6000 Blackwell is a high-end graphics card designed for professional applications such as CAD & CAE, AEC & BIM, rendering, AI, video editing and VFX. The RTX PRO 6000 Blackwell is the highest performing graphics card based on the Blackwell architecture which supports improved ray tracing and AI performance. The RTX PRO 6000 Blackwell features 24,064 CUDA cores, 752 5th gen Tensor cores and 188 4th gen RT cores, combined with 96GB of ultra-reliable ECC memory.

The graphics card is the most important component in a workstation PC as its responsible for rendering applications, performing simulations and running AI models. More powerful, and more expensive, graphics cards enable you to work at higher resolutions as well as producing higher frame rates, improving user experience.

A graphics card comprises two main components, the GPU (Graphics Processing Unit), effectively the brain of the card plus VRAM (Video Random Access Memory) which the GPU uses to store data such as models, textures as well as buffering rendered frames.

A CUDA (Compute Unified Device Architecture) core is the primary processor inside NVIDIA GeForce GPUs responsible for rendering graphics and other tasks. Each GPUs has thousands of CUDA cores working together in parallel; the more cores, the higher the performance. Some tasks are accelerated by specialist RT or Tensor cores, see the separate entry in this FAQ for further detail on these types of core.

You can see how advanced a CUDA core is by the name of the architecture, to name some recent examples Blackwell (2025), Ada Lovelace (2022) and Ampere (2020).

RT (Ray Tracing) cores work alongside CUDA cores inside NVIDIA GeForce GPUs to accelerate ray traced effects in applications. Unlike traditional rasterised graphics, ray tracing simulates how light beams and shadows actually work, providing far more rich and realistic graphics. The more RT cores, the faster the framerate when ray tracing is enabled.

You can see how advanced the RT cores are in a GPU by its generation, i.e. 5th gen is more advanced than 4th gen etc.

Tensor cores work alongside CUDA cores inside NVIDIA GeForce GPUs to accelerate AI workloads and applications. The more Tensor cores, the faster the AI performance will be.

You can see how advanced the Tensor cores are in a GPU by its generation, i.e. 5th gen is more advanced than 4th gen etc.

DLSS (Deep Learning Super Sampling) is an advanced AI-powered rendering technique unique to NVIDIA graphics cards that uses the Tensor cores to increase frame rates. DLSS enables you ramp up the resolution to enjoy higher quality graphics, without sacrificing performance.

NVLink is a proprietary bus that is used to link pairs of compatible NVIDIA cards together, combining their shared memory into a single pool for improved performance. Not all NVIDIA workstation GPUs support NVLink, so check compatibility carefully.

A FLOP (Floating-Point Operation per Second) is a measure of how quickly GPUs calculate the complex floating-point maths behind complex calculations such as simulations and training AI models. The more FLOPS, the faster the performance.

Depending on the software, floating point calculations can be calculated to as great or little precision as required. This is commonly shorted to FPx where x is a number typically in the range of 4 to 64. See the separate entry in this FAQ for further detail.

Like other metrics in computer science such as KB, MB, GB and TB when it comes to RAM or storage, the number of FLOPS is commonly shortened to GFLOPS (giga, 10 to the power of 9), TFLOPS (tera, 10 to the power of 12) and PFLOPS (peta, 10 to the power of 15).

These are a shortened representation of how precisely a floating-point calculation is performed. The most common levels are in order of increasing precision:

A TOP (Trillion Operations per Second) is a measure of how quickly a GPU can perform the integer maths that is commonly used when inferencing AI models. The more TOPS, the faster the performance.

NVIDIA RTX PRO graphics cards such as the RTX PRO 6000 can be used for gaming as they feature the same CUDA cores, RT cores, Tensor cores and DLSS support as NVIDIA GeForce RTX graphics cards. However, their primary use case is professional applications such as CAD & CAE, AEC & BIM, rendering, AI, video editing and VFX.

Workstation graphics cards are expensive because they are designed and built for commercial rather than consumer operation. Their higher pricing reflects greater memory capacities, more reliable hardware, certified software, and enterprise-level technical support.