A high-performance workstation used to mean buying powerful hardware for every designer, engineer or editor and then finding a secure way to reach it remotely. GPU-powered Amazon WorkSpaces offers another option: the applications run on a managed desktop in AWS while the user receives the interactive display on their local device.

In March 2026, AWS launched new Graphics G6, Gr6 and G6f bundles for WorkSpaces Personal and WorkSpaces Core. They are built on Amazon EC2 G6-family infrastructure and powered by NVIDIA L4 Tensor Core GPUs. AWS reports up to twice the graphics and ML inference performance of the previous G4dn generation on representative workloads. The result is a broader set of cloud workstation options, including fractional GPU bundles for users who do not need an entire GPU. See the AWS launch announcement for the published comparisons.

If you are still comparing desktop platforms, our AWS WorkSpaces, Azure Virtual Desktop and Citrix comparison explains the wider architectural differences.

What AWS Added to WorkSpaces

The G6 family provides three ways to match GPU and memory capacity to the work instead of assigning the same expensive workstation to everyone.

Bundle family Resource profile A sensible starting point
Graphics G6f Fractional NVIDIA L4 GPU, from 1/8 to 1/2 GPU Lighter CAD, design review, basic content creation and occasional GPU acceleration
Graphics G6 One complete NVIDIA L4 GPU with 24 GB of GPU memory and a balanced 1:4 vCPU-to-RAM ratio Demanding 3D, rendering, video and engineering applications
Graphics Gr6 / Gr6f Higher 1:8 vCPU-to-RAM ratio, with full or fractional GPU options Large CAD assemblies, simulations and data-heavy visualizations that need more system memory

AWS lists 12 configurations across these families. Full-GPU G6 and Gr6 configurations provide 24 GB of GPU memory, while high-memory variants reach up to 256 GB of system RAM. The WorkSpaces bundle documentation identifies CAD/CAM, graphic design, video transcoding, 3D rendering, data processing, and ML training and inference among the supported workload categories.

Those specifications are useful for shortlisting. They do not replace a test with the actual application, plugins, project files and display setup your team uses.

Where GPU-Powered WorkSpaces Fit

Advanced 3D modelling and rendering

Interactive 3D work depends on more than raw CPU capacity. Complex scenes, detailed textures, ray tracing and high-resolution viewports place sustained demand on GPU memory and system RAM. G6 provides a modern full-GPU option, while Gr6 is designed for projects where the model or dataset also consumes substantial system memory.

The desktop and project data can remain in AWS rather than travelling between a local workstation and cloud storage for every operation. This can simplify access for distributed teams, but storage placement and throughput must be designed alongside the desktop.

Video editing and production

NVIDIA L4 GPUs include hardware acceleration for graphics, AI and video workloads. AWS positions the G6 bundles for video editing, transcoding, animation and visual effects, including high-resolution production workflows.

A cloud environment can place the editing desktop close to shared media and render outputs, while giving contractors a controlled workstation without copying the project to an unmanaged computer. Bundle choice depends on codec, resolution, effects, storage and application. Light editing may fit G6f; sustained high-resolution production is a candidate for a full GPU and a tested storage path.

Data visualization and machine learning

G6 combines graphics acceleration with fourth-generation Tensor Cores. That makes it relevant to interactive data visualization, image and video analysis, AI-assisted creative tools, and compatible ML training or inference tasks.

The distinction between a workstation and a compute platform matters. An interactive desktop for visualization, development and moderate GPU jobs may fit WorkSpaces G6; large multi-GPU training may be better served by purpose-built EC2 or managed ML services.

CAD and engineering

AWS specifically calls out architecture, engineering and construction applications such as Autodesk Revit, AutoCAD and Bentley MicroStation in its launch material. Large assemblies, infrastructure models and simulations are also the main reason AWS introduced the higher-memory Gr6 variants.

For an engineering team, the benefit is access to a consistent workstation from the office, home or a project location. CAD deployments still require checks for licensing, versions, plugins, 3D mice, monitors, printing and document-management integrations.

Choosing Between G6, Gr6 and G6f

Choose from measured workload behaviour, not from the user’s job title. Two engineers can have very different requirements if one reviews drawings while the other opens large assemblies and renders them throughout the day.

  • Start with G6f when GPU use is light or intermittent and a fractional allocation can meet the acceptance criteria.
  • Start with G6 when the workflow needs a complete GPU for sustained viewport, rendering, video or inference work.
  • Start with Gr6 when system memory, rather than GPU memory alone, limits large models, simulations or visualizations.
  • Consider Gr6f when the workload needs high system memory but only moderate GPU capacity.

A controlled pilot helps avoid both wasted capacity and poor user experience. It should use representative project files and record model load, viewport response, playback, export or render time, memory consumption and user feedback.

AWS published selected price and performance comparisons, but they are tied to stated regions, billing choices and representative tests. A production estimate should use current regional pricing and expected usage, storage, data transfer and licensing.

What Computer On Site Provides as an AWS Partner

We turn the selected GPU bundle into a working, supportable cloud workstation service. The AWS console can create a WorkSpace; the project work is making applications, identity, data, security and user experience work together.

Our delivery scope can include:

  • application, plugin, dataset, peripheral and user discovery;
  • AWS Region, availability, quota and G6-family sizing checks;
  • directory, identity, access, network and endpoint design;
  • desktop images, application packaging and update planning;
  • storage and security controls for models, media and datasets;
  • software-licensing validation with the relevant vendor;
  • pilot delivery, acceptance criteria, rollout and operational documentation.

WorkSpaces supports Active Directory and non-AD designs, with documented paths for Microsoft Entra ID through AWS IAM Identity Center. The correct model depends on the identities, devices and controls already in use. See the current Amazon WorkSpaces capabilities.

What a GPU WorkSpaces Pilot Should Prove

A useful pilot answers business questions, not just whether the desktop starts.

  1. Compatibility: install the exact application versions, plugins, render engines and drivers.
  2. Real-project performance: complete a representative model, edit, render, export or visualization.
  3. User experience: measure latency and display quality from actual user locations.
  4. Data path: validate file, cache, media and output performance with the proposed storage.
  5. Peripherals: test monitors, 3D mice, tablets, audio, printing and specialist hardware.
  6. Security and licensing: confirm identity, data controls, patching, administration and vendor terms.
  7. Cost: model monthly and usage-based choices against real working patterns.

The rollout should use agreed pass/fail criteria. Finding that a specialist task belongs on a physical workstation is a useful result, not a failed pilot.

Frequently Asked Questions

Can AWS WorkSpaces G6 run AutoCAD, Revit, Premiere Pro or DaVinci Resolve?

AWS identifies these workload categories, but that does not guarantee every version, plugin, codec or licence. We validate the exact stack during the pilot.

Can employees connect from an ordinary laptop?

In many cases, yes. Applications and GPU work run in AWS, but the endpoint, client, network, monitors and peripherals must still meet project requirements.

Is a fractional G6f bundle suitable for professional work?

It can be. G6f suits lighter CAD, review and intermittent GPU demand. Full G6 or high-memory Gr6 is a better pilot candidate for sustained rendering or large models.

Can G6 replace every physical workstation?

No. Specialist hardware, vendor restrictions, local data or very low latency can favour a physical workstation. The goal is to identify where cloud desktops fit.

Is AWS WorkSpaces cheaper than buying workstations?

It depends on utilisation, bundle, storage, licensing, data transfer and hardware refresh. A fair comparison needs the lifecycle cost on both sides.

Start With One Real Workload

AWS WorkSpaces G6 gives design, engineering, media and data teams access to current GPU-powered desktops without tying the working environment to one physical computer. The successful design is the one that matches the right bundle, storage, network and controls to a real project.

Computer On Site can assess your workload, build the AWS environment and run a pilot with your applications and users. Talk to us about an AWS GPU WorkSpaces pilot, or review our wider cloud workspace services.