ArcGIS Pro on AWS: GPU-Accelerated Deployment and Performance

Organizations across industries continue to modernize how they deliver ArcGIS Pro. As GIS workflows become more complex and organizations increasingly adopt cloud infrastructure, Amazon Web Services (AWS) provides several options for delivering GPU-accelerated ArcGIS Pro environments with flexibility, scalability, and centralized management.

ArcGIS Pro can be delivered through Amazon WorkSpaces, Amazon WorkSpaces applications, or dedicated GPU-enabled Amazon EC2 instances. Selecting the right approach depends on the GIS workloads being performed, how desktops and applications need to be delivered, where supporting data resides, and how much control the organization requires over the underlying infrastructure.

Choosing the Right AWS Platform

Selecting the right AWS platform for ArcGIS Pro starts with four considerations:

1. Workload requirements: 2D mapping, editing, 3D visualization, imagery, geoprocessing, and GPU-accelerated analysis can place very different demands on infrastructure.
2. Deployment model: Consider the delivery model required for the environment, such as a persistent GPU-enabled desktop, streamed access to ArcGIS Pro, or a dedicated GPU-enabled EC2 instance.
3. Scalability: Consider the number of users, expected concurrency, and how quickly resources may need to scale as demand changes.
4. Operational ownership: Determine how much control your organization requires over the operating system, GPU drivers, applications, and underlying infrastructure.

These considerations help determine whether Amazon WorkSpaces, Amazon WorkSpaces applications, or dedicated Amazon EC2 GPU instances provide the best fit for the environment.

Amazon WorkSpaces: Persistent GPU-Enabled Desktops

Amazon WorkSpaces provides persistent cloud desktops that can be configured with GPU resources for ArcGIS Pro. This provides organizations with a managed desktop environment while reducing the need to manage the underlying physical infrastructure.

AWS currently provides multiple generations of GPU-enabled WorkSpaces. Existing Graphics G4dn configurations remain viable for ArcGIS Pro, while newer Graphics G6f and G6 configurations provide NVIDIA L4 GPU resources. Graphics G7 provides a newer high-end option using NVIDIA RTX PRO Blackwell GPUs.

For new deployments, select the configuration based on the expected ArcGIS Pro workload rather than simply choosing the smallest available GPU-enabled bundle.

Recommended Starting Configurations

The fractional NVIDIA L4 configurations available with Graphics G6f can provide an effective balance between GPU resources and cost for many ArcGIS Pro users. For more demanding workflows, Graphics G6 provides access to a full NVIDIA L4 GPU, while Graphics G7 provides newer NVIDIA RTX PRO Blackwell GPU resources.

These configurations are starting points. Validate representative ArcGIS Pro projects and workflows before deploying broadly.

Existing Graphics G4dn WorkSpaces

Graphics.g4dn and GraphicsPro.g4dn use NVIDIA T4 GPUs and remain options for existing ArcGIS Pro deployments. Organizations already using these configurations do not necessarily need to migrate solely because newer GPU families are available.

For new deployments, evaluate the newer Graphics G6 configurations alongside the requirements of the ArcGIS Pro workload.

Amazon WorkSpaces Applications: Stream ArcGIS Pro on Demand

Amazon WorkSpaces applications, formerly Amazon AppStream 2.0, provides a managed application-streaming platform for delivering ArcGIS Pro without requiring users to maintain a persistent virtual desktop.

ArcGIS Pro can be delivered from GPU-enabled fleets that scale based on user demand. This can make WorkSpaces applications useful for training environments, temporary or project-based users, contractors, classrooms, and other scenarios where users need access to ArcGIS Pro without requiring a dedicated persistent desktop.

GPU Options for ArcGIS Pro

Several generations of NVIDIA GPU instances are available:

• Graphics G4dn – NVIDIA T4: An established option that remains suitable for many existing ArcGIS Pro environments.
• Graphics G5 – NVIDIA A10: A strong and widely deployed option for ArcGIS Pro visualization, editing, 3D, and analysis workloads.
• Graphics G6 – NVIDIA L4: A newer GPU platform offering fractional and full-GPU configurations, providing flexibility across a range of ArcGIS Pro workloads.
• Graphics G6e – NVIDIA L40S: A higher-end option for workloads requiring additional GPU performance, GPU memory, or CUDA compute capability.
• Graphics G7 – NVIDIA RTX PRO Blackwell: A newer high-performance GPU option for demanding graphics and GPU-accelerated workflows.

For new deployments, Graphics G6 provides a strong contemporary starting point, while G5 remains a capable established platform and G4dn remains relevant for existing environments. Higher-end G6e and G7 configurations can support workflows with greater GPU compute and memory requirements.

Instance selection should be based on the ArcGIS Pro workload, CPU and system memory requirements, GPU resources, expected concurrency, and the performance of the supporting data infrastructure.

As with other virtualized ArcGIS Pro environments, test representative projects and workflows before deploying broadly.

Amazon EC2 GPU Instances: Maximum Flexibility and Control

For organizations that require direct control over the virtual machine, operating system, GPU drivers, storage, and supporting infrastructure, Amazon EC2 GPU instances provide a flexible platform for ArcGIS Pro.

Several generations of NVIDIA GPU instances can support ArcGIS Pro. G6 provides the newer NVIDIA L4 GPU platform, while G5 remains a strong and widely deployed NVIDIA A10 option. G4dn, based on the NVIDIA T4, continues to support existing deployments and remains a practical option where it is already in use.

Single-Session Sizing (One User per Instance)

For new deployments, G6 provides a strong contemporary starting point. G5 remains a capable and widely deployed platform, and existing G4dn environments do not necessarily need to migrate solely because newer GPU generations are available.

These configurations are starting points. Select an instance based on the expected ArcGIS Pro workload and validate representative projects and workflows before production deployment.

Additional GPU Options

G6f – NVIDIA L4 – Fractional GPU instances can provide a cost-effective option for lighter ArcGIS Pro workloads. Because smaller G6f configurations provide limited CPU, system memory, and fractional GPU resources, carefully size and test these instances rather than assuming the smallest available configuration will meet ArcGIS Pro requirements.

G6e – NVIDIA L40S – A high-end option for demanding ArcGIS Pro workloads requiring greater GPU performance, GPU memory, or CUDA compute capability. The additional GPU resources can benefit complex 3D visualization, imagery, GPU-intensive analysis, and other demanding workflows.

Newer GPU instance families may also provide additional options as AWS expands its accelerated-compute portfolio. Evaluate GPU architecture, GPU memory, CPU, system memory, driver support, regional availability, and workload requirements when considering these configurations.

Deployment and Storage Guidance

ArcGIS Pro performance in AWS depends on the complete environment, including the EC2 or WorkSpaces configuration, storage architecture, network performance, data location, and the characteristics of the GIS workload.

When designing the environment:

1. Keep compute and data close: Locate ArcGIS Pro compute resources and frequently accessed GIS data as close together as practical. Minimizing latency between ArcGIS Pro and its data is particularly important for workflows that perform frequent file-based operations.
2. Select storage for the workload: AWS provides several storage options, including Amazon EBS and shared file-storage services. The appropriate architecture depends on whether data is dedicated to an individual system, shared between users, or accessed by other GIS components.
3. Size storage for performance: Evaluate latency, throughput, IOPS, concurrency, and workload characteristics when sizing storage. Storage capacity alone should not be used as an indication of expected performance.
4. Use local storage appropriately: Some EC2 configurations provide local instance storage that can offer high performance for temporary data, caches, and processing workspaces. Instance store data is temporary and should not be used as the only location for persistent GIS data.
5. Validate the complete data path: Test representative ArcGIS Pro projects using the same storage, network, and data architecture planned for production. Performance observed with data stored locally on the virtual machine may differ significantly from performance when the same data is accessed across a network.

Storage architecture should be designed around the requirements of the GIS data and workload rather than selecting a storage service based solely on its advertised performance specifications.

Why GPU Acceleration Matters

GPU acceleration is an important component of a well-performing virtualized ArcGIS Pro environment. ArcGIS Pro uses the GPU extensively for rendering and visualization, particularly with 3D scenes, imagery, complex symbology, and other graphics-intensive workflows.

GPU capabilities are also increasingly important beyond visualization. Some ArcGIS Pro geoprocessing and analysis workflows use NVIDIA CUDA libraries for GPU acceleration. This makes GPU architecture, available GPU memory, and driver compatibility important considerations when selecting and configuring an AWS environment.

More GPU resources do not automatically result in better ArcGIS Pro performance. GPU selection should reflect the requirements of the workload and be considered alongside CPU performance, system memory, storage, network performance, and data location.

For GPU-accelerated analysis, verify the requirements of the specific ArcGIS Pro workflow, including GPU memory and supported NVIDIA driver and CUDA versions.

Performance and User Experience

ArcGIS Pro performance in AWS depends on the complete environment. CPU performance, GPU resources, system memory, storage, network latency, data location, and workload characteristics all contribute to the user experience.

When designing and sizing the environment:

• CPU: Favor modern processors with strong per-core performance and approximately 3.0 GHz or greater clock speeds where possible.
• GPU: Size GPU and GPU memory for the expected visualization and analysis workload. More demanding 3D, imagery, and GPU-accelerated workflows may require additional GPU resources.
• System memory: Provide sufficient memory for ArcGIS Pro, the operating system, supporting applications, and expected workloads.
• Storage and data: Consider the complete path between ArcGIS Pro and its data. Storage latency, throughput, IOPS, data location, and concurrency can significantly affect performance.
• Network: Minimize latency between users, ArcGIS Pro compute resources, and GIS data. Interactive and file-based workflows can be particularly sensitive to network latency.
• Testing: Validate configurations using representative ArcGIS Pro projects and workflows. The ArcGIS Pro Performance Assessment Tool (PAT) can help establish repeatable performance baselines and compare configurations.

Selecting a larger instance does not necessarily result in better ArcGIS Pro performance. Increasing CPU, memory, or GPU resources provides limited benefit when another component—such as storage, network latency, or data access—is the primary constraint.

Size the environment for the workload, identify performance constraints through testing, and scale the resources that address those constraints.

Bringing It All Together

AWS provides several options for delivering GPU-accelerated ArcGIS Pro environments, allowing organizations to select an architecture based on workload requirements, user experience, scalability, management, and infrastructure control.

Amazon WorkSpaces provides persistent managed desktops, Amazon WorkSpaces applications provides application streaming for users who do not require a persistent desktop, and Amazon EC2 GPU instances provide greater control over the operating system, GPU drivers, storage, and supporting infrastructure.

Regardless of the deployment model, successful ArcGIS Pro environments should be designed as complete systems. CPU, GPU, memory, storage, network performance, data location, and workload characteristics should be evaluated together rather than sizing any single component in isolation.

Start with a configuration appropriate for the expected workload, test representative ArcGIS Pro projects and workflows, identify constraints, and scale the resources that address those constraints.