Compute Resources
Roman Research Nexus provides cloud-based compute resources that support a wide range of scientific workflows, including interactive notebook sessions, software development, scientific analysis, simulations, and machine learning applications. Users can select from several compute configurations that balance CPU cores, memory capacity, GPU resources, and credit consumption to match the needs of their work.
Planned Update - The server configurations described in this article represent the next generation of Roman Research Nexus compute resources and are scheduled to become available during the second week of December 2026. Current users may notice differences between the options shown here and those currently available in the platform during the transition period.
Compute Resources and Account Types
The compute resources available to you depend on your account type:
- Individual accounts can launch General-Purpose Servers, which provide resources for notebook-based analysis, software development, tutorials, and routine scientific workflows.
- Team accounts can launch all available compute resources, including Parallel Computing Servers, Memory-Optimized Servers, and GPU Servers.
Any user may request a team account through the standard account request process. For information on creating and managing team accounts, see the Team Accounts documentation.
General Purpose Servers
General-purpose servers are available to all users and are the recommended starting point for most users of the Roman Research Nexus. They support all currently available Roman Research Nexus tutorials and are well suited for notebook-based analysis, exploration of Roman data products, visualization, routine scientific analysis, and software development and testing. The Table of General Purpose Server Configurations summarizes the servers available to all users, including their CPU capacity, memory allocation, and hourly credit consumption.
General Purpose Server Configurations
| Tier | Instance | vCPU | Memory | Credits / Hour |
|---|---|---|---|---|
| Small | r5.large | 2 | 16 GB | 0.126 |
| Medium | r5.2xlarge | 8 | 64 GB | 0.504 |
Team Compute Resources
Team accounts provide access to additional compute resources for analyses and applications that require more CPU cores, memory, or GPU acceleration than are available through the general-purpose server tiers.
GPU resources differ from the standard CPU-based server offerings. While the CPU server tiers include the Roman software stack, tutorials, and mission-supported tools, GPU servers provide a more basic configuration intended for GPU-enabled applications. Users are responsible for installing and maintaining any additional software required by their workflows. As a result, GPU resources are best suited for experienced users who are familiar with the requirements of their applications and can manage their own software stack.
GPU resource availability depends on Amazon Web Services (AWS) capacity and cannot be guaranteed at all times. Users planning GPU-intensive scientific analyses, particularly those requiring substantial GPU resources during a specific time period, are encouraged to contact the Science Operations Center (SOC) through the Roman Help Desk in advance. The SOC will work with users to coordinate resource allocation based on their anticipated needs and available AWS capacity.
The Table of Team Server Configurations summarizes the compute resources available to team accounts, including CPU capacity, memory allocation, GPU resources, and hourly credit consumption.
Team Server Configurations
| Tier | Instance | vCPU | Memory | Credits / Hour |
|---|---|---|---|---|
| Parallel Computing Resources | ||||
| Parallel-16 | m7a.4xlarge | 16 | 64 GB | 0.93 |
| Parallel-32 | m7a.8xlarge | 32 | 128 GB | 1.85 |
| Parallel-64 | m7a.16xlarge | 64 | 256 GB | 3.71 |
| Memory Optimized Resources | ||||
| Large-MEM | r5.4xlarge | 16 | 128 GB | 1.008 |
| XXL-MEM | r7i.16xlarge | 64 | 512 GB | 4.23 |
| GPU Resources | ||||
| GPU-L | g7.2xlarge | 8 | 32 GB | 2.52 |
| GPU-XL | g7e.2xlarge | 8 | 64 GB | 3.63 |
GPU specifications: GPU-L provides one NVIDIA RTX PRO 4500 Blackwell GPU with 32 GB of GPU memory, while GPU-XL provides one NVIDIA RTX PRO 6000 Blackwell GPU with 96 GB of GPU memory.
Choosing a Server and Managing Compute Costs
Selecting the right server depends on the computing requirements of your application. In general, users should choose the smallest server that comfortably supports their work and move to larger resources only when additional CPU cores, memory, or GPU capabilities are required.
- General-Purpose Servers are the recommended starting point for most users and support all currently available Roman Research Nexus tutorials.
- Parallel Servers provide additional CPU cores and are generally the most cost-effective option when your application can take advantage of parallel processing.
- Memory-Optimized Servers provide the same number of CPU cores as comparable parallel server tiers, but with substantially more memory.
- GPU Servers should be selected only when your software explicitly requires GPU acceleration. Roman Research Nexus tutorials, workflows, and mission-supported software are designed for the standard CPU server tiers.
Compute resources consume credits while a server is running. Credit consumption depends on the selected server type and the duration of use. Credits are charged for the entire time a server is running, regardless of utilization. A server waiting for user input consumes credits at the same rate as a server actively executing code. Larger servers consume credits more quickly because they provide additional compute, memory, or GPU resources. To make efficient use of available credits:
- Select the smallest server that meets the needs of your work.
- Move to a larger server only when additional resources are needed.
- Use Memory-Optimized servers only when additional memory is required, as they consume more credits than comparable Parallel server tiers.
- Use GPU servers only when your software explicitly requires GPU acceleration.
- Stop servers when they are no longer needed.
Managing Idle Servers
To conserve shared resources and reduce unnecessary credit consumption, Nexus automatically stops servers that are not actively being used.
Two rules determine when a server is stopped:
If no notebook kernels or terminals are running, the server stops after 30 minutes. This typically occurs after all notebooks and terminals have been shut down.
If notebook kernels or terminals remain open but are idle, they are automatically shut down after 4 hours of inactivity. A notebook kernel is considered idle when it is not executing code. A terminal is considered idle when no activity is occurring within it. Once all kernels and terminals have been shut down, the 30‑minute shutdown timer begins. As a result, a notebook left open but unused will typically cause the server to stop after approximately 4.5 hours.
A notebook cell that is actively executing code is not considered idle and will continue running. However, long-running jobs launched in the background of a terminal may appear idle to the system. To prevent these jobs from being terminated, run them within notebook cells or configure them to produce periodic output.
When a server is stopped, all running kernels, terminals, and processes are terminated. Files that have been saved to persistent storage remain available, but any unsaved work or data held only in memory is lost. Users should save their work regularly and stop servers when they are no longer needed.
Monitoring Credit Usage
Roman Research Nexus provides tools for monitoring compute, storage, and data egress consumption. Regularly reviewing resource usage can help individuals and teams manage their credit allocations and plan future work. Individual users can monitor the credits associated with their personal allocations, while team members and team administrators can monitor resource consumption across team accounts.
The Cost Monitor dashboard provides visibility into:
- Compute credit consumption
- Storage usage
- Data egress usage
- Remaining credit balances
- Historical usage trends
Users running long-duration analyses or using higher-capacity compute resources should periodically review their usage to ensure credits are being consumed as expected. For detailed information about viewing and managing resource consumption, see the documentation on the Roman Research Nexus.
For additional questions not answered in this article, please contact the Roman Help Desk.
