- Target model
- Precision & context
- Workload
- …
Model selected: assess hardware options
Use the target model, precision, context, and workload to assess memory and resource needs, and compare hardware and deployment configurations.
Optimizer
BitCloud Atlas
AI infrastructure planning platform
Make informed choices about models and compute.
Turn model and hardware requirements into deployment plans.
How much compute does your model need? Which models fit your existing devices? Does a configuration meet deployment requirements? BitCloud Atlas brings capacity assessment, model and hardware selection, and comparison together, helping teams understand candidate configurations, constraints, and supporting evidence.
Candidate options
Plan and compare both ways
Constraints
Planning illustration
- Device type
- Device count
- Operating conditions
- …
Hardware available: explore model options
Use device types, quantities, and operating conditions as constraints to explore suitable models and configurations.
Combination selected: check deployment conditions
Check capacity, parallel configuration, and other constraints for a model–hardware combination. Review unmet conditions and suggested adjustments.
- Capacity & resources
- Parallel configuration
- Limits & suggestions
Compare plans on a common basis.
Compare capacity, deployment constraints, and operating costs.
Capacity & workload
Estimate resource needs from model weights, runtime cache, context, and concurrency.
Configuration & deployment constraints
Check device counts, parallel configuration, and full replicas to identify limitations.
Cost & operating conditions
Compare estimated power and cloud resource costs using selected configurations and inputs to inform planning discussions.
Understand the results and the evidence behind them.
Distinguish estimates, official deployment configurations, and measurements, with sources and applicable conditions.
Theoretical estimates
Calculated from model and hardware parameters and assumptions. Useful for initial planning, but not equivalent to runtime performance.
Official deployment configurations
Reference model, hardware, and runtime configurations from official documentation. They inform deployment choices, not performance conclusions.
Measured results
Review records for specific devices, models, versions, and workloads. Reassess applicability when conditions differ.
Use Atlas in your team’s workflow.
Multiple ways to use Atlas across your team’s workflows.
Web interface
Explore models and hardware, assess capacity, and compare options.
API
Query deployment suggestions with consistent calculations and result explanations.
Read-only MCP
Let agents query catalogs and deployment suggestions in read-only mode to support selection discussions.
Clarify the key conditions before choosing a plan.
Support model selection, resource use, and technical reviews.
Choose hardware for a target model
Compare resource requirements before procurement or deployment, and identify conditions that need further validation.
Assess options for existing devices
Explore models and configurations on available hardware to inform resource use and expansion.
Organize a technical review
Bring requirements, configurations, estimates, and sources together so presales, technical, and delivery teams can discuss trade-offs.
Planning and execution serve different tasks.
Atlas: selection, resource assessment, and conditions to confirm. BitTune: deployment, tuning, and validation.
BitCloud Atlas
Capacity assessment and planning
BitTune
Deployment, tuning and validation
FAQ
Does Atlas deploy models or run tuning directly?
This page covers capacity assessment, selection, and deployment planning. For deployment, tuning, and validation in an actual environment, explore BitTune and discuss your devices and requirements.
Are calculated results the same as actual performance?
No. Estimates, official deployment configurations, and measurements serve different purposes. Actual performance depends on devices, models, engines, configurations, and workloads, and needs validation in the target environment.
Does a catalog entry mean a model or device has been validated?
No. Catalog entries describe model and hardware information. Deployment evidence and validation scope must be reviewed separately against the version and configuration.
Can I start with hardware before choosing a model?
Yes. Explore model and configuration options from available hardware, or assess hardware requirements after selecting a target model.
Does the public project include every capability described here?
The public project provides its implementation and usage documentation. Available capabilities, deployment methods, and licensing depend on the project documentation and version.
How do I get started?
Explore the public project first. For a planning discussion, email BitCloud with your known model or hardware, expected workload, and current questions.
BitCloud Atlas