UM HPC Cost Calculator

A modern web application for calculating job costs on the University of Michigan Great Lakes and Armis2 High Performance Computing clusters.

Features

• Accurate rate calculations using published rates and TRES weights
• Multiple clusters: Great Lakes and Armis2
• Multiple partitions per cluster (standard, largemem, gpu, and more)
• Real-time cost updates as you adjust parameters
• Responsive design for desktop and mobile
• Detailed breakdown of resources and dominant billing factor

Great Lakes partitions

• Standard/Debug (CPU=2505, Memory=358)
• Large Memory (CPU=7704, Memory=185)
• GPU (CPU=1370, Memory=304, GPU=27391)
• SPGPU (CPU=4520, Memory=377, GPU=18079)
• GPU MIG40 (CPU=3424, Memory=221, GPU=27391)

Armis2 partitions

• Standard/Debug (derived weights CPU≈2900, Memory≈414)
• Large Memory (derived weights CPU≈8037, Memory≈299)
• GPU (derived weights CPU≈2900, Memory≈414, GPU≈28157)

Note

Billing weights were pulled using scontrol on 07/01/2025.

Getting Started

Prerequisites

• Node.js (version 16 or higher)
• npm or yarn

Installation

1. Install dependencies:

   npm install

2. Start the development server:

   npm run dev

3. Open your browser and navigate to http://localhost:5173

Building for Production

npm run build

The built files will be in the dist directory.

Usage

1. Select Cluster: Choose Great Lakes or Armis2
2. Select Job Type: Choose between Single Core, Multicore, or Array job
   • Single Core Job: Limited to 1 core (1 node, 1 task, 1 core)
   • Multicore Job: For jobs requiring multiple cores with shared memory
   • Array Job: For multiple independent jobs with identical resource requirements
3. Select Partition: Choose the appropriate partition for your job
4. Configure Resources:
   • Set the number of CPU cores needed (automatically set to 1 for Single Core jobs)
   • Specify memory requirements
   • For GPU partitions, set the number of GPUs
   • For Array jobs, set the number of jobs in the array
5. Set Runtime: Enter the expected job duration (days, hours, minutes, seconds)
6. View Cost: The estimated cost will be calculated automatically, including detailed breakdown
7. Generate SLURM Script: Expand the SLURM script section to view and copy an example batch script

Cost Calculation

The calculator uses the official Great Lakes TRES billing weights and calculates costs as follows:

• Billing Formula:
  billing = max(cpu_weight × cores, mem_weight × memory_GB, gpu_weight × gpus)
  (weights depend on partition; see above)
• Cost Formula:
  cost = (total_minutes × billing) ÷ 10,000,000

The dominant resource (CPU, memory, or GPU) determines the billing for each job. The calculator displays a detailed breakdown of each component and which factor is dominant.

For Armis2, weights are derived to match published per-minute rates for baseline configurations:

• standard/debug: 1 core & 7 GB ≈ $0.000290046/min
• largemem: 1 core & 26.89 GB ≈ $0.000803704/min
• gpu: 1 GPU with 5 cores & 15 GB ≈ $0.002815741/min If you have exact TRES weights, update them in CLUSTER_CONFIG.armis2 in src/App.jsx.

Important Notes

• Costs shown are maximum estimates and don't account for:

  • UMRCP (University of Michigan Research Computing Package) allocations
  • Unit cost-sharing programs
  • Other funding sources

• For accurate billing information, always use the official my_job_estimate command on Great Lakes or Armis2

• This calculator now uses the same TRES billing weights and formulas as the official system for maximum accuracy.

Rates Source

Rates are based on the official service rates:

• Great Lakes: https://its.umich.edu/advanced-research-computing/high-performance-computing/great-lakes/rates
• Armis2: https://its.umich.edu/advanced-research-computing/high-performance-computing/armis2/rates

Partition limits and node capacities used in defaults:

• Great Lakes:
  • Defaults, Limits, and Storage
  • Configuration
• Armis2:
  • Defaults, Limits, and Storage
  • Configuration