The ExaScale Index (ESI) is a structured way to summarize how close a computing system, platform, or ecosystem is to achieving exascale capability—commonly understood as order-of-magnitude compute at or exceeding 1018 operations per second.
Rather than treating exascale as a single threshold, ESI expresses readiness across multiple dimensions that must advance together for exascale capability to be practical, repeatable, and sustainable.
Purpose
- High-level comparison of architectures and system designs
- Consistent progress tracking over time
- Distinguishing peak demonstrations from deployable capability
- Encouraging transparent assumptions
Core Dimensions
1) Compute Throughput
Operations per second under defined workload and precision.
2) Energy Efficiency
Performance per watt including infrastructure overhead.
3) Scalability & Interconnect
Growth efficiency as nodes are added.
4) Memory & Data Movement
Ability to feed compute with sufficient data bandwidth.
5) Reliability & Fault Tolerance
Error handling and recovery at extreme scale.
6) Practical Deployability
Cost, manageability, and repeatability.
Scope Notes
- ESI is not a vendor benchmark.
- Workload definition must be stated with any score.
- Applicable to HPC, AI, and hybrid systems.