-
Notifications
You must be signed in to change notification settings - Fork 5
Performance Benchmarks
Christopher Lorton edited this page Feb 20, 2024
·
1 revision
- 754 LGAs from 2015 Nigeria population data
- 183M agents total
| Machine | CPU/GPU | Est. 10 year simulation time* | Notes |
|---|---|---|---|
| IDM Self-Service VM | Nvidia A40 | 00:55 | Taichi implementation on CUDA |
| Apple MacBook Pro | Apple M1 Max | 01:53 | Taichi implementation on Metal |
| Lenovo ThinkPad X1 Extreme | Nvidia GeForce GTX 1650 | 4:58 | Taichi implementation on CUDA |
| Apple MacBook Pro | Apple M1 Max | 25:55 | Numba optimized NumPy implementation |
| GitHub codespace | AMD EPYC 7763 | 36:46 | Numba optimized NumPy implementation on 16 cores |
| Lenovo ThinkPad X1 Extreme | Intel Core i7-1085H | 46:11 | Numba optimized NumPy implementation on 6 cores (WSL Ubuntu) |
*Extrapolated from 1-year simulation time.
Note: These simulations started with an initial, fully susceptible population and simulated an outbreak in this population which is compute intensive. We need an updated simulation representing steady-state behavior with typical measles prevalence (single digit or less prevalence and significant fraction of the population in the recovered/immune state).