Dear all,
A friend of mine got working docker on the M1 processor and ran a few benchmarks both on top Intel processors (available on Mac) and on M1. The benchmarks are from the paper Tikidji-Hamburyan et al. 2015 Software for Brain Network Simulations: A Comparative Study and were slightly modified compared with the original version.
Hardware
- MacMini Intel(R) Core™ i7-8700B CPU @ 3.20GHz (6C 12T) 64GB RAM
Docker 12C 16GB RAM - MacBook Pro 13" M1 16GB
Docker 8C 10GB RAM
Results for the single thread
Testing 10 times: 5000 leaky integrate-and-fire neurons with random connections
Task | intel | m1 |
---|---|---|
Building time Simulation time |
26.21 s 28.83 s
|
17.17 s 19.69 s
|
Building time Simulation time |
26.18 s 11.66 s
|
17.18 s 14.71 s
|
Building time Simulation time |
7.04 s 3.88 s
|
3.56 s 2.37 s
|
Building time Simulation time |
7.06 s 3.86 s
|
3.51 s 2.36 s
|
Building time Simulation time |
7.10 s 13.38 s
|
3.54 s 2.36 s
|
Building time Simulation time |
7.12 s 3.87 s
|
3.58 s 2.35 s
|
Building time Simulation time |
7.03 s 1.38 s
|
3.51 s 0.63 s
|
Building time Simulation time |
7.10 s 3.85 s
|
3.53 s 2.37 s
|
Building time Simulation time |
7.04 s 3.88 s
|
3.48 s 0.63 s
|
Building time Simulation time |
7.04 s 1.40 s
|
3.49 s 2.36 s
|
Testing 10 times: 500 HH neurons with sparse double exponential connections
Results | intel | m1 |
---|---|---|
Building time Simulation time |
5.45 s 13.19 s
|
3.64 s 8.54 s
|
Building time Simulation time |
5.05 s 5.43 s
|
3.47 s 3.15 s
|
Building time Simulation time |
0.18 s 4.43 s
|
0.09 s 5.74 s
|
Building time Simulation time |
0.18 s 9.13 s
|
0.09 s 2.40 s
|
Building time Simulation time |
0.18 s 4.43 s
|
0.09 s 2.38 s
|
Building time Simulation time |
0.19 s 4.43 s
|
0.09 s 2.39 s
|
Building time Simulation time |
0.18 s 4.44 s
|
0.09 s 2.39 s
|
Building time Simulation time |
0.19 s 4.47 s
|
0.09 s 2.39 s
|
Building time Simulation time |
0.18 s 4.40 s
|
0.09 s 2.41 s
|
Building time Simulation time |
0.18 s 4.44 s
|
0.09 s 2.39 s
|
Results OpenMP multithreading
Testing 10 times: 5000 leaky integrate-and-fire neurons with random connections
Results | intel | m1 |
---|---|---|
Building time Simulation time |
7.05 s 4.60 s
|
3.43 s 3.88 s
|
Building time Simulation time |
7.03 s 5.49 s
|
3.46 s 4.50 s
|
Building time Simulation time |
7.06 s 5.51 s
|
3.58 s 4.24 s
|
Building time Simulation time |
7.04 s 5.12 s
|
3.47 s 4.41 s
|
Building time Simulation time |
6.99 s 5.37 s
|
3.46 s 4.30 s
|
Building time Simulation time |
7.04 s 5.53 s
|
3.47 s 4.35 s
|
Building time Simulation time |
7.01 s 4.43 s
|
3.47 s 4.42 s
|
Building time Simulation time |
7.07 s 5.18 s
|
3.47 s 3.75 s
|
Building time Simulation time |
7.00 s 5.11 s
|
3.58 s 4.55 s
|
Building time Simulation time |
7.08 s 5.28 s
|
3.45 s 4.07 s
|
and Testing 10 times: 500 HH neurons with sparse double exponential connections
Results | intel | m1 |
---|---|---|
Building time Simulation time |
0.19 s 3.70 s
|
0.09 s 2.36 s
|
Building time Simulation time |
0.18 s 3.79 s
|
0.08 s 2.30 s
|
Building time Simulation time |
0.20 s 3.78 s
|
0.09 s 2.36 s
|
Building time Simulation time |
0.19 s 4.05 s
|
0.09 s 2.27 s
|
Building time Simulation time |
0.18 s 3.67 s
|
0.08 s 2.34 s
|
Building time Simulation time |
0.18 s 3.72 s
|
0.09 s 2.34 s
|
Building time Simulation time |
0.18 s 3.80 s
|
0.09 s 2.37 s
|
Building time Simulation time |
0.18 s 3.75 s
|
0.09 s 2.27 s
|
Building time Simulation time |
0.18 s 3.84 s
|
0.09 s 2.29 s
|
Building time Simulation time |
0.18 s 3.74 s
|
0.09 s 2.31 s
|
If anyone wants to run a quick test on M1, let me know.