The new M1 MacBook Pro 13-inch has a shared CPU/CPU memory architecture. Here we compare the speed of the Apple M1 GPU to some other Macs.
M1 Speedtest devices
|MacBook Pro (2020)||M1 Apple||16 GB||M1 Apple|
|MacBook Pro (2018)||2,9 GHz 6-Core Intel Core i9||32 GB 2400 MHz DDR4||Radeon Pro 560X 4 GB|
|Mac Pro (2019)||3.2 GHz 16-Core Intel Xeon W||96 GB 2933 MHz DDR4||AMD Radeon Pro Vega II Duo 32 GB|
|MacBook Air (Early 2014)||1,7 GHz Dual-Core Intel Core i7||8 GB 1600 MHz DDR3||Intel HD Graphics 5000 1536 MB|
Computing the energy grid
The energy grid is computed by dividing the unit cell in 128x128x128 voxels, and for each voxel the potential energy is computing by looping over all atoms in the unit cell. The interactions is a simple Lennard-Jones potential using a cutoff of 12 Angstrom. This implies that, to satisfy the minimum image convention, that unit cells that are too small are automatically using super cells. The number of atoms denoted in the table are therefore of the super cell and not of the unit cell.
M1 Speedtest data (in milliseconds; smaller is better).
|#atoms||M1 Apple||MacBook Pro 2018||Mac Pro 2018||Mac Air 2014|
|MIL-101 (14416 atoms)||989.6||757.0||228.6||5303.2|
|NU-1301 (50192 atoms)||3286.5||2354.0||397.9||11465.7|
Times in milliseconds. Smaller is better.
The results for the Apple M1 are excellent. They are on par with the Radeon Pro 560X and faster for up to about a thousand atoms. And all of this with a much lower battery cost. Note that the real power horse, the desktop Vega II duo is by far the fastest GPU. However, the future for the Apple M1 system-on-a-socket looks very bright.