mt.gpu

Submodules

• mt.gpu.arch
• mt.gpu.arch_amd64_cpu
• mt.gpu.arch_rock
• mt.gpu.arch_rpi
• mt.gpu.arch_tegra
• mt.gpu.version
  • Variables
    • MAJOR_VERSION
    • MINOR_VERSION
    • PATCH_VERSION
    • version

Functions

• detect_machine(): Detects the machine architecture.

• get_mem_info(): Returns a dictionary containing information about detected CPU/GPU devices and their memory usage, or None if the architecture is unknown.

• sort_gpu_devices(): Sorts the GPU devices in decreasing order of free memory. Only effective with CUDA hardare.

mt.gpu.detect_machine()

Detects the machine architecture.

Returns:

a unique name identifying the machine architecture

Return type:

[‘arm64-r5b’, ‘arm64-rpi’, ‘arm64-tx1’, ‘arm64-tx2’, ‘arm64-j43’, ‘amd64-cpu’, ‘amd64-nvidia’, ‘amd64-amd’]

Notes

The names correspond to the following architectures:

• “arm64-r5b” : a Rock 5B

• “arm64-rp4” : a Raspberry Pi Model 4

• “arm64-rp3” : a Raspberry Pi Model 3

• “arm64-tk1” : an Nvidia Tegra K1

• “arm64-tx1” : an Nvidia Tegra X1

• “arm64-tx2” : an Nvidia Tegra X2

• “arm64-xnx” : an Nvidia Jetson Xavier-NX

• “amd64-cpu” : an amd64 PC without any graphic card

• “amd64-nvidia” : an amd64 PC with Nvidia graphic card(s)

• “amd64-amd” : and amd64 PC with AMDGPU card(s)

mt.gpu.get_mem_info(print_bars=False)

Returns a dictionary containing information about detected CPU/GPU devices and their memory usage, or None if the architecture is unknown.

Parameters:

print_bars (bool) – whether or not to print some bars

mt.gpu.sort_gpu_devices()

Sorts the GPU devices in decreasing order of free memory. Only effective with CUDA hardare.