gpu_stats package

The GPU Stats Package handles collection of information from connected GPUs

gpu_stats._base module 

Parent class for obtaining Stats for various GPU/TPU backends. All GPU Stats should inherit from the _GPUStats class contained here.

class lib.gpu_stats._base.BiggestGPUInfo(card_id: int, device: str, free: float, total: float)

Bases: object

Dataclass for holding GPU Information about the card with most available VRAM.

card_id

Integer representing the index of the GPU device.

Type:: int

device

The name of the device

Type:: str

free

The amount of available VRAM on the GPU

Type:: float

total

the total amount of VRAM on the GPU

Type:: float

card_id: int

device: str

free: float

total: float

class lib.gpu_stats._base.GPUInfo(vram: list[int], vram_free: list[int], driver: str, devices: list[str], devices_active: list[int])

Bases: object

Dataclass for storing information about the available GPUs on the system.

Attributes:

vram: list[int]: List of integers representing the total VRAM available on each GPU, in MB.
vram_free: list[int]: List of integers representing the free VRAM available on each GPU, in MB.
driver: str: String representing the driver version being used for the GPUs.
devices: list[str]: List of strings representing the names of each GPU device.
devices_active: list[int]: List of integers representing the indices of the active GPU devices.

devices: list[str]

devices_active: list[int]

driver: str

vram: list[int]

vram_free: list[int]

lib.gpu_stats._base.set_exclude_devices(devices: list[int]) → None

Add any explicitly selected GPU devices to the global list of devices to be excluded from use by Faceswap.

Parameters:: devices (list[int]) – list of GPU device indices to exclude

Example

>>> set_exclude_devices([0, 1]) # Exclude the first two GPU devices

gpu_stats.apple_silicon module 

Collects and returns Information on available Apple Silicon SoCs in Apple Macs.

class lib.gpu_stats.apple_silicon.AppleSiliconStats(log: bool = True)

Bases: _GPUStats

Holds information and statistics about Apple Silicon SoC(s) available on the currently running Apple system.

Notes

Apple Silicon is a bit different from other backends, as it does not have a dedicated GPU with it’s own dedicated VRAM, rather the RAM is shared with the CPU and GPU. A combination of psutil and Tensorflow are used to pull as much useful information as possible.

Parameters:: log (bool, optional) – Whether the class should output information to the logger. There may be occasions where the logger has not yet been set up when this class is queried. Attempting to log in these instances will raise an error. If GPU stats are being queried prior to the logger being available then this parameter should be set to False. Otherwise set to True. Default: True

gpu_stats.amd module 

gpu_stats.cpu module 

Dummy functions for running faceswap on CPU.

class lib.gpu_stats.cpu.CPUStats(log: bool = True)

Bases: _GPUStats

Holds information and statistics about the CPU on the currently running system.

Notes

The information held here is not useful, but _GPUStats is dynamically imported depending on the backend used, so we need to make sure this class is available for Faceswap run on the CPU Backend.

The base _GPUStats handles the dummying in of information when no GPU is detected.

Parameters:: log (bool, optional) – Whether the class should output information to the logger. There may be occasions where the logger has not yet been set up when this class is queried. Attempting to log in these instances will raise an error. If GPU stats are being queried prior to the logger being available then this parameter should be set to False. Otherwise set to True. Default: True

gpu_stats.directml module 

gpu_stats.nvidia_apple module 

Collects and returns Information on available Nvidia GPUs connected to Apple Macs.

class lib.gpu_stats.nvidia_apple.NvidiaAppleStats(log: bool = True)

Bases: _GPUStats

Holds information and statistics about Nvidia GPU(s) available on the currently running Apple system.

Notes

PyNvx is used for hooking in to Nvidia’s Machine Learning Library and allows for pulling fairly extensive statistics for Apple based Nvidia GPUs

Parameters:: log (bool, optional) – Whether the class should output information to the logger. There may be occasions where the logger has not yet been set up when this class is queried. Attempting to log in these instances will raise an error. If GPU stats are being queried prior to the logger being available then this parameter should be set to False. Otherwise set to True. Default: True

gpu_stats.nvidia module 

Collects and returns Information on available Nvidia GPUs.

class lib.gpu_stats.nvidia.NvidiaStats(log: bool = True)

Bases: _GPUStats

Holds information and statistics about Nvidia GPU(s) available on the currently running system.

Notes

PyNVML is used for hooking in to Nvidia’s Machine Learning Library and allows for pulling fairly extensive statistics for Nvidia GPUs

Parameters:: log (bool, optional) – Whether the class should output information to the logger. There may be occasions where the logger has not yet been set up when this class is queried. Attempting to log in these instances will raise an error. If GPU stats are being queried prior to the logger being available then this parameter should be set to False. Otherwise set to True. Default: True

gpu_stats.rocm module 

Collects and returns Information about connected AMD GPUs for ROCm using sysfs and from modinfo

As no ROCm compatible hardware was available for testing, this just returns information on all AMD GPUs discovered on the system regardless of ROCm compatibility.

It is a good starting point but may need to be refined over time

class lib.gpu_stats.rocm.ROCm(log: bool = True)

Bases: _GPUStats

Holds information and statistics about GPUs connected using sysfs

Parameters:: log (bool, optional) – Whether the class should output information to the logger. There may be occasions where the logger has not yet been set up when this class is queried. Attempting to log in these instances will raise an error. If GPU stats are being queried prior to the logger being available then this parameter should be set to False. Otherwise set to True. Default: True