pyrsmi

pyrsmi

Python Bindings for System Management Library for AMD GPUs

---

• pyrsmi is a python package of rocm-smi-lib, for providing a limited features of ROCm System Management Library for assisting Python development involving AMD GPUs.
• It is based on (rocm-smi-lib)[https://github.com/RadeonOpenCompute/rocm_smi_lib], so its scope of support should be similar to that of the latter.

Requirements

• pyrsmi runs on latest ROCm-supported Instinct MI-series GPU systems. Scope of tested systems is limited; please create tickets for any issues encountered.

How to install from source

• Clone this repo:
  • git clone https://github.com/ROCm/pyrsmi
  • cd pyrsmi
• python -m pip install -e .
• pyrsmi can be installed as PyPA-compatible Python package.

How to install Python packages

• Install build package:
  • pip install build
• At the top directory (where setup.py is), run: python -m build
• Then by default packages (both sdist and wheel) will be built under dist directory.
• The packages can be either installed with pip install, or be uploaded to PyPI (release or test) repo, or an artifactory of your choice. The latter can be installed liked beflow.

How to install from PyPI

• NOTE: Release versions are available at pypi.org, and the package can be installed with :
  • python -m pip install pyrsmi

How to use pyrsmi

• Environment setup:
  • pyrsmi searches for rocm-smi library from default ROCm environment variable ROCM_PATH. So, for standard ROCm installation, the library will be automatically detected. In some unusual case where the ROCm installation is moved, make sure to set the environment variable as so :

    $ mv /opt/rocm/ /usr/local/xyz/
    $ export ROCM_PATH=/usr/local/xyz/
    

• Running in python:

  from pyrsmi import rocml
  
  rocml.smi_initialize()
  print(rocml.smi_get_device_count())
  rocml.smi_shutdown()
  

Examples

• Examples directory contains a number of code snippets showing how to use the package.
• It also contains an example showing how to use pyrsmi to create a web-based system monitoring tool that displays various dashboards of system status, including memory, CPU/GPU utilization and process names.

List of API functions

• The list is not exhaustive. Please refer to the code for missing functions.

Function	Description	Argument	Return Type	Note
smi_initialize	initialize rsmi	None	None
smi_shutdown	shut down rsmi	None	None
smi_get_version	get version of rsmi	None	str	'major.minor.patch'
smi_get_kernel_version	get version of ROCm kernel driver	None	str
smi_get_device_id	get device id of GPU devices	None	uint64	id of devices
smi_get_device_count	get number of GPU devices	None	int	num of devices
smi_get_device_name	get name of GPU devices	None	str
smi_get_device_unique_id	get unique id of GPU devices	None	int	64bit integer
smi_get_device_utilization	get device utilization in % busy	device_id	int
smi_get_device_memory_used	get device memory usage	device_id	int	in Bytes, type 'VRAM'
smi_get_device_memory_total	get device's total memory	device_id	int	in Bytes, type 'VRAM'
smi_get_device_memory_busy	get percentage of time busy accessing memory	device_id	int	in percent time
smi_get_device_memory_reserved_pages	get info of reserved device memory	device_id	(# pages, ptr to block)
smi_get_device_pcie_bandwidth	get device's estimated PCIe bandwidth	device_id	float	in Bytes/sec
smi_get_device_compute_process	get list of pid of processes running on the system	None	List[int]
smi_get_device_average_power	get device's average power	device_id	float	power in Watt
smi_get_device_xgmi_error_status	get XGMI error status for the device	device_id	int
smi_reset_device_xgmi_error	get device's average power	device_id	float	power in Watt
smi_get_device_compute_partition	get device's compute partition	device_id	str	e.g. 'SPX', 'CPX'
smi_get_device_memory_partition	get device's memory partition	device_id	str	e.g. 'NPS1'
smi_get_device_link_type	gets hops and types of link between two devices	(device_id, device_id)	(int, int)	(n_hops, type)
smi_get_device_uuid	gets UUID of the device	(device_id, format)	str	default with 'GPU-' prefix