Python bindings to the NVIDIA Management Library
================================================
Provides a Python interface to GPU management and monitoring functions.
This is a wrapper around the NVML library.
For information about the NVML library, see the NVML developer page
http://developer.nvidia.com/nvidia-management-library-nvml
As of version 11.0.0, the NVML-wrappers used in pynvml are identical
to those published through [nvidia-ml-py](https://pypi.org/project/nvidia-ml-py/).
Note that this file can be run with 'python -m doctest -v README.txt'
although the results are system dependent
Requires
--------
Python 3, or an earlier version with the ctypes module.
Installation
------------
pip install .
Usage
-----
You can use the lower level nvml bindings
```python
>>> from clore_pynvml import *
>>> nvmlInit()
>>> print("Driver Version:", nvmlSystemGetDriverVersion())
Driver Version: 410.00
>>> deviceCount = nvmlDeviceGetCount()
>>> for i in range(deviceCount):
... handle = nvmlDeviceGetHandleByIndex(i)
... print("Device", i, ":", nvmlDeviceGetName(handle))
...
Device 0 : Tesla V100
>>> nvmlShutdown()
```
Or the higher level nvidia_smi API
```python
from clore_pynvml.smi import nvidia_smi
nvsmi = nvidia_smi.getInstance()
nvsmi.DeviceQuery('memory.free, memory.total')
```
```python
from clore_pynvml.smi import nvidia_smi
nvsmi = nvidia_smi.getInstance()
print(nvsmi.DeviceQuery('--help-query-gpu'), end='\n')
```
Functions
---------
Python methods wrap NVML functions, implemented in a C shared library.
Each function's use is the same with the following exceptions:
- Instead of returning error codes, failing error codes are raised as
Python exceptions.
```python
>>> try:
... nvmlDeviceGetCount()
... except NVMLError as error:
... print(error)
...
Uninitialized
```
- C function output parameters are returned from the corresponding
Python function left to right.
```c
nvmlReturn_t nvmlDeviceGetEccMode(nvmlDevice_t device,
nvmlEnableState_t *current,
nvmlEnableState_t *pending);
```
```python
>>> nvmlInit()
>>> handle = nvmlDeviceGetHandleByIndex(0)
>>> (current, pending) = nvmlDeviceGetEccMode(handle)
```
- C structs are converted into Python classes.
```c
nvmlReturn_t DECLDIR nvmlDeviceGetMemoryInfo(nvmlDevice_t device,
nvmlMemory_t *memory);
typedef struct nvmlMemory_st {
unsigned long long total;
unsigned long long free;
unsigned long long used;
} nvmlMemory_t;
```
```python
>>> info = nvmlDeviceGetMemoryInfo(handle)
>>> print "Total memory:", info.total
Total memory: 5636292608
>>> print "Free memory:", info.free
Free memory: 5578420224
>>> print "Used memory:", info.used
Used memory: 57872384
```
- Python handles string buffer creation.
```c
nvmlReturn_t nvmlSystemGetDriverVersion(char* version,
unsigned int length);
```
```python
>>> version = nvmlSystemGetDriverVersion();
>>> nvmlShutdown()
```
For usage information see the NVML documentation.
Variables
---------
All meaningful NVML constants and enums are exposed in Python.
The NVML_VALUE_NOT_AVAILABLE constant is not used. Instead None is mapped to the field.
NVML Permissions
----------------
Many of the `pynvml` wrappers assume that the underlying NVIDIA Management Library (NVML) API can be used without admin/root privileges. However, it is certainly possible for the system permissions to prevent pynvml from querying GPU performance counters. For example:
```
$ nvidia-smi nvlink -g 0
GPU 0: Tesla V100-SXM2-32GB (UUID: GPU-96ab329d-7a1f-73a8-a9b7-18b4b2855f92)
NVML: Unable to get the NvLink link utilization counter control for link 0: Insufficient Permissions
```
A simple way to check the permissions status is to look for `RmProfilingAdminOnly` in the driver `params` file (Note that `RmProfilingAdminOnly == 1` means that admin/sudo access is required):
```
$ cat /proc/driver/nvidia/params | grep RmProfilingAdminOnly
RmProfilingAdminOnly: 1
```
For more information on setting/unsetting the relevant admin privileges, see [these notes](https://developer.nvidia.com/nvidia-development-tools-solutions-ERR_NVGPUCTRPERM-permission-issue-performance-counters) on resolving `ERR_NVGPUCTRPERM` errors.
Release Notes
-------------
- Version 2.285.0
- Added new functions for NVML 2.285. See NVML documentation for more information.
- Ported to support Python 3.0 and Python 2.0 syntax.
- Added nvidia_smi.py tool as a sample app.
- Version 3.295.0
- Added new functions for NVML 3.295. See NVML documentation for more information.
- Updated nvidia_smi.py tool
- Includes additional error handling
- Version 4.304.0
- Added new functions for NVML 4.304. See NVML documentation for more information.
- Updated nvidia_smi.py tool
- Version 4.304.3
- Fixing nvmlUnitGetDeviceCount bug
- Version 5.319.0
- Added new functions for NVML 5.319. See NVML documentation for more information.
- Version 6.340.0
- Added new functions for NVML 6.340. See NVML documentation for more information.
- Version 7.346.0
- Added new functions for NVML 7.346. See NVML documentation for more information.
- Version 7.352.0
- Added new functions for NVML 7.352. See NVML documentation for more information.
- Version 8.0.0
- Refactor code to a nvidia_smi singleton class
- Added DeviceQuery that returns a dictionary of (name, value).
- Added filter parameters on DeviceQuery to match query api in nvidia-smi
- Added filter parameters on XmlDeviceQuery to match query api in nvidia-smi
- Added integer enumeration for filter strings to reduce overhead for performance monitoring.
- Added loop(filter) method with async and callback support
- Version 8.0.1
- Restructuring directories into two packages (pynvml and nvidia_smi)
- Adding initial tests for both packages
- Some name-convention cleanup in pynvml
- Version 8.0.2
- Added NVLink function wrappers for pynvml module
- Version 8.0.3
- Added versioneer
- Fixed nvmlDeviceGetNvLinkUtilizationCounter bug
- Version 8.0.4
- Added nvmlDeviceGetTotalEnergyConsumption
- Added notes about NVML permissions
- Fixed version-check testing
- Version 11.0.0
- Updated nvml.py to CUDA 11
- Updated smi.py DeviceQuery to R460
- Aligned nvml.py with latest nvidia-ml-py deployment
- Version 11.4.0
- Updated nvml.py to CUDA 11.4
- Updated smi.py NVML_BRAND_NAMES
- Aligned nvml.py with latest nvidia-ml-py deployment (11.495.46)
- Version 11.4.1
- Fix comma bugs in nvml.py
- Version 11.5.0
- Updated nvml.py to support CUDA 11.5 and CUDA 12
- Aligned with latest nvidia-ml-py deployment (11.525.84)
- Version 11.5.4 CLORE
- removed versioneer
- fixed nvmlDeviceGetGpcClkMinMaxVfOffset, nvmlDeviceGetMemClkMinMaxVfOffset
Raw data
{
"_id": null,
"home_page": "http://www.nvidia.com/",
"name": "clore-pynvml",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.6",
"maintainer_email": null,
"keywords": null,
"author": "NVIDIA Corporation",
"author_email": "rzamora@nvidia.com",
"download_url": "https://files.pythonhosted.org/packages/bc/62/e570918e15ea9cce3110210ea3ed8f0f54b0be8f9d59d8a35b5715ed140b/clore_pynvml-11.5.4.tar.gz",
"platform": null,
"description": "Python bindings to the NVIDIA Management Library\n================================================\n\nProvides a Python interface to GPU management and monitoring functions.\n\nThis is a wrapper around the NVML library.\nFor information about the NVML library, see the NVML developer page\nhttp://developer.nvidia.com/nvidia-management-library-nvml\n\nAs of version 11.0.0, the NVML-wrappers used in pynvml are identical\nto those published through [nvidia-ml-py](https://pypi.org/project/nvidia-ml-py/).\n\nNote that this file can be run with 'python -m doctest -v README.txt'\nalthough the results are system dependent\n\nRequires\n--------\nPython 3, or an earlier version with the ctypes module.\n\nInstallation\n------------\n\n pip install .\n\nUsage\n-----\n\nYou can use the lower level nvml bindings\n\n```python\n>>> from clore_pynvml import *\n>>> nvmlInit()\n>>> print(\"Driver Version:\", nvmlSystemGetDriverVersion())\nDriver Version: 410.00\n>>> deviceCount = nvmlDeviceGetCount()\n>>> for i in range(deviceCount):\n... handle = nvmlDeviceGetHandleByIndex(i)\n... print(\"Device\", i, \":\", nvmlDeviceGetName(handle))\n...\nDevice 0 : Tesla V100\n\n>>> nvmlShutdown()\n```\n\nOr the higher level nvidia_smi API\n\n```python\nfrom clore_pynvml.smi import nvidia_smi\nnvsmi = nvidia_smi.getInstance()\nnvsmi.DeviceQuery('memory.free, memory.total')\n```\n\n```python\nfrom clore_pynvml.smi import nvidia_smi\nnvsmi = nvidia_smi.getInstance()\nprint(nvsmi.DeviceQuery('--help-query-gpu'), end='\\n')\n```\n\nFunctions\n---------\nPython methods wrap NVML functions, implemented in a C shared library.\nEach function's use is the same with the following exceptions:\n\n- Instead of returning error codes, failing error codes are raised as\n Python exceptions.\n\n ```python\n >>> try:\n ... nvmlDeviceGetCount()\n ... except NVMLError as error:\n ... print(error)\n ...\n Uninitialized\n ```\n\n- C function output parameters are returned from the corresponding\n Python function left to right.\n\n ```c\n nvmlReturn_t nvmlDeviceGetEccMode(nvmlDevice_t device,\n nvmlEnableState_t *current,\n nvmlEnableState_t *pending);\n ```\n\n ```python\n >>> nvmlInit()\n >>> handle = nvmlDeviceGetHandleByIndex(0)\n >>> (current, pending) = nvmlDeviceGetEccMode(handle)\n ```\n\n- C structs are converted into Python classes.\n\n ```c\n nvmlReturn_t DECLDIR nvmlDeviceGetMemoryInfo(nvmlDevice_t device,\n nvmlMemory_t *memory);\n typedef struct nvmlMemory_st {\n unsigned long long total;\n unsigned long long free;\n unsigned long long used;\n } nvmlMemory_t;\n ```\n\n ```python\n >>> info = nvmlDeviceGetMemoryInfo(handle)\n >>> print \"Total memory:\", info.total\n Total memory: 5636292608\n >>> print \"Free memory:\", info.free\n Free memory: 5578420224\n >>> print \"Used memory:\", info.used\n Used memory: 57872384\n ```\n\n- Python handles string buffer creation.\n\n ```c\n nvmlReturn_t nvmlSystemGetDriverVersion(char* version,\n unsigned int length);\n ```\n\n ```python\n >>> version = nvmlSystemGetDriverVersion();\n >>> nvmlShutdown()\n ```\n\nFor usage information see the NVML documentation.\n\nVariables\n---------\n\nAll meaningful NVML constants and enums are exposed in Python.\n\nThe NVML_VALUE_NOT_AVAILABLE constant is not used. Instead None is mapped to the field.\n\nNVML Permissions\n----------------\n\nMany of the `pynvml` wrappers assume that the underlying NVIDIA Management Library (NVML) API can be used without admin/root privileges. However, it is certainly possible for the system permissions to prevent pynvml from querying GPU performance counters. For example:\n\n```\n$ nvidia-smi nvlink -g 0\nGPU 0: Tesla V100-SXM2-32GB (UUID: GPU-96ab329d-7a1f-73a8-a9b7-18b4b2855f92)\nNVML: Unable to get the NvLink link utilization counter control for link 0: Insufficient Permissions\n```\n\nA simple way to check the permissions status is to look for `RmProfilingAdminOnly` in the driver `params` file (Note that `RmProfilingAdminOnly == 1` means that admin/sudo access is required):\n\n```\n$ cat /proc/driver/nvidia/params | grep RmProfilingAdminOnly\nRmProfilingAdminOnly: 1\n```\n\nFor more information on setting/unsetting the relevant admin privileges, see [these notes](https://developer.nvidia.com/nvidia-development-tools-solutions-ERR_NVGPUCTRPERM-permission-issue-performance-counters) on resolving `ERR_NVGPUCTRPERM` errors.\n\n\nRelease Notes\n-------------\n\n- Version 2.285.0\n - Added new functions for NVML 2.285. See NVML documentation for more information.\n - Ported to support Python 3.0 and Python 2.0 syntax.\n - Added nvidia_smi.py tool as a sample app.\n- Version 3.295.0\n - Added new functions for NVML 3.295. See NVML documentation for more information.\n - Updated nvidia_smi.py tool\n - Includes additional error handling\n- Version 4.304.0\n - Added new functions for NVML 4.304. See NVML documentation for more information.\n - Updated nvidia_smi.py tool\n- Version 4.304.3\n - Fixing nvmlUnitGetDeviceCount bug\n- Version 5.319.0\n - Added new functions for NVML 5.319. See NVML documentation for more information.\n- Version 6.340.0\n - Added new functions for NVML 6.340. See NVML documentation for more information.\n- Version 7.346.0\n - Added new functions for NVML 7.346. See NVML documentation for more information.\n- Version 7.352.0\n - Added new functions for NVML 7.352. See NVML documentation for more information.\n- Version 8.0.0\n - Refactor code to a nvidia_smi singleton class\n - Added DeviceQuery that returns a dictionary of (name, value).\n - Added filter parameters on DeviceQuery to match query api in nvidia-smi\n - Added filter parameters on XmlDeviceQuery to match query api in nvidia-smi\n - Added integer enumeration for filter strings to reduce overhead for performance monitoring.\n - Added loop(filter) method with async and callback support\n- Version 8.0.1\n - Restructuring directories into two packages (pynvml and nvidia_smi)\n - Adding initial tests for both packages\n - Some name-convention cleanup in pynvml\n- Version 8.0.2\n - Added NVLink function wrappers for pynvml module\n- Version 8.0.3\n - Added versioneer\n - Fixed nvmlDeviceGetNvLinkUtilizationCounter bug\n- Version 8.0.4\n - Added nvmlDeviceGetTotalEnergyConsumption\n - Added notes about NVML permissions\n - Fixed version-check testing\n- Version 11.0.0\n - Updated nvml.py to CUDA 11\n - Updated smi.py DeviceQuery to R460\n - Aligned nvml.py with latest nvidia-ml-py deployment\n- Version 11.4.0\n - Updated nvml.py to CUDA 11.4\n - Updated smi.py NVML_BRAND_NAMES\n - Aligned nvml.py with latest nvidia-ml-py deployment (11.495.46)\n- Version 11.4.1\n - Fix comma bugs in nvml.py\n- Version 11.5.0\n - Updated nvml.py to support CUDA 11.5 and CUDA 12\n - Aligned with latest nvidia-ml-py deployment (11.525.84)\n- Version 11.5.4 CLORE\n - removed versioneer\n - fixed nvmlDeviceGetGpcClkMinMaxVfOffset, nvmlDeviceGetMemClkMinMaxVfOffset\n",
"bugtrack_url": null,
"license": "BSD",
"summary": "Python Bindings for the NVIDIA Management Library",
"version": "11.5.4",
"project_urls": {
"Homepage": "http://www.nvidia.com/"
},
"split_keywords": [],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "3a24305d851e2c2d7f8c966838640273f60532fe9e1238f5a55b6e7cb82555cc",
"md5": "2e4c76fa6d87a192d584e98bddc4e2d7",
"sha256": "8177f257be14d99ccbbc503d2d90889c2c81bcd1b448cba480713027a6b53609"
},
"downloads": -1,
"filename": "clore_pynvml-11.5.4-py3-none-any.whl",
"has_sig": false,
"md5_digest": "2e4c76fa6d87a192d584e98bddc4e2d7",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.6",
"size": 54231,
"upload_time": "2024-05-23T13:26:26",
"upload_time_iso_8601": "2024-05-23T13:26:26.397022Z",
"url": "https://files.pythonhosted.org/packages/3a/24/305d851e2c2d7f8c966838640273f60532fe9e1238f5a55b6e7cb82555cc/clore_pynvml-11.5.4-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "bc62e570918e15ea9cce3110210ea3ed8f0f54b0be8f9d59d8a35b5715ed140b",
"md5": "07ddf93b4af1119f066010e83e23e963",
"sha256": "c8ebda89adb7d36c9897270e8c1dc067a52d3b48490e4a396d1859362ae60bf7"
},
"downloads": -1,
"filename": "clore_pynvml-11.5.4.tar.gz",
"has_sig": false,
"md5_digest": "07ddf93b4af1119f066010e83e23e963",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.6",
"size": 55497,
"upload_time": "2024-05-23T13:26:28",
"upload_time_iso_8601": "2024-05-23T13:26:28.672078Z",
"url": "https://files.pythonhosted.org/packages/bc/62/e570918e15ea9cce3110210ea3ed8f0f54b0be8f9d59d8a35b5715ed140b/clore_pynvml-11.5.4.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-05-23 13:26:28",
"github": false,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"lcname": "clore-pynvml"
}
JSON