gpu_occupancy_monitor

Monitor GPU usage, occupy it when usage is below the threshold.

监视gpu使用率，低于阈值就占满，防止被kill

描述

占卡脚本，监控 GPU 利用率，并在 GPU 利用率低于指定阈值时计算，防止利用率不足被kill。

环境需求

• Python 3.x
• NVIDIA GPU 和驱动
• PyTorch
• nvidia-ml-py3 库（安装命令：pip install nvidia-ml-py3）

使用方法

1. 安装所需库：

   pip install torch nvidia-ml-py3

2. 运行脚本： nohup python3 fuck_gpu.py

---

Description

A script for occupying GPUs: Monitors GPU utilization and performs computations when GPU utilization is below a specified threshold to prevent being killed due to insufficient utilization.

Features

• Continuously monitors GPU usage using NVIDIA's NVML library.
• Automatically starts dummy matrix multiplication tasks on GPUs with low utilization.
• Stops dummy tasks when utilization rises above the specified threshold.
• Supports multi-GPU systems.
• Uses PyTorch for GPU computation tasks.

Requirements

• Python 3.x
• NVIDIA GPU and drivers
• PyTorch
• nvidia-ml-py3 library (install using pip install nvidia-ml-py3)

How It Works

1. The script initializes NVML to monitor GPU utilization.
2. It checks the utilization of each GPU at regular intervals.
3. If a GPU's utilization is below the threshold (default: 60%), it starts a dummy task that performs matrix multiplications on that GPU.
4. If the utilization exceeds the threshold, the dummy task is stopped.

Usage

1. Install the required libraries:

   pip install torch nvidia-ml-py3

2. Run the script: nohup python3 fuck_gpu.py