How to check GPU vs CPU percentage?

Part 1 (2019)
1

Let’s say a learning step takes 5 minutes in Kaggle. During this time I can see GPU being used on and off, while the CPU is mostly on. IS there a way to check the stats which would indicate the effifiency of the GPU utilization, the percentage of time spent in CPU vs GPU in Kaggle?

2

I’m not sure how to get CPU utilization but in this notebook Jeremy uses the following function to view how much GPU memory is used (and to clear memory):

import gc
def report_gpu():
    print(torch.cuda.list_gpu_processes())
    gc.collect()
    torch.cuda.empty_cache()

image
image1348×346 15.8 KB

Also found this Kaggle notebook that runs a custom script in the notebook with subprocess.Popen. Never used such a script so can’t speak to its performance.

1 Like
3

Hi!

Here’s a CPU part:

import psutil

def monitor_ram_cpu():
    # Get process information
    processes = psutil.process_iter()

    # Display RAM and CPU usage for each process
    for process in processes:
        try:
            memory_info = process.memory_info()
            cpu_percent = process.cpu_percent()
            print(f"Process: {process.name()}")
            print(f"RAM usage: {memory_info.rss / 1024**3:.2f} GB")
            print(f"CPU usage: {cpu_percent:.2f}%")
            print("---")
        except psutil.NoSuchProcess:
            pass

monitor_ram_cpu()
2 Likes