Pytorch 에서 CUDA 호출이 비동기식이기 때문에
타이머를 시작 또는 중지 하기 전에 torch.cuda.synchronize() 를 통해 코드를 동기화 시켜주어야 한다.
start = torch.cuda.Event(enable_timing=True)
end = torch.cuda.Event(enable_timing=True)
start.record()
z = x + y
end.record()
# Waits for everything to finish running
torch.cuda.synchronize()
print(start.elapsed_time(end))
참고자료 1 : https://discuss.pytorch.org/t/best-way-to-measure-timing/39496
Best way to measure timing?
Hello, I’m looking for the best way to measure the timing of a process: time.perf_counter or time.process_time? I have seen in several topics that people use more perf_counter but process_time is process-wide (1). But in the docs, you can see that proces
discuss.pytorch.org
참고자료 2 : https://discuss.pytorch.org/t/how-to-measure-time-in-pytorch/26964/2
How to measure time in PyTorch
There are many things you can do CPU-only benchmarking: I’ve used timeit as well as profilers. CUDA is asynchronous so you will need some tools to measure time. CUDA events are good for this if you’re timing “add” on two cuda tensors, you should sa
discuss.pytorch.org
'AI Development > PyTorch' 카테고리의 다른 글
| [PyTorch] 모델 파라미터 일부만 업데이트 하기 (0) | 2023.04.06 |
|---|---|
| [PyTorch] PyTorch 모델을 저장하는 방법 및 고려해야할 점 (2) | 2022.03.23 |
| [PyTorch] 파이토치에서 TensorBoard 사용하기 (3) | 2022.02.04 |
| [Pytorch] 이미지 데이터세트에 대한 평균(mean)과 표준편차(std) 구하기 (0) | 2021.05.11 |
| [Pytorch] 장치간 모델 불러오기 (GPU / CPU) (1) | 2020.01.09 |