Thank you for responding ptrblck.
I’ve tried using random tensor inputs and the gpu process is utilized 100%.
#testing if gpu process is working
a = torch.rand(20000,20000).cuda()
end = time.time()
while True:
print(time.time())
a += 1
a -= 1
end = time.time()
However, while profiling the data loading time, I ended up getting erratic elapsed time.
Results for the actual data loader:
