GPU-utils rate: 0~50%

[hedaobaishui]

I have run this code by change some parameter。

1 setting channels num smaller.

context_channels =384//4
forecast_steps = 20
forenum = 20
input_channels = 3 #use my dataset

2 Hardware parameters

CPU 9cores Intel(R) Xeon(R) Silver 4210R CPU @ 2.40GHz
GPU　3090*1 memory 256GB
PyTorch 1.10.0
Python 3.8
Cuda 11.3

3 train parameter

    traindataset = MyDataset(datapath=opt.datapath,dataname= "./train/TestAname.npy",presize=opt.prenum,foresize=opt.forenum,inputchannel=opt.input_channels)        
    tran_loader = DataLoader(traindataset, batch_size=opt.batchsize, num_workers=9,pin_memory=True,shuffle=True,prefetch_factor=8,persistent_workers=True)

3 Issue

GPU-utils rate: 0~50%
when use torch.profiler.profile to analyze model
Output information is in the analyze.md.
analyze.md
Any good advice plz?

[JackKelly]

Is the bottleneck loading data into RAM?

[peterdudfield]

@all-contributors please add @hedaobaishui for question

[allcontributors]

@peterdudfield

I've put up a pull request to add @hedaobaishui! 🎉