Rtx 2080 benchmarks

Hi,
I bought a rtx 2080 but can’t get better results with respect to gtx 1080. Can anybody share terminal output of pytorch official examples with alexnet and vggnet on default setup for the first few iterations. I want to compare with your results.
My computer ryzen 1700x, rtx 2080, 32 gb 2400 mhz ram, and ssd.
python 3.7, pytorch 1.0
Thx in advance.

Mines are as follows
=> creating model ‘alexnet’
Epoch: [0][ 0/5005] Time 3.243 ( 3.243) Data 1.928 ( 1.928) Loss 6.9077e+00 (6.9077e+00) Acc@1 0.00 ( 0.00) Acc@5 0.78 ( 0.78)
Epoch: [0][ 10/5005] Time 0.158 ( 0.528) Data 0.114 ( 0.348) Loss 6.9093e+00 (6.9078e+00) Acc@1 0.00 ( 0.11) Acc@5 0.00 ( 0.50)
Epoch: [0][ 20/5005] Time 1.449 ( 0.529) Data 1.403 ( 0.402) Loss 6.9091e+00 (6.9076e+00) Acc@1 0.00 ( 0.07) Acc@5 0.39 ( 0.52)
Epoch: [0][ 30/5005] Time 0.189 ( 0.487) Data 0.144 ( 0.379) Loss 6.9086e+00 (6.9076e+00) Acc@1 0.00 ( 0.06) Acc@5 0.00 ( 0.49)
Epoch: [0][ 40/5005] Time 1.364 ( 0.489) Data 1.319 ( 0.391) Loss 6.9061e+00 (6.9077e+00) Acc@1 0.00 ( 0.10) Acc@5 1.17 ( 0.56)
Epoch: [0][ 50/5005] Time 0.092 ( 0.466) Data 0.030 ( 0.373) Loss 6.9096e+00 (6.9075e+00) Acc@1 0.00 ( 0.11) Acc@5 0.00 ( 0.54)
Epoch: [0][ 60/5005] Time 1.571 ( 0.483) Data 1.527 ( 0.393) Loss 6.9081e+00 (6.9076e+00) Acc@1 0.39 ( 0.10) Acc@5 0.39 ( 0.53)
Epoch: [0][ 70/5005] Time 0.092 ( 0.470) Data 0.000 ( 0.382) Loss 6.9067e+00 (6.9074e+00) Acc@1 0.00 ( 0.10) Acc@5 0.39 ( 0.57)
Epoch: [0][ 80/5005] Time 1.389 ( 0.478) Data 1.343 ( 0.391) Loss 6.9038e+00 (6.9075e+00) Acc@1 0.00 ( 0.10) Acc@5 0.78 ( 0.56)
Epoch: [0][ 90/5005] Time 0.098 ( 0.468) Data 0.000 ( 0.383) Loss 6.9005e+00 (6.9073e+00) Acc@1 0.39 ( 0.09) Acc@5 0.39 ( 0.54)
Epoch: [0][ 100/5005] Time 1.509 ( 0.473) Data 1.464 ( 0.389) Loss 6.9101e+00 (6.9069e+00) Acc@1 0.39 ( 0.09) Acc@5 0.39 ( 0.54)
Epoch: [0][ 110/5005] Time 0.093 ( 0.465) Data 0.000 ( 0.382) Loss 6.9029e+00 (6.9068e+00) Acc@1 0.39 ( 0.10) Acc@5 0.78 ( 0.56)

You are bottlenecked by data loading time. The average time spent loading data in your log is 382 ms out of 465 ms. Make sure your data is on a local SSD and try increasing the number of data loading workers (-j in the ImageNet example code)

I did benchmark in https://github.com/u39kun/deep-learning-benchmark and results are below. Is there a way other than to increase number of workers in the data loader?

rtx 2080

Framework Precision vgg16 eval vgg16 train resnet152 eval resnet152 train densenet161 eval densenet161 train
pytorch 32-bit 39.3ms 127.5ms 62.3ms 211.8ms 65.8ms 206.8ms
pytorch 16-bit 27.8ms 98.4ms 38.0ms 144.2ms 43.7ms 154.6ms

gtx 1080

Framework Precision vgg16 eval vgg16 train resnet152 eval resnet152 train densenet161 eval densenet161 train
pytorch 32-bit 55.2ms 171.0ms 83.6ms 279.6ms 86.3ms 289.2ms
pytorch 16-bit 47.2ms 152.7ms 65.0ms 235.4ms 66.3ms 237.7ms

You could also set pin_memory=True