Implementation time and device?

[108618035guotingliao]

Hi,

Thanks you for your sharing~
I've implemented it for a while. I have some questions about the time because I spend lots of time on reaching 500K steps. However, the original authors pointed out that they only spent 3 hrs per 100K steps with Nvidia V100. And I spent about 2 days per 100K steps with RTX 3080. I doubt if my GPU ability was't be properly installed. My GPU's memory is full. But power only reach about 200 Watt, 340 Watt in total.

So, I want to check that how much time you spent and which device you trained? The following is my device.

• GPU: RTX 3080
• CPU: Intel i7-10700
• CUDA: v11.2
• Cudnn: v8.0.5
• Pytorch: v1.8.0
• cudnn_benchmark: True

Thanks you!!!