Spec for Running Tensorflow Version on Multiple GPUs

capreolus-ir / capreolus

A toolkit for end-to-end neural ad hoc retrieval

https://capreolus.ai

Apache License 2.0

95 stars 32 forks source link

Spec for Running Tensorflow Version on Multiple GPUs #191

Closed Pourbahman closed 3 years ago

Pourbahman commented 3 years ago

Hi Andrew,

According to https://github.com/capreolus-ir/capreolus/blob/master/docs/reproduction/PARADE.md I need 48GB RAM to run your code on GPU.

I provided 5*Gigabyte GEFORCE GTX 1080 Ti TURBO (Pascal), 11,72GB to run your tensorflow version of your implementation on multiple GPUs.

Would you please tell me if I can run it by this spec?

Thanks in advance, Kind Regards, Zahra

andrewyates commented 3 years ago

Hi Zahra, we have only tested on single 48GB GPUs. There may be inefficiencies in how the model is split across multiple GPUs (which is handled by Tensorflow), so I don't know for sure whether 5x12GB will be enough memory.