Any new TensorFlow-DirectML benchmark for Radeon vs Geforce Tensorflow CUDA?

[orangpelupa]

i only managed to find this one from January, and it shows Radeon is still way slower for training

the problem is, on October or September, AMD released a new driver that boost tensorflow-directml performance up to 4x. But i cannot find any benchmark comparing TensorFlow-DirectML benchmark for Radeon vs Geforce Tensorflow CUDA with that new driver.

btw does tensorflow-directml means i can use anything coded for tensorflow to work with radeon GPU with zero fiddling?

EDIT:

it seems it does requires zero fiddling! (other than installing tensorflow-directml)

Once TensorFlow-DirectML is installed, it works seamlessly with existing model training scripts.

https://devblogs.microsoft.com/wind...g-models-on-any-gpu-with-tensorflow-directml/

 

[Rikimaru]

I've tried to use it but it only implements TensorFlow 1.15 and stuff i try is 2.x now.

 

[orangpelupa]

I've tried to use it but it only implements TensorFlow 1.15 and stuff i try is 2.x now.

Thanks for tha info. That's quite a bummer.