Question about low GPU utilisation

[hakunin]

Hi,
I'm using gym_trading to learn about RL for algotrading.

Compared to some simple models I did with RNNs I noticed the learning takes significantly longer.
So I switched to running it on my 2080 ti. Runs a bit faster than on CPU.
But then I notice the total GPU utilization is only about 7-10% and that's on a desktop machine.

Is there a way I can speed up the learning and use more GPU power?

[hakunin]

If I make the neural net really big, it gets to a higher utilisation (~30%)

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #
=================================================================
 Dense_1 (Dense)             (None, 1024)              11264

 Dense_2 (Dense)             (None, 1024)              1049600

 dropout (Dropout)           (None, 1024)              0

 Output (Dense)              (None, 3)                 3075

=================================================================
Total params: 1,063,939
Trainable params: 1,063,939
Non-trainable params: 0
_________________________________________________________________

When it says the model is sequential, it basically means it can't get any faster even if the network is small due to the sheer amount of iterations and no paralelism?