# Backward pass only uses ~50% GPU utilization?

**URL:** <https://discuss.pytorch.org/t/backward-pass-only-uses-50-gpu-utilization/97020>\
**Category:** Uncategorized\
**Created:** [September 21, 2020, 4:04pm UTC](https://discuss.pytorch.org/t/backward-pass-only-uses-50-gpu-utilization/97020 "2020-09-21T16:04:11Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![gebbissimo](https://discuss.pytorch.org/letter_avatar_proxy/v4/letter/g/e0b2c6/32.png) [@gebbissimo](https://discuss.pytorch.org/u/gebbissimo)\
**Post date:** [September 21, 2020, 4:04pm UTC](https://discuss.pytorch.org/t/backward-pass-only-uses-50-gpu-utilization/97020/1 "2020-09-21T16:04:11Z")

</div>

Hello there!

The graph below visualizes my GPU utilization per timestamp as a progress bar. It is mostly 99%, but has some drops in between. Interestingly, those drops correlate consistently to the times of the “loss.backward()” call. Any idea why the backward pass might not use 100% GPU or how to improve it?

If helpful: The final loss is an L1-loss between two RGB images. One of those images is created using F.grid\_sample() and some pinhole camera backprojection and projection functions.

Thanks in advance!

 ![image](https://discuss.pytorch.org/uploads/default/original/3X/2/7/27c42e5bba003a27ab7d365f19fce5ee39c53a9b.png)