# multi-GPU training

**URL:** <https://discourse.scverse.org/t/multi-gpu-training/199>\
**Category:** scvi-tools\
**Tags:** gpu\
**Created:** [October 13, 2021, 4:42am UTC](https://discourse.scverse.org/t/multi-gpu-training/199 "2021-10-13T04:42:25Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![thisisyichenzhang](https://yyz1.discourse-cdn.com/flex035/user_avatar/discourse.scverse.org/thisisyichenzhang/32/40_2.png) [@thisisyichenzhang](https://discourse.scverse.org/u/thisisyichenzhang)\
**Post date:** [October 13, 2021, 4:42am UTC](https://discourse.scverse.org/t/multi-gpu-training/199/1 "2021-10-13T04:42:25Z")

</div>

- Does any of the provided method support multi-GPU training？
- I am trying to develop my custom model with your fantastic tools. I noticed that in **scvi.trainer** , it has an option to set **gpus**  
, how should I use this feature to build a model supporting multi-GPU training?

---

<div class="post-metadata">

**Author:** ![adamgayoso](https://yyz1.discourse-cdn.com/flex035/user_avatar/discourse.scverse.org/adamgayoso/32/100_2.png) [@adamgayoso](https://discourse.scverse.org/u/adamgayoso)\
**Post date:** [October 13, 2021, 3:46pm UTC](https://discourse.scverse.org/t/multi-gpu-training/199/2 "2021-10-13T15:46:48Z")

</div>

So ideally you’d just set `gpus>1` but our data loader does not currently work for multi-gpu training. It’s something we are currently looking into.
