# NVIDIA Tensor Cores Usage

**URL:** <https://community.konduit.ai/t/nvidia-tensor-cores-usage/3072>\
**Category:** Tuning Help\
**Created:** [February 13, 2024, 8:57am UTC](https://community.konduit.ai/t/nvidia-tensor-cores-usage/3072 "2024-02-13T08:57:54Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![seger7024](https://avatars.discourse-cdn.com/v4/letter/s/a88e57/32.png) [@seger7024](https://community.konduit.ai/u/seger7024)\
**Post date:** [February 13, 2024, 8:57am UTC](https://community.konduit.ai/t/nvidia-tensor-cores-usage/3072/1 "2024-02-13T08:57:54Z")

</div>

Hi!

Thanks for the great project. I have used DL4J in the past and I am about to begin again a small project with it. I hope its development will continue.

I have a rather newbie question but I have not come across an answer here yet. It is about the usage of the nvidia tensor cores of the RTX20/30/40 GPUs. On this page ([https://developer.nvidia.com/blog/optimizing-gpu-performance-tensor-cores/](https://developer.nvidia.com/blog/optimizing-gpu-performance-tensor-cores/)) it is said that mixed precision must be used and there are some descriptions for other frameworks on how to activate this. I can set to half precision in DL4J. Would that make the training use the cores or how does it work in DL4J if it does?

Thanks,  
Stefan