# CUDA Kernel with customised device

**URL:** <https://dev-discuss.pytorch.org/t/cuda-kernel-with-customised-device/2918>\
**Category:** Uncategorized\
**Created:** [April 17, 2025, 2:06am UTC](https://dev-discuss.pytorch.org/t/cuda-kernel-with-customised-device/2918 "2025-04-17T02:06:07Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Ishu2001](https://avatars.discourse-cdn.com/v4/letter/i/d2c977/32.png) [@Ishu2001](https://dev-discuss.pytorch.org/u/Ishu2001)\
**Post date:** [April 17, 2025, 2:06am UTC](https://dev-discuss.pytorch.org/t/cuda-kernel-with-customised-device/2918/1 "2025-04-17T02:06:07Z")

</div>

is there way to launch cuda kernels with the customised device, useprivate1… or to customise cuda and create new device on top of it, by o verloading some operations

---

<div class="post-metadata">

**Author:** ![artyom-beilis](https://yyz2.discourse-cdn.com/flex036/user_avatar/dev-discuss.pytorch.org/artyom-beilis/32/207_2.png) [@artyom-beilis](https://dev-discuss.pytorch.org/u/artyom-beilis)\
**Post date:** [April 17, 2025, 1:18pm UTC](https://dev-discuss.pytorch.org/t/cuda-kernel-with-customised-device/2918/2 "2025-04-17T13:18:27Z")

</div>

What devices? Generally there is a cuda alternative called OpenCL - and it supports various GPUs - currently it works as out of tree backend. It supports significant amount of torch operators but far from being full featured:

> **[GitHub - artyom-beilis/pytorch\_dlprim: DLPrimitives/OpenCL out of tree backend for...](https://github.com/artyom-beilis/pytorch_dlprim)**
>
> DLPrimitives/OpenCL out of tree backend for pytorch