[Feature]: Hope we can add a GPU for translation

[xiaofufu520]

Expect to happened

I noticed that when using it, CPU computation for translation has some delay, and I hope to be able to use my GPU for computation. I checked the dependencies of ctranslate2.dll—it doesn't reference cudart/nvrtc (key libraries of the CUDA runtime), which shows that the author used a CPU-only build when packaging. I hope this can be available in the next version, thanks.

Expected APP Version

next newest version

Expected Runtime Environment

Windows 11

Country/Region

Qatar(example)

Expected Output

Sharing features in images

![DESCRIPTION](LINK.png)

Code of Conduct

• I agree to follow this project's Code of Conduct

[misyaguziya]

Hi @xiaofufu520 — you're right that the standard (CPU) package ships a CPU-only CTranslate2, so ctranslate2.dll has no CUDA references there.

GPU translation is already available in the GPU edition: install the GPU/CUDA edition (choose it in the v3.5.0 installer), then in Settings → "Processing device for AI translation", select your GPU. The local translation model (M2M100 / NLLB via CTranslate2) then runs on CUDA.

Docs :

• https://misyaguziya.github.io/VRCT-Docs/docs/config-advanced/#general
• https://misyaguziya.github.io/VRCT-Docs/docs/config-translation#ctranslate2-model-and-settings

If you're on the GPU edition and translation still runs on CPU, reply with your GPU model and I'll investigate. Closing as already supported.