Hello! Developer here.
Some of you asked whether OCR could run on Intel NPUs or Intel Arc / Iris Xe GPUs. We built it with OpenVINO, and it works on our test rig, but we do not own an Intel NPU machine, so we cannot measure real-world speed ourselves. If you have an Intel Core Ultra (NPU) or an Intel Arc / Iris Xe GPU and a few minutes to spare, we would appreciate your help. No obligation at all: if nobody has time, that is fine, and we will decide what to do with the feature by around August 31.
What is in the beta build
- Settings > Translation > Options shows an "OCR accelerator" row (Auto / Intel NPU / Intel GPU) when an Intel NPU or Intel GPU is detected. On non-Intel PCs the row is hidden and nothing changes.
- Selecting Intel NPU or Intel GPU moves text recognition to that device. The first start compiles the models once (about 40 to 60 seconds; the control bar shows "Optimizing for Intel (first time only)" and translation keeps working meanwhile). Later starts use a cache and start immediately.
- "OCR device" (added in v1.28) then reads "NPU (OpenVINO)" or "Intel GPU (OpenVINO)", so you can see it really switched.
- The default remains DirectML; nothing changes unless you pick an accelerator. The build is about 220 MB larger because the OpenVINO runtime is bundled.
How to join
Right-click the app in your Steam library > Properties > Betas > choose "beta" (no password). To leave, choose "None" and Steam rolls you back to the normal version.
What we would love to know
Your CPU/GPU model and driver version; whether the "OCR accelerator" row appeared; whether the first-time optimization finished and how long it took; whether "OCR device" changed; whether translation felt faster, slower or the same compared with Auto; any crash or fallback. Even one or two lines help. Please post in this thread, where a copy-paste checklist is also available:
Beta discussion thread and checklist
Thank you very much!