Closed realkibou closed 1 year ago
Can you provide more information about your system, your SoC/CPU/GPU, please? For OpenVINO to be able to detect and use your GPU certain modules - like OpenCL - need to be installed.
A simple check is to install OpenVINO and run the tool hello_query_device
(from OpenVINO and/or Open-Model-Zoo).
If your system has multiple GPUs (like internal/embedded GPU(s) and discrete GPU(s)) then each supported PGU gets a different identifier, like GPU.0
and GPU.1
, like described under "https://docs.openvino.ai/latest/openvino_docs_OV_UG_supported_plugins_GPU.html".
11th Gen Intel(R) Core(TM) i7-1165G7 @ 2.80GHz 2.70 GHz
OpenVINO is already installed. With the CPU I can render images, just no GPU support.
There is only 1 GPU. The command I tried was python demo.py --device "GPU" --prompt "Street-art painting of Emilia Clarke in style of Banksy, photorealism" and python demo.py --device GPU --prompt "Street-art painting of Emilia Clarke in style of Banksy, photorealism" and python demo.py --device "GPU.0" --prompt "Street-art painting of Emilia Clarke in style of Banksy, photorealism" and None of them works.
EDIT: This command did it: python demo.py --device GPU.0 --prompt "Street-art painting of Emilia Clarke in style of Banksy, photorealism" its at 5.32s/it now. Before was 7-8. Thank you!
That is how it is supposed to work. I don't have a TigerLake at hand - but this is how I can run it on e.g. AlderLake's GPU I have available right now.
Are you under MS-Win or Linux?
Do you have an OpenVINO installation and could do a quick test with hello_query_device
to check if OpenVINO is able to detect the GPU and initialize the GPU-plugin successfully?
I can remember there was something specific for TigerLake, e.g. see here: "https://hub.docker.com/r/openvino/ubuntu20_dev_no_samples"; pre-built Docker containers were made available with the postfix _tgl
, which sounds like those containers had special drivers or patches included...
Hi,
I used the --device GPU but it crashes without any info. It just stops and returns back to the command input. Did anyone managed to let the diffusion run on the GPU?
Greetings