Closed seiyajin closed 1 year ago
Nice suggestion. I'll be looking into it for the next update.
Currently I have a Linux machine with a integrated graphics and a Nvidia one, so I will be testing there.
I will be releasing rc builds when I manage to implement it, stay tuned.
Hi, during my launch today I started the implementation of multi gpu support, follow this PR to see the progress: https://github.com/OLoKo64/upscale-rs/pull/12
I didn't had time to test it yet but it should be already working, as the correct arguments are being correctly passed to the binary that does the upscaling.
The UI needs a bit of work still.
The documentation of the commands can be found in this link: https://github.com/xinntao/Real-ESRGAN#usage-of-portable-executable-files
thanks alot for replying my request, this is will be great update
On Sat, 25 Feb 2023, 00:00 Reinaldo Rozato Junior, @.***> wrote:
Hi, during my launch today I started the implementation of multi gpu support, follow this PR to see the progress: #12 https://github.com/OLoKo64/upscale-rs/pull/12
I didn't had time to test it yet but it should be already working, as the correct arguments are being correctly passed to the binary that does the upscaling.
The UI needs a bit of work still. [image: image] https://user-images.githubusercontent.com/49915167/221240655-c2ec3b39-0726-4cac-89cc-fe05c2d8eaf4.png
The documentation of the commands can be found in this link: https://github.com/xinntao/Real-ESRGAN#usage-of-portable-executable-files
— Reply to this email directly, view it on GitHub https://github.com/OLoKo64/upscale-rs/issues/13#issuecomment-1444027159, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJCKAB6V75VITFNVFGAB6WDWZDSLVANCNFSM6AAAAAAVDM3O3M . You are receiving this because you authored the thread.Message ID: @.***>
@seiyajin I just released the first cross plataform release candidate for the v0.2.2
which includes the advanced options for multi GPU as well as added support for webp
images.
Give it a spin and tell me what you think.
v0.2.2-rc1 link: https://github.com/OLoKo64/upscale-rs/releases/tag/v0.2.2-rc1
thanks for rc build update, seem work ok, for my old GPU (gtx660 + gtx750ti)
but i think for some reason multiGPU string not worked so well (case happen when i'm using cupscale too) i think this problem happen from realESRGAN script, not pytorch modul
Hey @seiyajin , thanks for testing so quickly.
So the selection of GPU is working but not the parallelization when using multi GPU.
I can think in 3 possibilities:
gpu-id: 0,1
set the tile-size: 0,0
and the load:proc:save: 1:2,1:2
as well. Can you try that? My notebook does not recognize the integrated graphics as a second GPU so I cant test it.ok, i will try with other parameter, like you said parallelization multi gpu seem complicated, i tried searching solution about this problem, but i can not find the right answer. by the way, can i make request, for next update, can you add option for gfgan model inclued(optional), for face restoration, thank you
On Sun, 26 Feb 2023, 22:38 Reinaldo Rozato Junior, @.***> wrote:
Hey, thanks for testing so quickly.
So the selection of GPU is working but not the parallelization when using multi GPU.
I can think in 3 possibilities:
- Maybe you need to pass the other parameters as well, for example, when using gpu-id: 0,1 set the tile-size: 0,0 and the load:proc:save: 1:2,1:2 as well. Can you try that? My notebook does not recognize the integrated graphics as a second GPU so I cant test it.
- The documentation was not clear to me if the multi GPU support works only when selecting a directory, not individual files, so the parallelization occurs that way. This is a bit of a problem as I designed the app to select files only, so you can select the ones you need, not folders. This would require a bit more work to get right.
- The last option is like you said, maybe is something wrong with the realESRGAN binary.
— Reply to this email directly, view it on GitHub https://github.com/OLoKo64/upscale-rs/issues/13#issuecomment-1445391273, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJCKAB75UABTKS4RM6P7F5DWZN2F3ANCNFSM6AAAAAAVDM3O3M . You are receiving this because you were mentioned.Message ID: @.***>
@seiyajin Some updates.
I tested some things to make the multi GPU work but sadly I was not successful, it really seems that it is made for when you select a folder, but as it would require a good chunk of work and it just split the work between the images, not really a multi GPU.
Two things make me not want to add the upscale per folder:
by the way, can i make request, for next update, can you add option for gfgan model inclued(optional), for face restoration
In this app I'm using the binary, not the Python script, and it still has the feature "Support face restore - GFPGAN" as a TODO.
For now im going to finish this update including the following:
.webp
images.In the future if I manage to find a way to make the multi GPU work I will update it, same with the face restoration, as soon as they update it i will as well.
can you add multi gpu support or gpu selector, for your next update, i think it will good performance and faster for upscaling image