i have some mini shuttles with i5-12500T with UHD 770. I can run my patch in skia or as skiatexture in stride. Skia runs with 60fps, but when using skia in stride I get 18fps, while taskmanager shows 88% GPU usage.
I was wondering if there is a bottleneck somewhere, of if it is hardware related. maybe the GPU sharing ram with the CPU? I am also not sure if some (deactivated) stride nodes maybe hog some resources.
the stride pipeline is pretty empty atm: I display the skiatexture on a quadrender entity after scene, and in-scene there is just an orbitcamera and a background. some lights (directional, skybox, ambient) are connected but deactivated. PostFX sits in an deactivated if region but is connected to the graph.
yellow is the load without teamviewer. looks like the machine could do a little bit better
Try SkiaRenderer instead of SkiaTexture… Otherwise, the shuffle between OpenGL/Angle and DirectX will definitely cost something. It is possible that the Intel GPU driver has no idea what to do and uses a non-optimized way to solve the issue.
The Skia to Stride performance issue was also the reason why @bjoern recently reworked a whole UI system from Skia to work in vvvv beta style with pre-rendered textures from Photoshop and Stride rendering nodes. The performance gain was huge.
If Stride is not the main part of your app, you can use TextureToImage to display the Stride texture content in Skia with some frames delay. Depending on what you do and how big the resolutions are, that might work well.
Thank you, looks definitely faster using skiarenderer! would like to stay with stride as main renderer, maybe migrate the text rendering to it. I saw there was a new stride text node.
I keep carrying this skia vram hack with me, is this still necessary? I remember the vram limit was increased at some point, and I also don’t use 4GB of skia textures. is it possible this steals vram from stride?
Actually most of the textures were created with (cached) TextureFX.
But yes it was a lot faster.
Which internally uses Skia…
I think it was raised to 512 MB initially but then lowered again because of some issue. @robotanton ?