Hi,
I am attempting to render an animation. The local node behaves normally, but remote nodes reload their render kernels every frame (see attached)
-- which sort of defeats the point of a KERNEL, but whatever. Anyway, this really slows down the render speed, since ~95% of the time spent per frame is used building kernels instead of rendering the scene.
This occurs on my Windows 7 x64 node running x2 4GB RX550s and Blender 2.79b (current stable version as of 06/16/18.)
Am I missing something? How can I fix this?
Hi Ryan,
Thanks for the update :)
v015 should be released in about a week from now and should be a lot better in many regards. Resyncing should hopefully have been improved as well.
Thank you for the fast response James.
I should specify that my kernel build times aren't unusually long. This scene is really basic; each frame takes maybe a second to render without the overhead of rebuilding the kernel. With a sufficiently complex scene, kernel rebuild time would become negligible relative to render time.
I am looking forward to progress on this issue. My only other complaint is that nodes are sometimes finicky to resync, but even still CrowdRender is the best solution to distributed rendering with Blender that I've found so far. Keep it up!
Hi Ryan,
I'm not that familiar with the RX550, but on most of my tests using Nvidia this kernel building stage is fairly quick and doesn't take up so much of the overall render time.
As far as what I can see from your post, you're not missing anything and using CrowdRender correctly, it seems as though your long kernel build times might be specific to radeon cards -> https://blender.stackexchange.com/questions/97259/long-loading-render-kernels-on-opengl-with-amd-pro-duo
The reason the kernels are recompiled on each frame is that our addon doesn't yet have tight enough integration with cycles to load the kernels once and then just request parts of each frame to render. We're working on a solution to this so that we can do just that though so stay tuned :D. There will be improvements coming that will hopefully eliminate the need to re-compile the kernels.