Using AMD OpenCL, it needs to compile before rendering. However it uses 10% of the CPU to do that

(BigBlend) #1

Couldn’t it be multi-threaded? Instead of waiting 30s to 60s it can be done in 10s.

I’m just wonder why it hasn’t been done? Is it a technical impossibility?

(mathieu) #2

It has been done, just yet another patch that takes dust.

(BigBlend) #3

Actually. It is kinda good news. Now we know it is possible but maybe the implementation won’t be made until the other high priority stuff gets done. However it would be nice if blender foundation had it in their road map so we know we can expect it in the future.

At least they should have put it in the stable official release of Blender. As it won’t get the newer updates of 2.8.

Thanks bliblubli for the quick reply.

(jbakker) #4

The mentioned patch is in blender2.79-buildbot and blender 2.80. Together with many other Cycles OpenCL related patches.

(BigBlend) #5

Yeah. They have implemented a faster opencl compiling. It takes about 20 seconds now instead of 1 minute. Still slow but much better.

(iszotic) #6

For some reason, there’s still a part of the kernel that can only be compiled with one thread that takes 20sec

(doublebishop) #7

They have implemented a faster opencl compiling.

The person you responded to is the opencl dev for blender :stuck_out_tongue:

They are also working on this patch – which will precompile all the kernels for your computer, so it will act like NVIDIA cards.

(BigBlend) #8

I knew that… :rofl: Kinda embarrassing but how could I have known.

I hope jbakker had a good laugh :slight_smile: