In my setup (AMD A4), using opencl with darktable is about twice as fast as using the dual-core cpu.
It's not working out of the box, though. Not even if you have installed the OpenCL SDK by AMD. Because my BIOS only allows me to set up to 512M as vram for my GPU/APU. But then I thought let's just try what happens and set the required memory in darktable down to 250M. And it worked. And it worked great. I get a great speed increase:
OpenCL (zoom to 200%): [dev_process_image] pixel pipeline processing took 3,574 secs (0,460 CPU)
GPU (zoom to 200%): [dev_process_image] pixel pipeline processing took 6,442 secs (11,640 CPU)
And as you can see, the CPU is hardly used anymore, so it could do something else at the same time!
Conclusion: It appears you can set the requirements lower. A 2x speed increase for standard activities is really worth it! See the next post for how to set up Darktable for AMD APUs.