Render very slow using GPU from Radeon Sapphire RX 580 vs CPU

Question

After installing Blender in Windows 10 and configuring the cycles compute device to use OpenCL with Radeon RX 580 Series, the render using cycles and GPU Compute takes much more time tan single CPU rendering...

CPU Rendering: 2.05 seconds

GPU Rendering: 63.44 seconds

The render is just so simple as the cube that appears when you open blender.

Graphic card is a Radeon Sapphire Nitro+ RX 580 8Gb GDDR5 and I updated the drivers to have the last ones.

Could you please help me with this issues?

Thanks in advance.

Answer 1

First of all the GPU needs to compile the Kernel first ( which takes more time in case of your cube than the actual render itself ) - this is done on the cpu.

Second Radeon need bigger tiles to have enough work to even get remotely saturated. 256x256 is a bare minimum and in case 2.79 or below is used you should even go Fullscreen ( 1 or 2 tile ) for best performance ( dont use progressive ! ) - each call to the Gpu halts it for a short time - the more of theese calls for example when progressive is used come together - like when eavery sample needs to be updated to the cpu the gpu spends most of the time waiting for data.

Take a more complex scene and try again.