I am using Blender 2.77 and an R7 250 GPU on Windows.

Compute Device >> OpenCL (Oland).

OpenSubdiv >> (Tested with all).

Device >> GPU Compute.

Enter image description here


That is the normal rendering behavior of a GPU.

GPUs don't have multiple discrete cores in the sense that a CPU has; GPUs have many (hundreds maybe) small processing units that act together as as a highly parallelized process that forms a single rendering bucket.

Since on it's own each single unit is relatively 'stupid' in the sense that it can only understand simple basic instructions they can't each alone process a full image, but their strength lies in the fact that they act together as a single 'processing thread' that may end up being faster that the multiple cores of your CPU (depending on the performance of hardware involved of course).

EDIT: As Cegaton mentions bellow you would get multiple buckets if you had multiple graphics cards in your system without a pairing (SLI/Crossfire) solution.

Also as Gandalf says, for this reason GPUs tend to like larger bucket sizes while CPUs benefit from smaller ones, although due to the limited amount of VRAM usually present in graphics cards increasing too much may start having a negative impact on rendering speed.

| improve this answer | |
  • 3
    $\begingroup$ If you had more than one GPU you'd see one tile for each GPU being used. $\endgroup$ – user1853 Sep 17 '16 at 3:31
  • $\begingroup$ Thought this would happen. :) But it's actually happening for per tile. $\endgroup$ – Sourav Paul Sep 17 '16 at 3:38
  • 6
    $\begingroup$ Also, larger tile sizes are often faster for GPU rendering, while smaller tiles are usually faster for CPUs. See blender.stackexchange.com/q/44428/599 $\endgroup$ – gandalf3 Sep 17 '16 at 3:43
  • $\begingroup$ @gandalf3 thanks for the link. more time saved. :) $\endgroup$ – Sourav Paul Sep 17 '16 at 4:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.