I have a gtx 1080 and ryzen 2700x. The gpu is a little faster when testing render times. I would like to balance render load between the two because the system is rendering mostly on the CPU when I hit F12, 100% CPU and about 20% GPU utilization. I know Cycles utilizes both processing that's why I am asking. I am 2 weeks into Blender coming from C4D so I am a Blender novice with good 3D knowledge.

I am basically testing materials and lighting using a smooth shaded Suzanne. On this scene I have a Sub division modifier divided 2x and quality set to 2. Solidifier modifier with thickness set to 0.02m. Glass Bsdf IOR set to 1.5. I have an hdri hooked to the colour of the world shader/light

I am using Blender version: 2.80 (sub 72) build date: 27/05/2019, 18:09 and 4.5.0 NVIDIA 388.13 drivers on Windows 10 1903.

Any help is highly appreciated.

  • $\begingroup$ Hi, dkeyboo. Your question is unclear. What are you talking about? What are you rendering and how? You haven't even mentioned what render engine or Blender version you are using. Please use the edit link below your question to edit it and add all the necessary relevant detail. $\endgroup$ Jun 1, 2019 at 19:19
  • $\begingroup$ @MartinZ They did tag the question cycles. That aside, how are you rendering with both CPU and GPU at once? What are your render device settings? $\endgroup$
    – gandalf3
    Jun 1, 2019 at 19:30
  • $\begingroup$ Thanks @MartinZ thanks for your response. I have edited the question. Hopefully it is clear now. $\endgroup$
    – dkeyboo
    Jun 2, 2019 at 1:36
  • $\begingroup$ @gandalf3 Under System Preferences, Cycles Render Devices, under CUDA tick both CPU and GPU. I think that's the trick;) $\endgroup$
    – dkeyboo
    Jun 2, 2019 at 1:40

1 Answer 1


I think I found the answer. Under Cycles, Sampling, Integrator choose "Branched Path Tracing" instead of the default "Path Tracing". I don't know how I missed this the first time.

For the scenes I have tested, GPU utilization is on average 60% and going as high as 100% at some point in some scenes.

  • $\begingroup$ The Branched Path Tracing is another rendering way that blender can use. But it is not directly relative to your original problem. The problem seems to be the rendering tile or task is to heavy to CPU and too easy for GPU. Changing to the Branched Path Tracing might change the complexity of rendering process. You might want to check other problem for a more specified solution $\endgroup$
    – HikariTW
    Jun 2, 2019 at 1:48
  • $\begingroup$ You are absolutely right @Hikariztw. Choosing the "Branched Path Tracing" is not the button that makes cycles share/balance workload between CPU and GPU, it only changes the way cycles traces light rays between the camera, light sources and objects in the scene. "Branched Path Tracing" is more complex than "Path Tracing" and my guess is that complexity presents more calculations to be solved some of which are off loaded to the GPU, the CPU also gets more calculations to solve. $\endgroup$
    – dkeyboo
    Jun 2, 2019 at 6:19
  • $\begingroup$ You might change the rendering tiles size to manually balance the multi-process that cycles use to rendering with multi-thread. While CPU like a small size tile, GPU often need a large size tile to utilize the power of it. See this: blender.stackexchange.com/questions/141271/… $\endgroup$
    – HikariTW
    Jun 2, 2019 at 6:32
  • 2
    $\begingroup$ I have tried large and small render tiles, Blender defaults to 64*64. The fastest registered time on mine is 32*32. It doesn't quite seem like large tiles utilise more GPU, I find that if more tiles concentrate in a local region that solves slowly, GPU usage is high. In fact when using 512*512 tiles, both CPU and GPU usage starts high but quickly drops to 20% CPU and about 2% GPU for a total of 7mins and 48secs. The same scene took 2mins and 33 secs with 32*32 tiles. GPU usage is high when solving complex calculation. Each tile utilises a single CPU thread hence low CPU usage with big tiles $\endgroup$
    – dkeyboo
    Jun 2, 2019 at 8:27
  • $\begingroup$ Tile size importance is different in 2.80. It may no longer be true, that GPUs require bigger tiles for faster rendering in all cases. $\endgroup$ Jun 2, 2019 at 8:52

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .