Cycles hybrid render in blender

[Janettebot]

Hello, fixing my pc soon and I will be using it for blender however I have a question, will hybrid rendering be faster or slower than gpu only when an i3 10100 (equivalent to i7 7700) is paired with a gtx 1050ti utilising cuda acceleration. I know this might be kinda hard to gauge so I wont be disappointed if you cant answer. Thanks in advance. 

[DigitalGoat]

31 minutes ago, Janettebot said:

Hello, fixing my pc soon and I will be using it for blender however I have a question, will hybrid rendering be faster or slower than gpu only when an i3 10100 (equivalent to i7 7700) is paired with a gtx 1050ti utilising cuda acceleration. I know this might be kinda hard to gauge so I wont be disappointed if you cant answer. Thanks in advance. 

It could be a lot slower, the GPU will finish rendering it's tiles before the CPU so the process will be constantly waiting for the CPU to catch up.

The GPU can only render tiles assigned to it, it can't "help out" the CPU and vice versa. A good technique is to use only GPU or CPU, test to see which renders quicker (almost definitely the GPU with CUDA) then when setting up your render on GPU set the tile size to the output size divided by 4, so if output is 1920 x 1080, tile size will be 480 and 270, for CPU 16, 16 - 24, 24 should be good.

To speed up renders without sacrificing too much quality set render sampling lower and increase output size (sampling of 64 for 1920 is quite acceptable).

Obviously alter the settings to suit your exact setup, do lots of test renders with a control scene you set up till you are happy with the results, then save the project as a start up file to maintain those settings.

For quicker test renders check "fast GI approximation" or use Evee.

[Janettebot]

26 minutes ago, DigitalGoat said:

It could be a lot slower, the GPU will finish rendering it's tiles before the CPU so the process will be constantly waiting for the CPU to catch up.

The GPU can only render tiles assigned to it, it can't "help out" the CPU and vice versa. A good technique is to use only GPU or CPU, test to see which renders quicker (almost definitely the GPU with CUDA) then when setting up your render on GPU set the tile size to the output size divided by 4, so if output is 1920 x 1080, tile size will be 480 and 270, for CPU 16, 16 - 24, 24 should be good.

To speed up renders without sacrificing too much quality set render sampling lower and increase output size (sampling of 64 for 1920 is quite acceptable).

Obviously alter the settings to suit your exact setup, do lots of test renders with a control scene you set up till you are happy with the results, then save the project as a start up file to maintain those settings.

For quicker test renders check "fast GI approximation" or use Evee.

Thank you, currently asking because I am doing 8k renders on a celeron cpu only :/. I look forward to future renders with faster hardware. With 8k im doing 72 samples with branched path tracing and adaptive sampling.