I run blender on an Imac with 16 gig. This is a horrible machine to try to do 3d modeling… Anyway, I was playing around with the performance section of the render properties panel. I rendered a very simple 1080p scene using Blender internal.
Using the default s in the performance section:
Auto Detect threads
tile size 64x64
The scene took 16.91 to render.
I played around with the tile size and eventual figured out that 256x256 was the optimal setting for this scene.
The resulting render time was 7.72 (Less than half the time!)
I then started playing with the threads and after several tries I came up with fixed 20 threads gave me the best times.
The resulting render time was 6.21
The moral of the story is that from now on, when doing one of those “all weekend render sessions” I will take the few minutes I need to change the settings and see what gives me the best performance. If I can cut my render time by more than 60% for a project, it is well worth the time!
I have had similar time savings on my PC running windows XP with 2 gig.
I came a to a similar conclusion when I started playing around with 2.65. 192 tiles is the optimum number on my machine for Blender Internal. This number is different when using Cycles, however. I had not tried playing with the thread count. I assumed you just would leave that at the number of cores in your machine. I’ll have to try that.
It’s for Blender Internal, on my tests very small tiles in Blender Internal get (much) longer rendering time for the whole scene overall, very different from Cycles apparently.
Yeah exactly. I’ve got an i7, so four cores, two threads per core with hyperthreading, so I set (or rather blender detects) it at 8 threads. But in your first post you noted that a setting of 20 threads improved performance on an Imac, which could be anything from a G3 to i7 cpu. How does that work? Unless I’m interpreting the meaning of threads in blender incorrectly.
There are probably a few hundred threads worth of processes running at this point in time on your computer… it just means that only 8 (if you have a quad core with hyperthreading) are executed at any one point in time… Setting it higher then 8, means that you are relying more on the internal thread management system (something that linux is especially good at)… You should only do this if you see cores becoming idle for too long whilst rendering. and even then, setting it to 16 should be a bit more efficient then setting it to 20… But, in all cases, this will vary from scene to scene.
I read a university peace recently that had said rendering at 32 produce the best realistic rendering results this was about the time tiles was into. Finally getting a gpu card and always working in cpu. I have to say rendering 32x32 feels, looks better faster. While 16x16 renders faster. Presently have a major glitch with the new card. I have to say I have been using the view port way to much. I have the mouse setup to hit the F12 button. God i hate the button.
