I started working on Blender 3D rendering tests using Yafaray raytrace render tests. I saw that people were scoring around 3 minutes on an Intel i7 920 with the test scene that features a glass block, semi-glossy floor and two Suzanne the Blender Monkey solid models.
I gave it a whirl and: less than 3 seconds
Okay, but it looks like an early 80s pop music video, all burned out and nothing like the intended output. With that fixed, I ran it again: 27 minutes
WTF? How is it that slow? I looked at Windows Task Manager and saw that my Two CPUs and 4 threads were severely underutilized. So I quit for the night and played World of Warcraft with my wife. Suzanne the Monkey could wait.
Still, I couldn't help but wonder if my rack of Dual Xeon 2.4 nodes would still be woefully underpowered due to age at this rate. 27 minutes a frame divided by six nodes is 4.5 minutes and my workstation is using Dual 3.2GHz 553MHz FSB Xeons. Uh oh. Even slower.
But the test I ran is under Windows XP Pro 32-bit and that's insanely slow compared to Linux, based on the other benchmarks I was seeing.
Today I tried again. This time I understood things a bit better. It would seem the exporter is now required and I have to set the parameters from the right-hand menu that comes up when I select the exporter from the render menu. It's set for one thread. That explains a lot. Since I had disabled Hyperthreading to see if that was the issue, I was running with the two CPUs 1:1 style. This time: just under 15 minutes.
That's a lot better. Now to re-enable Hyperthreading and see what that gets me. Should net a 20% decrease in render time. Time to test for that. Reboot!
10.5 minutes under Windows XP Pro when Hyperthreading. That'll do.
Soon I shall see how Linux fares.
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment