Cycles is consistently slower in 2.74 #44337

New Issue

gandalf3 · 2015-04-10T08:32:53+02:00

gandalf3 commented

2015-04-10 08:32:53 +02:00

System Information
Archlinux 64

Blender Version
Broken: 2.74
Worked: 2.73

Short description of error
While rendering this test scene (originally made to test the new black background optimization in 2.74),
I found that cycles is consistently slower in 2.74. I ran a test scene five times each in both 2.73 and 2.74 (downloaded from blender.org).
All the 2.74 render times are slower than even the slowest 2.73 time.
The average render times from my tests:

2.73: 3:33.20
2.74: 3:41.67

I can't see any subjective difference in noise level (just by looking at it the renders anyway).
Renders were done from the CLI.

Exact steps for others to reproduce the error

Open black_bg_performance_test.blend
Render

**System Information** Archlinux 64 **Blender Version** Broken: 2.74 Worked: 2.73 **Short description of error** While rendering this test scene (originally made to test the new black background optimization in 2.74), I found that cycles is consistently slower in 2.74. I ran a test scene five times each in both 2.73 and 2.74 (downloaded from blender.org). All the 2.74 render times are slower than even the slowest 2.73 time. The average render times from my tests: 2.73: `3:33.20` 2.74: `3:41.67` I can't see any subjective difference in noise level (just by looking at it the renders anyway). Renders were done from the CLI. **Exact steps for others to reproduce the error** 1. Open [black_bg_performance_test.blend](https://archive.blender.org/developer/F160409/black_bg_performance_test.blend) 2. Render

gandalf3 commented

2015-04-10 08:32:53 +02:00

Changed status to: 'Open'

gandalf3 commented

2015-04-10 08:32:53 +02:00

Added subscriber: @gandalf3

Sergey Sharybin was assigned by Bastien Montagne

2015-04-10 12:03:12 +02:00

Bastien Montagne commented

2015-04-10 12:03:12 +02:00

Added subscriber: @mont29

Bastien Montagne commented

2015-04-10 12:03:12 +02:00

Sergey, I think to remember you have worked on that already?

Sergey Sharybin commented

2015-04-10 12:50:50 +02:00

I've solved some typos in the distance heuristic in QBVH and did some workarounds for MSVC stupidness.

It' unclear from the report if it's CPU or GPU rendering, plus attaching system-info.txt is always a good idea.

I've solved some typos in the distance heuristic in QBVH and did some workarounds for MSVC stupidness. It' unclear from the report if it's CPU or GPU rendering, plus attaching system-info.txt is always a good idea.

gandalf3 commented

2015-04-11 00:16:30 +02:00

CPU rendering (sorry, I don't know how I forgot to mention that..).
system-info.txt

CPU rendering (sorry, I don't know how I forgot to mention that..). [system-info.txt](https://archive.blender.org/developer/F160744/system-info.txt)

Aaron Carlisle commented

2015-04-11 20:39:31 +02:00

Added subscriber: @Blendify

gandalf3 commented

2015-04-12 07:29:16 +02:00

On GPU 2.74 is slightly faster than 2.73, so it seems this only affects CPU rendering.

I did notice that GPU renders of the scene are noticeably more noisy than CPU renders with the same number of samples. In particular near the spotlight source.

On GPU 2.74 is slightly faster than 2.73, so it seems this only affects CPU rendering. I did notice that GPU renders of the scene are noticeably more noisy than CPU renders with the same number of samples. In particular near the spotlight source.

Sergey Sharybin commented

2015-04-20 15:24:16 +02:00

Changed status from 'Open' to: 'Resolved'

Sergey Sharybin closed this issue

2015-04-20 15:24:16 +02:00

Sergey Sharybin commented

2015-04-20 15:24:16 +02:00

You wouldn't see difference in the noise in this particular scene. The reason of slowdown is caused by watertight nature of ray-triangle intersection now, which was compensated with QBVH for more complex scenes.

I've committed some memory optimization (828abaf) which also caused better utilization of CPU cache and solves speed regression in your particular file. But it's still possible some other files are still slower and this needs investigation.

But for now considering the report solved.

You wouldn't see difference in the noise in this particular scene. The reason of slowdown is caused by watertight nature of ray-triangle intersection now, which was compensated with QBVH for more complex scenes. I've committed some memory optimization (828abaf) which also caused better utilization of CPU cache and solves speed regression in your particular file. But it's still possible some other files are still slower and this needs investigation. But for now considering the report solved.

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Cycles is consistently slower in 2.74 #44337