It has been a few months since the release of 2.78, and the Blender Foundation has been sitting on a bunch of performance enhancements in that time. Since 2.79 is still a couple of months off, they decided to “cherry pick” a bunch of them back into the 2.78 branch and push out an update to it. Most of these updates are things like multi-threading their shader compiler for Cycles, speeding up motion blur in Cycles, and reducing “fireflies” in Cycles renders, which indirectly helps performance by requiring less light samples to average out the noise.
I tried running two frames from different scenes of my upcoming PC enthusiast explanation video. While they’re fairly light, motion graphics sequences, they both use a little bit of motion blur (~half of a 60 Hz frame of integration) and one of the two frames is in the middle of three objects with volumetric absorption that are moving quite fast.
The "easier" scene to render.
When disabling my GTX 670, and only using my GTX 1080, the easier scene went from 9.96s in 2.78a to 9.99s in 2.78b. The harsher scene, with volumetric absorption and a big streak of motion blur, went from 36.4s in 2.78a to 36.31s in 2.78b. My typical render settings include a fairly high sample count, though, so it’s possible that I could get away with less and save time that way.
The "harsher" scene to render.
Blender is currently working on Agent 327, which is an upcoming animated feature film. Typically, these movies guide development of the project, so it makes sense that my little one-person motion graphics won’t have the complexity to show the huge optimizations that they’re targeting. Also, I had a lot of other programs running, which is known to make a significant difference in render time, although they were doing the same things between runs. No browser tabs were opened or closed, the same videos were running on other monitors while 2.78a and 2.78b were working, etc. But yeah, it's not a bulletproof benchmark by any means.
Also, some of the optimizations solve bugs with Intel’s CPU implementation as well as increase the use of SSE 4.1+ and AVX2. Unfortunately for AMD, these were pushed up right before the launch of Ryzen, and Blender with Cycles has been one of their go-to benchmarks for multi-threaded performance. While this won’t hurt AMD any more than typical version-to-version variations, it should give a last-minute boost to their competitors on AMD’s home turf.
Blender 2.78b is available today, free as always, at their website.
Well, they better to start to
Well, they better to start to do some performance optimizations.
Here is why: https://www.youtube.com/watch?v=-4qWKJf6je4
Some Blender builds by some others are offering 3x-5x more performance on the very same hardware…..
While I’m not really sure
While I'm not really sure what this video is supposed to be compared against, try changing your tile size to 16×16 (it's currently 32×32). CPUs tend to like smaller tile sizes, while GPUs tend to like bigger (~480×240 or so). Might shave a few seconds.
Also, I've seen benchmarks from Blender Foundation that shows Linux is significantly faster than Windows 10 (~2x… Windows 7 was somewhere in between)… although that was on CUDA and it could have been differences in NVIDIA drivers, not so much differences in operating systems. (I'm guessing Linux CUDA drivers are pretty mature, if only because of supercomputing.) Blender has also undergone several optimizations since then, too.
With official Blender
With official Blender version, the ryzen file(sample 150) takes around 205-207 seconds on my rig. Very same hardware with Stilt build same file(sample 150) takes under 57 seconds.
More to read to understand the results: https://forums.anandtech.com/threads/post-your-ryzen-blender-demo-scores-amd-clarifies-blender-benchmark-confusion-run-150-samples.2494600/page-12
The video contains 2 different result , 100 sample and 150 to compare the ryzen CPU performance.
Same operation system has been used: Win 10 pro 64 bit.
Anyway, official Blender builds are terrible when comes down to optimizations. No matter if you have Intel or AMD CPU’s.
The whole test made only on CPU. No GPU has been involved.
Best thing is, do a test on your rig and submit the result at the video or here 😀 Links are provided at the video’s description. unless , you are afraid to do the test 😀