can you share the bottle, i’ll try to render on my side and confirm same issue.
2.81 beta dated october 27th seems to speed things up a lot (although still slower than 2.80)
2.81 (oct 21) gpu + cpu (32x32 tile) - 10.56.05
2.81 (oct 27) cpu+gpu (32x32 tile) - 5:40.75
2.80 cpu +gpu (32x32 tile) - 5.26.70
Definitely a step in the right direction though.
Anyone else found the same?
Unfortunately not. Rendering speed remained the same. Checked on the 29th
Using the bottle file…
Blend ver | Tile 32 | Tile 64 | Tile 128 | Tile 256 |
---|---|---|---|---|
2.80 | 3.26 | 3.01 | 3.02 | 2.12 |
2.81 | 11.29 | 7.39 |
I can see that my dual Vega 56 are used at around 50% each when using small tiles
Did anyone do any followup checks? shoudl we raise a bug report on 2.81 ?
I tested on a fresh build again, the problem remains. think, need to report this problem.
Yes I have the same issue. Can post stats if useful - but it’s similar to what you’ve posted @Grzesiek
Anyone tried the 2.82 alphas?
I’ll test 2.82 when i get home tonight. then raise a bug report with link to this thread
While official support will remain focused on CUDA and OptiX, the next version of E-Cycles will have much better performance with OpenCL (2x faster compared to 2.81 on Vega with the official BMW and Barcelona scenes).
Edit: build based on latest Blender 2.82 is up, feedback is welcome
So did a final run… using builds from yesterday November 8th 2019 @bliblubli
Single Vega 64 - BMW scene from Blender.org site for testing. only tiles were changed
Blender | 256 | 128 | 64 | 32 | Build |
---|---|---|---|---|---|
2.80 | 1m28s | 1m30s | 1m39s | 1m56s | official |
2.81 | 2m.34s | 2m34s | 2m43s | 3m20s | eefd806afc15 |
2.82 | 2m.35s | 2m34s | 2m43s | 3m25s | 23e1fb365b65 |
Will still try to do Classroom test.
It’s all done with Blender’s buildbots right?
Yes, pulled from the web on November 8th.
So based on above I’m raising a bug report on this, hoping they will revert this slowdown before 2.81 is released…
issue raised:
https://developer.blender.org/T71479
Based on the developers:
Just confirmed that the new Voronoi is the source of the regression.
so we’ll see what they do, and we might get “full” speed in 2.81/2.82 “soon”?
Did you by any chance use “Hilbert Spiral” and Denoiser? Give “Right to Left” a try.
not really.
I re-ran default BMW scene from Blender demo test files. ANd used deffault settings as per file to ensure they can replicate the issue.
IT ended up that it was related to one of the shaders and based on last post they stated:
I think this is sufficiently resolved now for the 2.81 release.
During the 2.82 cycle we can look at getting back a bit more of the performance since it’s still not fully the same.
I just downloaded 2.81 latest build (November 14th) and render times are much better.
Origianl 2.80 - 1m28s
Pre Nov 14 2.81/2.82 - 2m34s
Nov 14 2.81/2.82 - 1m46s
Thanks for raising this @Grzesiek - very impressive how quickly the blender team addressed it too.
2.81 has brought things back slightly, definitely a step in the right direction.
I’ve been having good results with ecycles in the meantime… the latest version (based on 2.82) has really cut my render times on opencl.
Will post some comparison stats tomorrow if it’s not too off topic.
If you have time, just try to compare “Hilbert Spiral” with “Left to Right”. I had great differences in render times when using the Blender Denoiser. My scene was big though and had render times about 45 minutes.
Ok will give it a try.