My friend, codes changes… Different necessities requires different code architectures. Alpha or 2.79 early codes are not finished codes and not give you correct results.
I said you. If you get true results, render heavy scenes with long render times. After that, if you think these results invalid, send a bug report.
That is the point of this post, to get extra people to CONFIRM where the issue is so THAT I can REPORT A BUG.
As for “not give me correct results” since it is “not finished code” you do understand that 2.8 only the interface and Evvee were redesinged. Cycles wasn’t dractically changes, just few optimizations. Its very similar code between 2.8 and 2.79. And on average, 2.8 gives slightly faster resutls as it had more optimizations done.
So again, This post is about confirming that the bug exists for others before I bother the Blender Developers, as they have plenty of work on their hands.
I said what is what. If you don’t want listen to me, you know.
I would rather like for you to use BMW 2car scene and render at the various tile settings and report back.
I rendered with my old computer:
Intel i5 2500K + Geforce GTX 670
BMW (GPU, 256x256) --> 4.27
BMW (GPU, 32x32 ) --> 3.35
BMW (GPU, 16x16 ) --> 3.34
BMW (GPU+CPU, 32x32 ) --> 3.07
BMW (GPU+CPU, 16x16 ) --> 3.04
You can see, long render times always better for benchmark…
You see, render times are change case by case and depend on GPU capabilities. Some scenes give approximately 2x fast render with CPU+GPU, some scenes not too much. My graphic card slow when I use 256 tiles rather than 32 tiles. Your graphic card slow when you use 16 tiles rather than 128 tiles.
I’m just concentrating on the GPU side, and see that smaller tiles indeed give you slight speed improvements, which is the only part i’m interested in. ANd it shows that Nvidia is ammue to the bug.
Dam…
so I just need another OpenCL comparison… Though nvidia can run wiht OpenCL, as they don’t really optimize it will definitely have worse performance. Any chance you can test Nvidia card with OpenCL and report again?
My card not support OpenCL. But I can say this, depend on GPU capabilities. All GPU architecture different than other. Nvidia card architecture different than AMD cards and this diffirencies are normal, not a bug.
In 2.79 I got fast render with 256 tiles, but in 2.80, I can get fast render with 32 tiles. Codes changes, everything changes…
Hmm intresting, thought OpenCL is supported by most Nivida GPU’s.
As for scaling/architecture. Again, in 2.8 Alpha/Beta, AMD scaled identicially to the GTX cards on smaller tiles. Hence it is a Bug. I’m just trying to identify where the Big is before I report
You know!..
Hi,
I have couple of video projects which are pending. That’s why I am also investigating whether there is any kind of bug 
BMW
GPU
512x512: 1’14"
256x256: 1’12"
128x128: 1’12"
64x64: 1’20"
32x32: 1’40"
16x16: 3’30"
CPU+GPU
64x64: 1’20"
32x32: 1’10"
16x16: 1’43"
For me, hybrid rendering does not seem beneficial for BMW scene, even worse in Classroom.
Dam … so scaling down to 64x64 you almost dont’ loose any performance… Now I’m lost… 
What Operating System and drivers are you using?
So current benches are more and more perplexing…
| Source | Device\Tile | 256 | 128 | 64 | 32 | 16 | Conclusion |
|---|---|---|---|---|---|---|---|
| Birdnamnam | GTX1070 | 1.48.0 | 1.33.0 | 1.29.0 | 1.28.0 | 1.29.0 | –> Correct Scaling <– |
| egamentncr | Radeon VII | 1.12.0 | 1.12.0 | 1.20.0 | 1.40.0 | 3.30 | Scaling as it should… |
| Hikmet | GTX 670 | 4.27.0 | x | x | 3.35.0 | 3.34.0 | Also good scaling… |
| Grzesiek | Vega 56 | x | 1.38.6 | 1.49.2 | 2.19.6 | 4.11.7 | Wrong Scaling |
| Grzesiek | 2x Vega 56 | x | 1.17.0 | 1.19.6 | 1.39.3 | 2.19.7 | Bad bad… |
| Grzesiek | CPU - TR 1950x | x | x | 2.15.9 | 2.04.4 | 2.00.5 | |
| Grzesiek | CPU + 1 GPUs | x | x | 1.27.7 | 1.15.6 | 1.29.3 | |
| Grzesiek | CPU + 2 GPUs | x | x | 1.19.5 | 1.07.6 | 1.10.6 |
Windows 10 updated version.
Nvidia seems to perform better on small tiles on GPU than AMD. Maybe that’s why hybrid rendering is more beneficial with CUDA.
Just FYI:
Blender Cycles can render using OpenCL on NVIDIA cards.
You have to start it with this flag:
--debug-value = 256
But it’s not worth it. OpenCL is crippled on NVIDIA.
OpenCL not works good on Nvidia cards.
Just my saying…
I simply wasn’t sure if you are aware of the flag.
Anyways, let’s not get this OT.
Agreed on the tile size, but your differences are so significantly similar compared to mine.
So Windows 10, and I’m guessing the latest drivers…
Yes.
Is it necessary to share this thread with the developers?
This week/weekend I’ll do last dab. Reinstall the OS and fresh drivers. If nothing changes, then YES.
Everything I do results in the same. 2 GPU’s are not being used to their full potential. And tile settings for OPenCL is producing reduced performance when tiles are small enough to use the CPU…
I just want to be sure 100% it’s not my system.
I did a full driver cleanup using DCC tool, and after reinstall of 19.8.1 drivers, still same results. tiles performance is bad, and multi GPU is bad.
I raised an initial ticket, and will do more testing to support them. Now I need to locale the 2.79 version that came after where smaller TIles in OpneCL were not impacting performance that much. Will keep you posted