GTX 1080 Ti Various Benchmarks (BMW, Classroom, fishy cat)

Blender 2.78c
4770K Stock
32 GB Ram
GTX 1080 Ti Founders Edition (driver 378.78 - released 9th March 2017)
Windows 10

GPU when idle sits around 56 C (Corsair Link). Rises to 68% to 70% during renders. Room temp is 23 C

File used to bench: https://code.blender.org/2016/02/new-cycles-benchmark/

BMW Scene
F12 : 2 minutes 20 seconds
4 Tiles: 2 minutes
1 tile : 2 minutes 11 seconds

Classroom
F12: 5 minutes 7 seconds
4 tiles: 6 minutes 5 seconds

Fish Cat
F12: 6 Minutes 28 seconds

Koro
F12: 32 minutes 24 seconds

Pavallion
F12: 7 Minutes 49 seconds

1 Like

I think this is at 30 samples. It should be 1000 samples

1 Like

Nice! I get about 30 seconds with two 1070s so you’re right in line with the Titan X (pascal). man, two of them would kill it. Probably the sweet spot of the Pascal architecture.

If 30 samples is the default when you load that scene then yeah.

Hi.
Sorry, I’d given you a wrong link on the other thread (demo files) :o
Actually I did not know that in blender.org the different links had different configurations (it is somewhat confusing)
This is the correct one containing all the scenes (Cycles benchmark zip):
https://code.blender.org/2016/02/new-cycles-benchmark/

Could you do the tests again with those files, even bmw scene? (scenes for GPU)

Sorry for the inconvenience.

Thank you!

Edit:
I think you could get better render times with bigest tiles size than the default settings in those scenes.

Thanks. Please test with latest master and with the right benchmark pack files to make result comparable. https://download.blender.org/demo/test/cycles_benchmark_20160228.zip

Sure, will do it now.

With the new benchmark file posted these are the results:

BMW Scene
F12 : 2 minutes 20 seconds
4 Tiles: 2 minutes
1 tile : 2 minutes 11 seconds

Classroom
F12: 5 minutes 7 seconds
4 tiles: 6 minutes 5 seconds

Fish Cat
F12: 6 Minutes 28 seconds

Koro
F12: 32 minutes 24 seconds

Pavallion
F12: 7 Minutes 49 seconds

Cheers

are those times with latest master?

I used the file you posted.

‘bliblubli’ means most recent Blender development build from here (no need to install, it’s portable):
https://builder.blender.org/download/

For Windows are the first links, not the experimental ones below

@D3Pixel, could you please put those new results in the first message of the thread?
Also the link that I had given you to download the Cycles benchmark zip file

It’s just that I would feel guilty if there were future confusions. Because official benchmark results are the ones that are useful for future comparisons.

Thanks again and sorry for the inconvenience, is that you are the first with the 1080 Ti :slight_smile:

Updated original post to reflect latest benchmark file results. I used the Blender release for windows 64 from the Blender home page. Also, is the file that bliblubli posted the same as yours? If so will modify original post to reflect your link instead.

Edit: Just tested it, it is the same file.

Latest master got many optimisations, it would be great to see what 2.79 will be like. It would be great if you test with this version https://builder.blender.org/download/blender-2.78-96868a3-win64.zip :slight_smile:

Yes, the same. It is official Blender for benchmark files:

I did a quick test with this build and it rendered the BWM scene 2 seconds quicker. That was all I had time for.

Dang, quite a bit faster then my GTX 960 under 2.78c Win10, will try latest build.

BMW Scene
GTX 960 6:07.75

Classroom
GTX 960 18:00.33

Fish Cat
GTX 960 13:40.18

Koro
GTX 960 1:08:13.20

Pavallion
GTX 960 23:09.01

Just for comparison:

BMW scene
GTX 1050 Ti: 7 minutes 26 secs (F12)

That’s less than a third of the 1080 Ti. Oh well, this thing fits into almost any boring office computer.

Baah, it is a TitanXP (GP102) with 1GB memory less ( but faster GDDR5x), and a bit faster clock speed. so yes it will not change much. ( only price )

I’m using today’s daily build on my Dual Rx480 on Windows 10.

BMW Scene
2x Rx480: 2 minutes 3 secs (2 tiles)

Can you try running the 1080 Ti with OpenCL instead of Cuda?

Use 20 samples instead of the default 35 samples.