The new Cycles GPU 2.73 Benchmark

Hi all, the new benchmark Cycles.
It is GPU only.
Benchmark use now 256x256 tiles.
It is a compromise between rendertime and keep 3 or more GPU busy.
Please don´t upload images, render time and specs only, as stamped.

Please use official Blender 2.73 only, we can compare performance better if all use the same Blender version.

I add the 2.72 file here too, please render both files with 2.73.

Open files, hit F12, write down:
Be aware of long rendertime on low end GPU´s, my GTX 760 need 16 Minutes!
It should be possible to render the bench on 1 GB GPU.

OS
GPU
Rendertime:

Sponza: XX:XX.XX as stamped.
2.72 file: XX:XX.XX as stamped.



Original file and textures from Marko Dabrovic (Lightwave).
Adapted for Cycles from Pixal (Greg Zaal), many thanks.
http://portfolio.gregzaal.com/
I change some settings, reduce texture size for benchmarking.

File 1: https://www.dropbox.com/s/22624e4aggpumtu/sponza_cycles_benchmark.blend?dl=0

File 2: https://www.dropbox.com/s/w986r1mya7mppyr/cycles_bench_272.blend?dl=0

Spreadsheet Libre Office file: https://www.dropbox.com/s/ee1glhxme32ktk6/Cycles%20GPU%20Bench_273.ods?dl=0

Cheers, mib

My results:
OpenSUSE 13.2/64
GTX 760 16:10.01
GTX 560Ti 448 12:29.51 | The Empire strikes back! :slight_smile:
Both 07:25.29

Cheers, mib

I remember Sponza, I group I used years ago for 3dsmax used it as a benchmark back then, late 90’s I think it was. Good chice mib. Anyway the results :-

OS : Win 64
GPU : 2 x GTX 780 ti SC
Time : 04:04.15

** EDIT

Sorry, scrub the last results, I had “Auto Tile Size” enabled.I set it to 256x256 and got these results :-

OS : Win 64
GPU : 2 x GTX 780 ti SC
Time : 04:06.77

Not a massive difference.

Cheers, Wig

mib2berlin

Interesting that the Fermi card is beating the 760 in this one. I have 2 580’s doing nothing. Might drop them back in my spare machine jus for comparison.

Cheers, Wig

My results:

OS: W8.1 64Bit
GPU: 2x GTX 970’s @ Stock (Reference models)
Time: 04:29:61

GPU: 2x GTX 970’s OC @ 1515 MHz
Time: 04:10:17

My results:

OS: Windows 8.1 / 64bit
GPU: 2x GTX 780 OC 4:18.64
Time: 04:18.64

Time: with even size tiles 200x360 is 04:00:05.

Mib2berlin why not optimize tile size to fit evenly with the dimensions size? Like in previous benchmarks. It is noticeably faster as you can see above.

Edit: Corrected the tile size to 200x360.

My result:

OS Windows 7 pro 64 bit
GPU GTX 580 3gb
Rendertime: 10:41.55

OS: Windows 7 64bit
GPU: GTX 780 Ti OC
Time: 07:24.04

OS: Windows 8.1 64bit
GPU: GTX 980 Strix
Time: 07:07.43

What did you break lol

OS: Windows 8.1 64 bits
GPU: GTX 780 + Geforce Titan
256x256 > Time: 05:45:96
400x360 > Time: 05:28:62

Windows 10 x64
GTX 970 + GTX 560 Ti
Time 5:49.22

Hey, thanks for posting.
These are interesting results, it seams the 900 and the 500 cards are profit from the scene or latest Cuda development.
There are not much entries atm. but I post the first results.


@LordOdin, Your result is a bit annoying, as fast as 780+TITAN, or the 700 series lost a lot of performance.
@eklein, tile setting was discussed a lot in the last benchmark thread.
Absolute result is not important, we only want compare the GPU´s.

Cheers, mib

Ubuntu 14.04
GTX 760 - 14:57.36

CPU comparison:-
Intel 4930K @ 4.5 GHz
Tiles 16 x 16 - 17:03.53

OS: Windows 7 64bit
GPU: MSI GTX 780 6 Gb & GTX 580 3Gb
Time: 05:12.15

With only GTX 780 : 9:18.79
With only GTX 580 : 10:50.04

OS: Windows 7 64bit
GPU: GTX 660 Ti
Time: 14:56;94

Windows 8.1 x64
GTX 970
Time 8:10.57

If you want to compare GPU only you need scenes that dont take 20+ seconds to build the BVH

I know a 3960x can build BVH more than double the speed of my FX-8350 (rough tests could be wrong but its deffo faster)

Win 8.1 pro 64-bit
MSI 750ti gaming OC (stock)
time: 16:34.20

Overclocked @1289/1525
time: 14:18.93

Win 8.1 pro 64-bit
GTX 690 (otherwise a 5 year old computer: i7 920)
Time: 07:29.20