How fast does your GPU run the AMD Blender benchmark?

[cbn]

To run the benchmark...

1.) first install Blender 2.78c:

https://www.blender.org

2.) Launch Blender

3.) In the upper left hand corner of the interface locate "File" and left click on it.....A drop down menu will appear.

4.) On this drop down menu, left click on "User Preferences".

5.) Left click on "System" in "User preferences".

6.) In the lower left hand corner of "System" you will find "Cycles compute device". Select CUDA if you have an Nvidia GPU and Open CL if you have an AMD GPU. Click on "Save user Settings".

7.) Close Blender

8.) Then download the AMD file:

http://download.amd.com/demo/RyzenGraphic_27.blend

Blender will launch when you open the file.

9.) With Blender open locate "Render" in the right hand of the user interface and change device from "CPU" to "GPU Compute".

10.) Then scroll down and look in the lower right hand corner of the user interface to locate "Tiles". Change values of 32 for both X and Y to 256 for both X and Y.

11.) Press F12 to start the benchmark.

 

[cbn]

My GTX 660:

31.63 seconds

 

[Brunnis]

GTX 1080:

11.24 s

 

[DeathReborn]

Odd but... 56.87 s Single 290 PCS+ @ 1.25GHz

 

[SKORPI0]

R9 290X Crossfire

17.25 seconds.

 

[bbhaag]

43.4 on a MSI RX480 Gaming 4gb

 

[cbn]

Some slow scores for AMD, but looking forward to seeing how they bench (on this same file) with the upcoming Blender 2.79:

https://www.blendernation.com/2017/04/12/blender-cycles-opencl-now-par-cuda/

https://wiki.blender.org/index.php/Dev:Source/Render/Cycles/OpenCL

 

[TheELF]

1050ti ~22 sec

 

[ddogg]

1080Ti - 09.99

 

[Carfax83]

08.24 on a Titan Xp:

 

[Bacon1]

3770k - ~56sec (32 tiles)

Fury Nitro - 29.5 sec (256 tiles)

The 3770k had 8 going at once while my Fury only had a single part render at a time, not sure if that is normal or not.

At tile size of 512 render time went down to 20.13 sec on GPU.

Tile size of 300 render time was 24.41

1024 tile size was similar to 512 with 20ish sec.

Edit:

I think blender-2.78-6a546fc-win64-vc14.zip is the newest build which will be 2.79 and it does have multiple renders at a single time on my GPU similar to the CPU, but now it does it in multiple passes (very low but almost instant render which increases in quality). It takes about a minute so not sure if I broke a setting or if something is wrong with it, I'll test it more out later, but maybe other people can try it and see if its just me or something up with that build.

 

[MentalIlness]

26.2 withn FX 8120 @ 4.0 / GTX 760 rig in sig.

 

[wilds]

1070 @ 2050 Mhz core 9.4 Ghz memory

was about 0.8 seconds slower at stock speed.

 

[Bacon1]

1070 @ 2050 Mhz core 9.4 Ghz memory

was about 0.8 seconds slower at stock speed.

Your 1070 seems to be as fast / faster than ddog's 1080 Ti? Are you guys using different settings?

 

[wilds]

Your 1070 seems to be as fast / faster than ddog's 1080 Ti? Are you guys using different settings?

Followed exact instructions in OP. Any more 1070's?

 

[Bacon1]

Edit:

I think blender-2.78-6a546fc-win64-vc14.zip is the newest build which will be 2.79 and it does have multiple renders at a single time on my GPU similar to the CPU, but now it does it in multiple passes (very low but almost instant render which increases in quality). It takes about a minute so not sure if I broke a setting or if something is wrong with it, I'll test it more out later, but maybe other people can try it and see if its just me or something up with that build.

So I did some more testing and it looks like even though it showed it in the options, it wasn't actually doing GPU compute using my GPU but still using CPU while using the downloaded version I linked above... Explains a lot! Its also back to just rendering 1 square at a time vs multiple on the CPU rendering. I had to deselect and reselect it and then it worked.

So using the latest version I linked with 256x256 it took 20.21sec to render.

1024x1024 took 18.90 vs 21.25
512x512 took 19.15 vs 20.82
350x350 took 19.33 vs 25.43
300x300 took 19.51 vs 25.05

So it is definitely faster than before, esp at the lower tile sizes while the 512x512 appears to be just slightly faster. But 20 vs 29 sec is a huge 45% boost 🙂 @ 256x256

64x64 is 26.46 vs 2 min 31 sec so 5.8x faster
32x32 is 51.09 vs 7 min 24 sec(!!!) thats 8.7x faster in the new build!

So yeah with smaller workloads it's a massive increase in the new build.

 

[DaQuteness]

MSI GTX970 4GB @ 256x256 buckets : 15.89
MSI GTX960 4GB @ 256x256 buckets : 25.12
both on the same render (just for giggles - I have 2x GPUs in my PC, I work in Blender constantly) @ 256x256 : 10.94

Wanna hear the sad part? I had the 960 first, then added the 970 in the other slot... to realize that this crap motherboard limits second GPU to 4x...so my 970 is running 4x according to GPUz.

 

[cbn]

Would be interesting to see what some laptop iGPUs can do as well?

 

[4K_shmoorK]

GTX 1080 Ti
8.13s

Edit:
With +575 mem (~12350MHz)
7.62s

Now we just need 2x Xp/Ti...

 

[Bacon1]

1080Ti - 09.99

08.24 on a Titan Xp:

8.13s

Edit:
With +575 mem (~12350MHz)
7.62s

Wonder how much of this difference is in the memory clocks as it appears to make a big difference in timing and why your 3 cards are showing such different results.

 

[ddogg]

Wonder how much of this difference is in the memory clocks as it appears to make a big difference in timing and why your 3 cards are showing such different results.

I have no clue. There was also someone with a 1080 who had a higher time than the guy with the 1070 :shrug. I can try playing with my memory speeds to see if it makes a difference

 

[wilds]

I think memory latency has something to do with it.

 

[DaQuteness]

Would be interesting to see what some laptop iGPUs can do as well?

GT755m / 2GB - 1:11.96