Cycles AMD OpenCL Test

Hi, since 2.69 it is possible to activate OpenCL rendering in Blender.
It is a bit hidden/tricky to activate but here you are:

EDIT: Since 2.75 OpenCL is activated for AMD GPU.
If you want to activate for all devices change environment variable to

CYCLES_OPENCL_SPLIT_KERNEL_TEST=1

If you want to use old mega kernel use environment variable in the following way:

Windows:

Win Key + R type CMD press enter
go to your blender folder. ex: cd C:\Program Files\Blender Foundation\Blender

press enter and then type: set CYCLES_OPENCL_TEST=all

And now: start Blender with blender.exe

Inside blender go to File -> User preferences -> System

Under compute device select OpenCL and then your card

Close the preferences windows and go to the render panel and select device: GPU Compute.

LINUX:

Open xterm, change directory to your blender folder with “cd”.

Type “export CYCLES_OPENCL_TEST=all”

Enter

Start Blender with “./blender”

All without quotes.

Inside blender go to File -> User preferences -> System

Under compute device select OpenCL and then your card

Close the preferences windows and go to the render panel and select device: GPU Compute.

OSX:

  • Open a Terminal (e.g. via CMD+Spacebar to open spotlight, search for “Terminal”)
  • cd into the folder where you put Blender
  • “export CYCLES_OPENCL_TEST=all”
  • Run Blender via “./blender.app/Contents/MacOS/blender”

Procedure from DingTo

Please post your OS, AMD card name and driver version to help other user.
Keep in mind that nvidia Cards are working from the beginning but are slower with OpenCL.
So it is not needed to put Nvidia cards here!

Cheers, mib.

I tested 1055T, Juniper (5770) and Tahiti (7950) on Windows 8 with the 13.12 drivers and a tungerz build from GraphicAll. Didn’t work, testing on Linux next.

Hi Octol, could you test also with Experimental VS 2013 or Linux official from builder.blender.org ?
Thanks cusa123 for testing.

Cheers, mib.

juicyfruit msvc 2013
compile time render total time notes
1055T 6.57 1.44 8.01 default scene
1055T - 1.41 1.41 -||-
7950 199.72 199.03 398.75 -||-
7950 - 1.54 1.54 -||- (faster in viewport than 1055T)
5770 crash to desktop -||-
5770 crash to desktop -||-

1055T - 238.13 238.13 BMW1M-MikePan rev2 (no change in settings)
(scripts are auto disabled)
7950 - 551.38 551.38 -||- (200s. of sync)
7950 - 352.27 352.27 -||- (no sync)
7950 - 302.06 302.06 -||- (no sync) (960*540)

Still windows 8 and 13.12 drivers

(why oh why must my post be moderated)
compilation render total notes
1055T - 536.98 536.98 Cornell-mib2berlin (no change in settings)
(looks odd, normal cpu render looks correct though)
7950 - 1008.78 1008.78 -||- (looks correct)
7950 - 876.37 876.37 -||- (135135)
7950 - 691.38 691.38 -||- (540
540)



Also, the volume absorption and subsurface scattering appears to work as well, but not the the volume scatter (latest vs13 build from builder.blender.org).

wondering if worth to try it
set up seems a bit complicated

on win 8 how much faster is it then rendering on CPU?

and is this faster then working with latest mingw on win 8 ?

I got this HD8670 D

thanks

As things stand right now OpenCL on GPU is no faster than on the OpenCL CPU, or the normal CPU kernel. You can try running both the GPU and the CPU at the same time with OpenCL and that might be faster, however the 8670D isn’t exactly overflowing with abundant compute power.

Only way to know really would be to try. Please report any results here.

(Also, I doubt it will be faster than MinGW.)

but how much faster is it then on cpu ?
or compare to gpu ?
I mean on this card I cannot run GPU as on a GE card!

right now mingw is like 1.3 times faster then cpu on 2.69
which is nice and faster!

I might later on add a small GE 650 card which does not use much power - around 65 watts I think
and it should be faster then the HD card

thanks

Short answer: It’s not faster.

since 2.69 it is possible to activate OpenCL rendering in Blender.

What is the advantage of OpenCL?

MingW not exactly up-to-date though, "
[TR=“class: file”]
Windows 64 bit
Experimental MinGW
blender-2.67-r57348-mingw64.zip
65M
Mon Jun 10 13:23:25 2013
[/TR]

"
from buildbot

What is the advantage of OpenCL?

With a good card (200$ +) a GPU is much faster in some cases.
It seams Cycles is not optimized for OpenCL at the moment so there is not much advantage.
One thing is may be interesting, a CPU speak also OpenCL, you can render with CPU and GPU at the same time.

@Writer’s Block, I heard they kick the MinGW build, try the Experimental VS 2013 build.

Cheers, mib.

With a good card (200$ +) a GPU is much faster in some cases.

I have a GTX580 CUDA card that I use for Cycles & Octane GPU renders.

Cycles renders at least 20X faster using the GPU than the Phenom 955 CPU.

I don’t think OpenCL will help me at this point, but thanks for the info. :smiley:

Think of OpenCL as a open non proprietary standard for parallel computing that kind of works like Nvidia’s CUDA but instead it should eventually work for all processors including Nvidia, AMD, FPGAs, other GPUs and CPUs. If you have got a 12 core CPU it can use those other cores also if the software is set up to use them! OpenCL is kind of in its infancy and it is just gaining momentum. OpenCL is being promoted by the Khronos group the same people that do OpenGL for 3D graphics and OpenGL ES for 3D graphics on mobile devices.

Not sure if it’s helpful, but a week or 2 ago I did some quick testing on several machines using the BMW1M-MikePan.blend file.

On my home machine, which runs via a shortcut to set CYCLES_OPENCL_TEST=all I got the following render results:

Home Workstation (CPU) - 15:37 (Core2Duo @ 2.67Ghz, 8GB)
Home Workstation (GPU) - 6:18 (ATI 7870)
Home Workstation (CPU + GPU) 6:26

So yeah…it might be time to upgrade my processor :slight_smile:

Also, in case it’s useful I did some testing with the machines where I work:

(Blender is started using the standard shortcut. CYCLES_OPENCL_TEST=all was NOT set)

All CPU rendering:

Current workstation - 4:02 (i7-860 @ 2.8Ghz 4cores/8threads, 16GB)
test workstation - 2:31 (i7-4770 @ 3.4Ghz 4cores/8threads, 8GB)
test Server 1:45 (dual x5650 @ 2.67 12cores/24threads, 72GB)

Not able to render on my iMac.

OS X 10.8.5
AMD Radeon HD 6770M Driver Version: 01.00.544
2.7 GHz Intel Core i5
12 GB 1333 MHz DDR3

Compiling OpenCL kernel ...
OpenCL error (ATI Radeon HD 6770M): [CL_DEVICE_NOT_AVAILABLE] : OpenCL Error : Error: Build Program driver returned (517)
OpenCL error (ATI Radeon HD 6770M): OpenCL Warning : clBuildProgram failed: could not build program for 0x1021b00 (ATI Radeon HD 6770M) (err:-2)
OpenCL error (ATI Radeon HD 6770M): [CL_BUILD_ERROR] : OpenCL Build Error : Compiler build log:
Error returned by cvms_element_build_from_source

AMD quad, 64 bit running Win7 64bit.
ATI 6870 card.
I have the latest drivers and the latest official build but the kernal wont compile.

again my hopes have been dashed!

its been a while since i posted that. i was wondering if anyone knew a way to get the kernel to compile.

anyone ?

@octol: nice renders! btw, your posts only have to me moderated for the first 5 or so posts, then you can post away. i think its supposed to be a spam-check thing.