using r9 380 to render in blender

[kobzarj]

I have an msi r9 380 4 gb GPU. today I enable the OpenCl compute in blender and the GPU pops up as one of the options (along with the CPU). Now I've been reading for some time now how rendering with the GPU should be several times faster than with CPU, however, in my case the opposite is happening. A render that took the CPU to complete 8:24 min, took the GPU 9:22. I am using Cycles Render by the way, and my CPU is FX 8370e. So my CPU is on the slow side, that's why I'm surprised that my GPU turned out to be so crap at rendering. Is this normal, or am I missing something, like not having some setting enabled???

Thanks in advance.

 

[13thmonkey]

what clock speeds does your GPU hit whilst rendering?

 

[renz496]

Well not an expert in this stuff but i see on other forum that some people that use blender with opencl end up using cpu to do their job even when they have amd gpu. i heard it is due to amd driver issue with opencl.

 

[kobzarj]

what clock speeds does your GPU hit whilst rendering?

honestly I'm not exactly sure how to measure that. I have the MSI Gaming App and use the preset OC profile, which should clock the GPU speed at 1000mhz. However I'm not exactly sure how to check the actual speed while rendering.

 

[kobzarj]

so I dug a little more on the new and found the following:

1. AMD GPU work like crap on Blender and you need an NVIDIA CUDAs to render properly.
2. through the command prompt, you have to navigate to the blender forder then type "set CYCLES_OPENCL_TEST=ALL", press enter, then "start blender.exe".
3. In the user preferences navigate to the SYSTEM>compute and Under OpenCL tab choose "CPU+GPU" (it won't be exactly like this, rather it will have the name of your CPU and GPU)
4. In short watch this video "https://www.youtube.com/watch?v=91AHD8YYAPc"

Following this method, my render time was cut down to 4:24, which is almost twice as fast as I was getting on the CPU or GPU alone. I expected to see 9 tiles being rendered simultaneously (8 CPU cores + 1 tile GPU), but the result is 2 tiles being rendered. Guess the GPU renders 1 tile, and all of the CPU cores render the other. Now, although this provided good results I still don't feel satisfied. I was expecting the GPU alone to do this kind of rendering without burdening the CPU. Maybe the r9 380 is not good enough for this, but I have seen videos of much older Nvidia cards perform much better, so it might be the OpenCL itself. If anyone finds anything new on this, please inform me as I would love to hear about it.
Thanks

 

[13thmonkey]

I don't think it's a blender and AMD issue exactly, it's more that opencl isn't very good, regardless of the hardware running it and AMD is stuck with it, cuda is just more efficient. Small example in this link.

https://forums.adobe.com/thread/1894349?start=0&tstart=0