When trying to render using GPU, I get: “Cuda binary kernel for this graphics card compute capability (6.1) not found.”.
I tried renaming kernel_sm_52 to kernel_sm_61, but it now throws an error “Cuda error at cumoduleLoad: Invalid Source”
Until Blender/CUDA support your card, if you are interested you could do some tests with OpenCL (kernel split or big kernel). In Maxwell cards OpenCL take about twice as long compared with CUDA (in BMW scenes), so you may be able to get an idea. But I really do not know how Pascal cards performs with OpenCL.
Anyway, if you want to try:
Open Blender with BMW27.blend scene, hit Spacebar and search: debug
(or you can also access with Ctrl+Alt+D)
Choose that option and in Debug value put: 256
and Ok.
Go to the Render tab, and find ‘Debug’ menu. Under OpenCL Flags, in Kernel choose: Split (If ‘Split’ does not work, you choose ‘Mega’), and in Device: GPU.
Now go to File > User Preferences > System, under Compute Devices select OpenCL, and your GPU model.
Configure big tile sizes, at least 480x270. Render and wait until kernel is compiled. Render again and share the result.
Yep, just as it says in the linked report: For Cycles to work on a GPU, we need to have a CUDA Toolkit from NVidia that supports the GPU - that new toolkit isn’t officially released, though, and Brecht posted a comment somewhere that the RC version of it actually seems to slow down older cards…
I guess the CUDA 8 toolkit RC supporting Pascal is out. Is there any word on when there will be an official fix for the GTX 1000 series cards as of now! Really hoping they come out soon, I have just ordered a GTX 1060
The Nightly works for my GTX 1080 in Cycles, and the AA issue was fixed only after switching the Sampling from “Path Tracing” to “Branched Path Tracing.”
The performance is worse than my old (now dead) GTX 780 Ti by about 10 seconds. I was able to make up some of the performance time per render by about 5 seconds by changing my tile size to X: 256, Y: 256. I used to have it at X: 512, Y: 512 on my 780 which was the best performance I could get for that card in render time.