GTX 1080 does not support with Octane render

Hi Nvidia,

I’m motion graphic designer, my primary machine is Nvidia GPU and Octane Render.
I have 2 GTX 970 and need to upgrade to 2 GTX 1080 for speed up my present rendering jobs with deadline.
Today I just buy 2 of GTX 1080 and I plan to sell my 2 of GTX 970 so I make appointments with buyers.
After I install 2 of GTX 1080 I found that they still have not support with Octane render because of CUDA.
I download CUDA 8.0RC from developer site,
I try to Install CUDA 8.0 from Nvidia’s developer forum but it doesn’t work because the CUDA installation has displayed {this graphics driver could not find compatible} ,
I send this bug report to Nvidia developer so they reply
{As noted on the CUDA 8.0 RC download page, GeForce GTX 1080 developers must re-install the latest driver from www.nvidia.com/drivers after installing any of these CUDA Toolkits.

Please install the CUDA 8.0 RC Toolkit without selected driver components, and then install latest driver version - 368.39 for Windows 10 64-bit/GTX 1080, it’s compatible with the CUDA 8.0 RC Toolkit as well. Thanks.}
I follow their steps but the result still same.

I’m still waiting for your progress for across this nightmare, waiting for CUDA 8.0 official till August I think that too difficult and may be cause adversely broad because CUDA is very important no less than gaming.

It is possible that if you can help find a solution.
I still waiting for your support

Thanks,

I think the subject line would be more appropriately titled “Octane renderer does not support GTX 1080”?

If the Octane renderer uses a proper fat binary with a PTX fall-back path, it should in principle be able to run on the GTX 1080 with the latest drivers, using JIT compilation of any of the CUDA kernels used for acceleration. I wonder whether this renderer performs any specific checks on either CUDA version or GPU architecture that would preclude use of CUDA 8 or the sm_61 architecture with the currently shipping version. It would probably be worthwhile checking on these items with the vendor of this product.

If the Octane renderer imposes such whitelist restrictions, you may have to wait until the vendor adds support for CUDA 8 and/or the GTX 1080. It will depend on the Octane software how involved this is. For example Folding@Home supports new GPUs simply by adding their PCIe ID to a vendor white list, which can be done very quickly (and has already happened for GTX 1080 as far as I am aware).

An old saying states “Never change horses in midstream” and that pretty much still applies to modern project management: It is usually best not to change hardware configurations or software component versions in the middle of a project, especially when working towards a quickly approaching deadline. My suggestion would be to simply revert your setup to the last working configuration.

Octane renderer is not provided by NVIDIA. Have you contacted the provider?

Regarding CUDA 8 install on GTX 1080, did you follow the install guide including the verification steps?

If so, what was the result of running the verification steps after CUDA install? (Be specific - provide the exact output.)

It’s my understanding that the current build of Octane Render does NOT support or is not optimized for the Pascal Architecture of the GTX 1080. The Octane team has stated that they will release support for Pascal Architecture and Octane Render optimizations once the full release of CUDA 8 launches in August.

Whenever I attempt to fire off a render in the plugin I am greeted with an alert that reads “There is no CUDA device which is selected…” in the Standalone the renderview states “no CUDA GPUs”
However, it is interesting that the settings in the Octane Render plugin under devices, availabile GPUs, it shows gpu0:GeForce GTX 1080, so we know it sees the GPU.

In regards to njuffa, I have no idea if this is from whitelist restrictions on GPU or CUDA version. I could swear I remember reading that someone had the 1080’s working with the current release of Octane with CUDA 8 RC, but perhaps I misread and am falling victim to wishful thinking.

So far I have the same results as degreebeat (upgrading from a single aftermarket GTX 780 Ti to FE GTX 1080) more on that here: https://devtalk.nvidia.com/default/topic/942447/cuda-setup-and-installation/cuda-8-0-rc-couldn-t-detect-gtx-1080/

I am facing the same issue with all GPU renderer’s I’ve tried.

VRay’s GPU support - fails.
RedShift GPU renderer - crashes
Furryball GPU renderer - doesnt see any GPU, renders black with errors
Octane - same as OP.

I can’t believe NVidia released the cards before CUDA was completed. For content creators, this makes the card completely useless for, what, another 2 months?

I bought these cards to be a GPU rendering workhorse. My workflow has been impeded and I will surely be missing some deadlines now. I depend on WORKING products. My fault I guess for not checking support, but I had no indication that “Cuda” didn’t mean what it said. Apparently I’ll have to wait 2 more months, and then however long it takes the renderer’s to actually implement the new version of CUDA. By which time the projects I had intended to GPU render will already have come and gone.

NVIDIA doesn’t have control over 3rd party software.

It’s entirely possible to write an application that uses CUDA, and “just works” when a newer GPU or a newer CUDA version comes out. There are many examples of this.

It’s also possible for an application writer to explicitly check an installed GPU against a “whitelist”, and only turn on the GPU acceleration path if the detected GPU belongs to the “whitelist”. This might be what is happening in some of these cases where CUDA is installed correctly but the app does not recognize the GPU. In such a situation, there is really nothing that can be done from the NVIDIA/CUDA side to affect the application behavior, and presumably, the app writers want it to be that way.

CUDA 8 is in an “RC” stage right now, essentially beta. This is an opportunity for application developers and other advanced users to get an advanced look at new capability. NVIDIA doesn’t recommend that production work immediately be cut over to any RC toolkit the moment it is available. It is principally provided for advanced evaluation. It is expected that there will be bug fixes and other changes that differentiate the RC build with the production release build.