Optix 6 support for sm_75 (RTX2060)

I am still working with Optix 6 and it seems it is not possible to use the sm_75 arch setting. I can compile my .cu to ptx with --arch=compute_75 but on launch I get the error: in _rtProgramCreateFromPTXString “Failed to parse input PTX string”.
I searched the documentation but did not find any limits on GPU arch mentioned. Would upgrading to Optix 6.5 solve this issue? I dont see a need to upgrade to Optix 7 so far as this would lead to a lot of code changes.

Upgrading from OptiX 6.0.0 to 6.5.0 is definitely recommended.

Note that since OptiX 6.5.0 the OptiX core implementation and the denoiser ship with the display drivers and that means you get bugfixes and improvements with each driver release and there have been plenty in the seven months since OptiX 6.5.0 was released.

The PTX parser in OptiX 6.0.0 might not have been updated to the newest available SM version. This happened before, see: https://forums.developer.nvidia.com/t/rt-error-invalid-source-when-using-sm-61/53948

You should be fine when limiting your SM target to the minimum supported Maxwell GPUs which is SM 5.0. Even Kepler’s SM 3.0 will do as long as the CUDA toolkit supports it. OptiX rewrites the code internally to the underlying SM version and the final PTX assembler will optimize it again.

Hi,
and thanks for the answer.
I tried both in the meantime. Limiting Optix 6.0 to sm_70 works and also just compiling against Optix 6.5 does the job. I don’t see any serious performance issues on either option so any will do for me for the moment. Good to know that there are no API changes between 6.0 and 6.5 (at least for the code I am using currently).

This is good to known too.
Thank you

.