CUDA architecture for root?

ROOT
1

Please read tips for efficient and successful posting and posting code

_ROOT Version:6.22_06
_Platform:_fedora33
_Compiler:_gcc (GCC) 10.3.1 20210422, nvcc cuda_11.4.r11.4/compiler.30033411_0

I’m compiling root from source using cmake. There is now a request/warning that the flag -DCMAKE_CUDA_ARCHITECTURES
be set to a non-blank value. I can’t find a list of allowed architecture values - some examples give numbers like compute_50 but no explanation of their meaning. I don’t find values for this spected for my GPU. Root compiles and runs fine with the value blank, but is there an optimum value to be used for root to enhance performance, and how does this relate to the source code and the gpu hardware?
I have not found much by direct searches.

2

That’s already a good thing. I do not see this flag listed in the list of build options for ROOT. May be @bellenot will be able to give you more details when he will be back.

3

Hi, FYI, CUDA has not been ported/tested on Windows

4

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.