cuda install and run

1.install 

1.1、 Upload the CUDA used to the SUT. 

1.2、Execute the command: 

#./cuda_12.1.0_530.30.02_linux.run  

Run 

Win 

nbody.exe -benchmark -fp64 -device=X -numbodies=Y 

nbody.exe    ->  Command to executes N-Body simulation test CUDA sample code 

-benchmark     ->    N-Body simulation is a program with Windowed animation, if testing GPU is with display output port and monitor connected to it; however, for stress purpose, animation is not needed no matter display output or not, so add -benchmark parameter. 

-fp64    ->   Needed only for 100-ish or 800-ish GPU, proceed stress with Float-point 64bit format, like V100, A100\A800, H100\H800, these model is designed based on Double precision float point number(FP64), using -fp64 could better uplift TDP to maximum. 

-device=X  ->   X is GPU numbering, 0 based, if a system installed wiht 8 GPUs, numbering will be 0~7, distinguish target GPU to assign workload with this parameter among each of command line window. 

-numbodies=Y   ->    

Y is the data to proceed workload, it should be a number to make large amount of VRAM usage and GPU will be expected to run on it for over than 12 hours, 

This number depend on VRAM capacity, for a 10GB or more adapter, this number can be start from 51200000 to 102400000, 

If this number is too big for a specific VRAM capacity, nbody will be interrupted, then this number should be decreased with degree of 10240000 

Different GPU has different performance, if a GPU is top model and numbodies is too samll, like 10240000, stress test could be finished faster than 12 hours, so this number is recommended to be estimated from higher number, like 81920000