note that the NVidia profiler tool is broken...

t$ ./vaddd
-------------------------------------------------------------
        system name = Linux
          node name = titan
            release = 6.4.0-genunix
            version = #1 SMP Mon Jun 26 15:53:25 GMT 2023
            machine = x86_64
          page size = 4096
       avail memory = 540904357888
                    = 528226912 kB
                    = 515846 MB
-------------------------------------------------------------
INFO : number of host CPUs:     40
     : number of CUDA devices:  2
     : we need 13027889070 bytes of memory on a GPU
     :    0: Quadro GP100 totalGlobalMem = 17069309952
     :    0: Quadro GP100 is to be selected
     :    1: Quadro K6000 totalGlobalMem = 12791250944
     : min memory unit is 1: Quadro K6000 with 12791250944 bytes
     : max memory unit is 0: Quadro GP100 with 17069309952 bytes
     : selected Quadro GP100

     : Quadro K6000 selected

     : Vector addition of 516979725 double FP64 elements
     : random data loaded 8365043093 nsecs   8.365043 secs
     : Wallclock cudaMalloc(A)    4685645 nsecs  0.004685645 secs
     : Wallclock cudaMalloc(B)    4712680 nsecs  0.00471268 secs
     : Wallclock cudaMalloc(C)    4414429 nsecs  0.004414429 secs
     : Copy of vector A from host to device done.
     : Wallclock cudaMemcpy() 455459656 nsecs  0.4554596 secs
     : Copy of vector B from host to device done.
     : Wallclock cudaMemcpy() 463675894 nsecs  0.4636759 secs
     : CUDA kernel launch with 504864 blocks of 1024 threads
     : vectorAdd done.
     : Wallclock kernel launch 50036587 nsecs  0.05003659 secs
     : cudaEventElapsedTime claims 0.04999689 secs
     : Copy result vector C from device to host done.
     : cudaMemcpy() 1223154597 nsecs   1.223155 secs
     : A + B correct within error 2^(-48) epsilon
     : result check done 1513243663 nsecs   1.513244 secs
DONE : total time 16178688201 nsecs   16.17869 secs
t$