Not enough information for a real answer. What state are the cores in? Do you have 8 GPUs? Probably not, so if you run 8 jobs at once how does the GPU handle that concurrency? Are your cores cache thrashing? Do you have enough RAM for 8 jobs? Does your GPU have enough RAM for 8 jobs? If you have 1 thread per run but run the script in 8 different shells at once, does it perform differently? If you run 4 threads instead of 8, does it behave the same?