Hi, I've been trying to root cause why I see huge run to run variation in stream performance. Running the same binary on the MTL machine with the same background load, I see a stream Triad score of either: Function Rate (MB/s) Avg time Min time Max time Triad: 2400.5403 0.2000 0.2000 0.2001 or Triad: 36019.3566 0.0151 0.0133 0.0431 This is from an interactive login. I haven't tried this using the batch mode yet. This is the OpenMP version using 64 threads, though I've tried other thread counts as well. Does anyone have an idea as to what could be impacting my bandwidth by 15x run to run? Thanks
Para obtener más información sobre las optimizaciones del compilador, consulte el aviso sobre la optimización.