A recent paper describing a promising new mechanism has been brought to light at the 2023 International Conference for High Performance Computing, Networking, Storage and Analysis. Record-Replay (RR) is the new mechanism’s name. What it does? It speeds up applications on large scale graphics processing unit (GPU) systems by recording how a program runs on a GPU, and then replaying that recording to test different settings and finding the fastest way to run the program.
Developed by a Lawrence Livermore National Laboratory (LLNL)-led team, and with case studies already showing its effectiveness, the tool its said to help “autotune” large offload applications and therefore overcoming a major bottleneck for speed in scientific applications.
You can find out more about the RR tool on LLNL’s page.