• [Experience sharing] GPU CUDA uses memory padding to avoid bank conflict


      Welcome to pay attention to my public account [Jizhi Vision], reply 001 to get Google programming specifications   O_o   >_<   o_O   O_o   ~_~   o_O This article talks about how to use memory padding in GPU CUDA programming to avoid bank conflict. 1、Shared memory Shared memory is a small, low-latency on-chip […]

  • Install FIO from source and enable the GDS (cufile) engine


    Want to use benchmark to test GDS(GPUDirectStorage)The speed, generally use the official NvidiaGdsio,butFIOIn version 3.26, an engine that can use GDS is provided, so the author chooses to use FIO and Gdsio for experimental comparison.However, due to the low attention paid to GDS and the lack of unofficial information, even an English search has no […]