Platform: NVIDIA CUDA
  Device: NVIDIA GeForce RTX 4090
    Driver version  : 570.133.20 (Linux x64)
    Compute units   : 128
    Clock frequency : 2610 MHz

    Global memory bandwidth (GBPS)
      float   : 895.18
      float2  : 923.80
      float4  : 941.00
      float8  : 952.13
      float16 : 962.52

    Single-precision compute (GFLOPS)
      float   : 84098.24
      float2  : 81316.00
      float4  : 81409.93
      float8  : 80840.90
      float16 : 80439.52

    No half precision support! Skipped

    Double-precision compute (GFLOPS)
      double   : 1423.48
      double2  : 1421.67
      double4  : 1418.26
      double8  : 1411.35
      double16 : 1397.99

    Integer compute (GIOPS)
      int   : 44874.12
      int2  : 44940.15
      int4  : 44661.71
      int8  : 44780.54
      int16 : 44792.71

    Integer compute Fast 24bit (GIOPS)
      int   : 44748.47
      int2  : 44771.30
      int4  : 44763.04
      int8  : 44584.92
      int16 : 43890.11

    Integer char (8bit) compute (GIOPS)
      char   : 38131.23
      char2  : 37187.54
      char4  : 36365.29
      char8  : 30460.54
      char16 : 28798.10

    Integer short (16bit) compute (GIOPS)
      short   : 37390.55
      short2  : 36008.64
      short4  : 37428.57
      short8  : 32507.62
      short16 : 27648.33

    Transfer bandwidth (GBPS)
      enqueueWriteBuffer              : 16.58
      enqueueReadBuffer               : 17.27
      enqueueWriteBuffer non-blocking : 16.27
      enqueueReadBuffer non-blocking  : 16.69
      enqueueMapBuffer(for read)      : 18.00
        memcpy from mapped ptr        : 5.30
      enqueueUnmap(after write)       : 26.84
        memcpy to mapped ptr          : 17.77

    Kernel launch latency : 3.92 us

Device: NVIDIA GeForce RTX 4090
  Driver version  : PoCL 7.0 (Linux x64)
  Compute units   : 128
  Clock frequency : 2610 MHz

  Global memory bandwidth (GBPS)
    float   : 893.98
    float2  : 922.62
    float4  : 940.90
    float8  : 951.27
    float16 : 960.86

  Single-precision compute (GFLOPS)
    float   : 74350.94
    float2  : 77464.59
    float4  : 80243.05
    float8  : 80875.78
    float16 : 80187.65

  Half-precision compute (GFLOPS)
    half   : 44700.20
    half2  : 89401.85
    half4  : 89370.85
    half8  : 88731.35
    half16 : 87873.76

  Double-precision compute (GFLOPS)
    double   : 1422.87
    double2  : 1421.34
    double4  : 1418.00
    double8  : 1411.15
    double16 : 1397.38

  Integer compute (GIOPS)
    int   : 30744.91
    int2  : 30581.63
    int4  : 30607.29
    int8  : 30897.89
    int16 : 30301.60

  Integer compute Fast 24bit (GIOPS)
    int   : 30686.79
    int2  : 30517.12
    int4  : 30543.35
    int8  : 30831.12
    int16 : 30575.51

  Integer char (8bit) compute (GIOPS)
    char   : 14530.12
    char2  : 21110.57
    char4  : 20030.94
    char8  : 20232.04
    char16 : 21248.74

  Integer short (16bit) compute (GIOPS)
    short   : 14339.66
    short2  : 21305.61
    short4  : 22127.96
    short8  : 20693.65
    short16 : 21423.40

  Transfer bandwidth (GBPS)
    enqueueWriteBuffer              : 21.80
    enqueueReadBuffer               : 21.29
    enqueueWriteBuffer non-blocking : 21.79
    enqueueReadBuffer non-blocking  : 21.29
    enqueueMapBuffer(for read)      : 363980.28
      memcpy from mapped ptr        : 5.18
    enqueueUnmap(after write)       : 26.00
      memcpy to mapped ptr          : 18.52

  Kernel launch latency : 3634.32 us