Numerical Libraries

• 20 years ago

  • 1 Mflop/s - Scalar based
    • Linpack, Level 1 BLAS, loop unrolling

• 10 years ago

  • 1 Gflop/s - Vector & SMP computing, cache aware
    • LAPACK, Level 2 & 3 BLAS, block partitioned, latency tolerant

• Today

  • 1 Tflop/s - Highly parallel, network based, message passing
    • ScaLAPACK, data decomposition, communication/computation

• 10 years away

  • 1 Pflop/s - Many more levels MH, combination/grids & HPC
    • More adaptive, LT and bandwidth aware, fault tolerant, extended precision, attention to SMP nodes

Previous slide

Next slide

Back to first slide

View graphic version