cuda.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU",
    "gpuClass": "standard"
  },
  "cells": [
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "id": "IaY5IKw_2veM"
      },
      "outputs": [],
      "source": [
        "import numpy as np"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "array_cpu = np.random.randint(0,255, size=(4000,4000))"
      ],
      "metadata": {
        "id": "JwvjHsHuFvtX"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "array_cpu.nbytes/1e6"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fz6VkF3o3epe",
        "outputId": "5bc19755-72cf-4c36-ac5b-a6d173023453"
      },
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "128.0"
            ]
          },
          "metadata": {},
          "execution_count": 4
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "* **CuPy**: Drop-in NumPy replacement on GPU. Best when your workload is expressible via vectorized array ops (ufuncs, reductions, broadcasting) or uses libraries (cuBLAS/cuFFT/cuDNN). Very fast with minimal code. Also lets you write custom kernels via `ElementwiseKernel`, `ReductionKernel`, or `RawKernel` if needed.\n",
        "* **Numba (CUDA target)**: You write custom kernels with `@cuda.jit`. Great for **irregular** control flow, custom memory access patterns, or algorithms that don’t map cleanly to standard array ops. More control (shared memory, tiling, warps), more code. No automatic BLAS/FFT—those you’d call separately."
      ],
      "metadata": {
        "id": "DzoCRZ6FGETn"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import cupy as cp"
      ],
      "metadata": {
        "id": "hUEhBV0b3I-y"
      },
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "**cupy doesn't always mean faster**\n",
        "\n"
      ],
      "metadata": {
        "id": "vvff6z7wIlpq"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "%%timeit\n",
        "\n",
        "np.asarray(array_cpu)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "KkixcwBSZKBp",
        "outputId": "00eeb862-a630-4911-ad07-9bc59f9fcf0c"
      },
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "76.8 ns ± 29.8 ns per loop (mean ± std. dev. of 7 runs, 10000000 loops each)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "%%timeit\n",
        "\n",
        "cp.asarray(array_cpu)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "RsYfRDzI33bz",
        "outputId": "7bec0218-f19b-4ae4-9d28-8f9535d0798a"
      },
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "24.2 ms ± 364 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "type(array_gpu)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "w4P3FpZU4x5C",
        "outputId": "ae6ccc30-800c-4847-bdba-50ba40341ca8"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "cupy.ndarray"
            ]
          },
          "metadata": {},
          "execution_count": 19
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**using cupy scipy**\n",
        "\n",
        "\n",
        "\n"
      ],
      "metadata": {
        "id": "cxiSPqH1Iz0i"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "from scipy import fft"
      ],
      "metadata": {
        "id": "PPbiz1Y944zk"
      },
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "%%timeit\n",
        "\n",
        "fft.fftn(array_cpu)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "YW_5hsqN5Fa-",
        "outputId": "33d98f5a-224b-45a3-c8e1-742a438c63d5"
      },
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "373 ms ± 52.7 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from cupyx.scipy import fft as fft_gpu"
      ],
      "metadata": {
        "id": "3sUQe8465eyH"
      },
      "execution_count": 13,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "%%timeit\n",
        "\n",
        "fft_gpu.fftn(array_gpu)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "XPVOi14w5vZ4",
        "outputId": "b7a96087-3e36-40c5-e71e-e916c1d5bece"
      },
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "The slowest run took 4.16 times longer than the fastest. This could mean that an intermediate result is being cached.\n",
            "120 µs ± 79.4 µs per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
          ]
        }
      ]
    }
  ]
}

everything in colab is with ctrl + m + jupyter commands