Merge pull request #16 from MichaelClifford/gpu-test

add gpu test notebook

notebooks/confirm_gpu_available.ipynb

@@ -0,0 +1,218 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "16e3552b-6245-4e8a-ab1c-2e17729901d4",
+   "metadata": {},
+   "source": [
+    "# Confirm Notebook Can Access GPU \n",
+    "\n",
+    "This notebook provides a couple of small examples that confirm your notebook environment is able to connect to a GPU, either locally or on a cluster. The bottom section also provides a short example showcasing the increased performance of using the GPU over a CPU for a matrix multiplication task. \n",
+    "\n",
+    "\n",
+    "The one caveat to ensuring this notebook works correctly and you can communicate with the gpu, is that you must install pytorch.\n",
+    "\n",
+    "```\n",
+    "pip3 install torch==1.9.0+cu111 torchvision==0.10.0+cu111 torchaudio==0.9.0 -f https://download.pytorch.org/whl/torch_stable.html\n",
+    "```\n",
+    "\n",
+    "In the near future this requirement will be handled automatically. "
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "8578d326-157b-4512-9379-367cf4629d93",
+   "metadata": {},
+   "source": [
+    "## Am I connected to a GPU?"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "id": "a66360fb-ff15-477b-a815-62c229bb39b3",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import torch\n",
+    "import timeit"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "id": "ef63602d-c4e0-4056-898f-f43fba8eba35",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "tensor([[0.3546, 0.9113, 0.7730],\n",
+      "        [0.1329, 0.8297, 0.8176],\n",
+      "        [0.4446, 0.3666, 0.1261],\n",
+      "        [0.4993, 0.8514, 0.9197],\n",
+      "        [0.7537, 0.0293, 0.4049]])\n"
+     ]
+    }
+   ],
+   "source": [
+    "x = torch.rand(5, 3)\n",
+    "print(x)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "id": "e22ed182-b282-4ebd-aba2-91cce53c321d",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "True"
+      ]
+     },
+     "execution_count": 3,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "use_cuda = torch.cuda.is_available()\n",
+    "use_cuda"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "id": "840b1c7e-8832-4538-9173-90f07bab5be7",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "__CUDNN VERSION: 8005\n",
+      "__Number CUDA Devices: 1\n",
+      "__CUDA Device Name: Tesla K80\n",
+      "__CUDA Device Total Memory [GB]: 11.99702016\n"
+     ]
+    }
+   ],
+   "source": [
+    "if use_cuda:\n",
+    "    print('__CUDNN VERSION:', torch.backends.cudnn.version())\n",
+    "    print('__Number CUDA Devices:', torch.cuda.device_count())\n",
+    "    print('__CUDA Device Name:',torch.cuda.get_device_name(0))\n",
+    "    print('__CUDA Device Total Memory [GB]:',torch.cuda.get_device_properties(0).total_memory/1e9)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "18651412-75e9-4894-86a2-563fc89ed652",
+   "metadata": {},
+   "source": [
+    "The above outputs indicate that pytorch is installed, it can see the gpu, and it shows us some details about the device we are connected to. "
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "3dc87e6e-5c8a-4754-9d0b-76c44bb4cd04",
+   "metadata": {},
+   "source": [
+    "## Does the GPU speed things up?"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "id": "5364fc66-f118-439e-891b-64c0e50cca01",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def batched_dot_mul_sum(a,b):\n",
+    "    return a.mul(b).sum(-1)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "id": "d2be1632-0854-4feb-8d51-c6ffeaa18797",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "cpu = torch.device(\"cpu\")\n",
+    "gpu = torch.device(\"cuda\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "id": "e58e65b6-4e04-44e5-a3a5-dd45d1af0dab",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "x_cpu = torch.randn(4000,4000,device=cpu)\n",
+    "x_gpu = torch.randn(4000,4000, device=gpu)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "id": "4a2cfde3-b101-46db-b9e4-0b5d72a10ca2",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "mul_sum(x, x) cpu:  2.2202524829190224 seconds\n",
+      "mul_sum(x, x) gpu:  0.003196620149537921 seconds\n"
+     ]
+    }
+   ],
+   "source": [
+    "t0 = timeit.Timer(stmt='batched_dot_mul_sum(x,x)', setup='from __main__ import batched_dot_mul_sum',\n",
+    "                  globals={'x': x_cpu})\n",
+    "\n",
+    "t1 = timeit.Timer(stmt='batched_dot_mul_sum(x,x)', setup='from __main__ import batched_dot_mul_sum',\n",
+    "                  globals={'x': x_gpu})\n",
+    "\n",
+    "print(f'mul_sum(x, x) cpu:  {t0.timeit(100)} seconds')\n",
+    "print(f'mul_sum(x, x) gpu:  {t1.timeit(100)} seconds')"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "833d220d-bc5a-4698-9076-b7725ea6d257",
+   "metadata": {},
+   "source": [
+    "We can see from the output above that the speed up on this particular node is orders of magnitude faster using the GPU over the CPU for a 4000,4000 matrix multiplication. Yeah!\n",
+    "\n",
+    "If this notebook does not run for you, that means you are not connected to a GPU. "
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.8.3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}