Added RTX5090 performance on different operating system (Win11/WSL/Fedora42)

plot_comparison.ipynb

@@ -0,0 +1,191 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "ccd5f401",
+   "metadata": {},
+   "source": [
+    "Comparing RTX5090-32GB results from different operating system (Windows11/WSL(Debian)/Linux(Fedora)). Using A100-SXM4-40GB results as reference."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "a9af3974",
+   "metadata": {},
+   "source": [
+    "You need to change output dir `output_folder`."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ff083421",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import pandas as pd\n",
+    "import glob\n",
+    "import plotly.express as px\n",
+    "import plotly.graph_objects as go\n",
+    "import torchvision.models as models\n",
+    "output_folder = \"fig/5090/\""
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "e072b9c4",
+   "metadata": {},
+   "source": [
+    "You need to include the data you want in `CSV_DIC`."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "abd92c0a",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "MODEL_DIC = {\n",
+    "    'mnasnet': models.mnasnet.__all__[1:],\n",
+    "    'resnet': models.resnet.__all__[1:],\n",
+    "    'densenet': models.densenet.__all__[1:],\n",
+    "    'squeezenet': models.squeezenet.__all__[1:],\n",
+    "    'vgg': models.vgg.__all__[1:],\n",
+    "    'mobilenet': models.mobilenet.__all__[1:],\n",
+    "    'shufflenetv2': models.shufflenetv2.__all__[1:]\n",
+    "}\n",
+    "model_list = MODEL_DIC.keys()\n",
+    "\n",
+    "REV_MODEL_DIC = {}\n",
+    "for key, values in MODEL_DIC.items():\n",
+    "    for v in values:\n",
+    "        REV_MODEL_DIC[v] = key\n",
+    "\n",
+    "BENCHMARK_TYPE = {\n",
+    "    'half_model_train_benchmark',\n",
+    "    'half_model_inference_benchmark',    \n",
+    "    'float_model_train_benchmark',\n",
+    "    'float_model_inference_benchmark',\n",
+    "    'double_model_train_benchmark',\n",
+    "    'double_model_inference_benchmark'\n",
+    "}\n",
+    "\n",
+    "CSV_DIC = {\n",
+    "    'RTX5090/WIN11': glob.glob('results/5090/Win11/'+'*.csv'),\n",
+    "    'RTX5090/WSL': glob.glob('results/5090/WSL/'+'*.csv'),\n",
+    "    'RTX5090/Fedora': glob.glob('results/5090/Fedora/'+'*.csv'),\n",
+    "    'A100-SXM4-40GB': glob.glob('results/dgx-a100/'+'A100-SXM4-40GB_1*.csv')\n",
+    "}\n",
+    "gpu_list = CSV_DIC.keys()\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "bd6367a5",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "data_buffer = []\n",
+    "for btype in BENCHMARK_TYPE:\n",
+    "    for gpu in gpu_list:\n",
+    "        found_csv = [item for item in CSV_DIC[gpu] if btype in item]\n",
+    "        df = pd.read_csv(found_csv[0])\n",
+    "        df['run'] = df.index + 1\n",
+    "        df['gpu'] = gpu\n",
+    "        df['benchmark_type'] = btype\n",
+    "        df_melted = df.melt(id_vars=['run', 'gpu', 'benchmark_type'], var_name='model', value_name='latency')\n",
+    "        df_melted['model_type'] = df_melted['model'].map(REV_MODEL_DIC)\n",
+    "        data_buffer.append(df_melted)\n",
+    "all_data = pd.concat(data_buffer, ignore_index=True)\n",
+    "all_data = all_data.set_index(['benchmark_type', 'model_type'])\n",
+    "all_data = all_data.sort_index()\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "34460c49",
+   "metadata": {},
+   "source": [
+    "Preview a figure for testing."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "a600b299",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "fig = px.box(\n",
+    "    all_data.loc[('double_model_train_benchmark', 'resnet')],\n",
+    "    y=\"latency\",\n",
+    "    color=\"gpu\",\n",
+    "    facet_col=\"model\",  # facet_col / facet_row\n",
+    "    title=\"Latency by Model and GPU/\" + \"double_model_train_benchmark\"\n",
+    ")\n",
+    "fig.update_traces(boxmean=True)\n",
+    "#fig.update_yaxes(dtick=50)\n",
+    "\n",
+    "for annotation in fig.layout.annotations:\n",
+    "    annotation.text = annotation.text.replace(\"model=\", \"\")\n",
+    "    # annotation.textangle = -15\n",
+    "\n",
+    "fig.show()\n",
+    "#fig.write_image(\"output.png\", width=3000, height=1000)\n",
+    "#fig.write_image(\"output.png\", width=1500, scale=2)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "2fb8040d",
+   "metadata": {},
+   "source": [
+    "Generate all figures."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "134bec11",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "for idx in all_data.index.unique().tolist():\n",
+    "    fig = px.box(\n",
+    "        all_data.loc[idx],\n",
+    "        y=\"latency\",\n",
+    "        color=\"gpu\",\n",
+    "        facet_col=\"model\",  # facet_col / facet_row\n",
+    "        title=\"Latency by Model and GPU/\" + idx[0]\n",
+    "    )\n",
+    "    fig.update_traces(boxmean=True)\n",
+    "    for annotation in fig.layout.annotations:\n",
+    "        annotation.text = annotation.text.replace(\"model=\", \"\")\n",
+    "    fig.write_image(output_folder+idx[0]+'_'+idx[1]+\".png\", width=1500, scale=2)"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "base",
+   "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": "3.13.2"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

requirement.txt

@@ -5,3 +5,4 @@ pandas
 plotly
 cufflinks
 psutil
+kaleido