{
  "id": 570201,
  "title": "runnig on T4x2",
  "url": "/competitions/byu-locating-bacterial-flagellar-motors-2025/discussion/570201",
  "author_name": "",
  "post_date": "2025-03-26T13:44:37.699591200Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Has someone managed to run yolo inference on the two T4s? Did it double the speed?</p>",
  "messages": [
    {
      "id": "3160200",
      "postDate": "03/26/2025 13:44:37",
      "content": "<p>Has someone managed to run yolo inference on the two T4s? Did it double the speed?</p>",
      "rawMarkdown": "Has someone managed to run yolo inference on the two T4s? Did it double the speed?",
      "votes": null
    },
    {
      "id": "3160201",
      "postDate": "03/26/2025 13:47:31",
      "content": "<p>I didn't. But if you inference in pair batches it should. Remember to initialize the model correctly in both gpu.</p>",
      "rawMarkdown": "I didn't. But if you inference in pair batches it should. Remember to initialize the model correctly in both gpu.",
      "votes": null
    },
    {
      "id": "3160365",
      "postDate": "03/26/2025 17:17:05",
      "content": "<p>I did it. It not double the speed because main bottleneck is data loading and preprocessing(I did speed up this step as much as I can, now it utilize 400%cpu ^D), but it has significant boost.<br>\nCheck this notebooks: <a href=\"https://www.kaggle.com/code/fautei/byd-yolo-optimized-submission-notebook\" target=\"_blank\">https://www.kaggle.com/code/fautei/byd-yolo-optimized-submission-notebook</a> <a href=\"https://www.kaggle.com/code/fautei/byd-yolo-sahi-submission-notebook\" target=\"_blank\">https://www.kaggle.com/code/fautei/byd-yolo-sahi-submission-notebook</a></p>",
      "rawMarkdown": "I did it. It not double the speed because main bottleneck is data loading and preprocessing(I did speed up this step as much as I can, now it utilize 400%cpu ^D), but it has significant boost.\nCheck this notebooks: https://www.kaggle.com/code/fautei/byd-yolo-optimized-submission-notebook https://www.kaggle.com/code/fautei/byd-yolo-sahi-submission-notebook",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3160201,
      "author_name": "sacuscreed",
      "author_url": "",
      "post_date": "03/26/2025 13:47:31",
      "content": "<p>I didn't. But if you inference in pair batches it should. Remember to initialize the model correctly in both gpu.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3160365,
      "author_name": "fautei",
      "author_url": "",
      "post_date": "03/26/2025 17:17:05",
      "content": "<p>I did it. It not double the speed because main bottleneck is data loading and preprocessing(I did speed up this step as much as I can, now it utilize 400%cpu ^D), but it has significant boost.<br>\nCheck this notebooks: <a href=\"https://www.kaggle.com/code/fautei/byd-yolo-optimized-submission-notebook\" target=\"_blank\">https://www.kaggle.com/code/fautei/byd-yolo-optimized-submission-notebook</a> <a href=\"https://www.kaggle.com/code/fautei/byd-yolo-sahi-submission-notebook\" target=\"_blank\">https://www.kaggle.com/code/fautei/byd-yolo-sahi-submission-notebook</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3160200": "Has someone managed to run yolo inference on the two T4s? Did it double the speed?",
    "3160201": "I didn't. But if you inference in pair batches it should. Remember to initialize the model correctly in both gpu.",
    "3160365": "I did it. It not double the speed because main bottleneck is data loading and preprocessing(I did speed up this step as much as I can, now it utilize 400%cpu ^D), but it has significant boost.\nCheck this notebooks: https://www.kaggle.com/code/fautei/byd-yolo-optimized-submission-notebook https://www.kaggle.com/code/fautei/byd-yolo-sahi-submission-notebook"
  },
  "source": "meta"
}