Object detection is a painstaking taking and requires lots and lots of computational resources. I have been trying to train my model using GCLOUD with the following configurations.\n\nMemory optimized instance with 30GB RAM and 4 cores processor.\nNvidia K80 GPU with 12GB VRAM \n585GB SSD for storing data.\n \nThis took about 8 hours to train the model (on validation set) and then 5 hours to generate complete predictions.
This configuration does works flawlessly but does incur a lot of bill to me.
\n\nWhat system configurations have you been using and how does it performs?
", "messages": [ { "id": "590364", "postDate": "08/02/2019 06:17:00", "content": "Object detection is a painstaking taking and requires lots and lots of computational resources. I have been trying to train my model using GCLOUD with the following configurations.\n\nMemory optimized instance with 30GB RAM and 4 cores processor.\nNvidia K80 GPU with 12GB VRAM \n585GB SSD for storing data.\n \nThis took about 8 hours to train the model (on validation set) and then 5 hours to generate complete predictions.
This configuration does works flawlessly but does incur a lot of bill to me.
\n\nWhat system configurations have you been using and how does it performs?
", "rawMarkdown": "Object detection is a painstaking taking and requires lots and lots of computational resources. I have been trying to train my model using GCLOUD with the following configurations.\n``` \nMemory optimized instance with 30GB RAM and 4 cores processor.\nNvidia K80 GPU with 12GB VRAM \n585GB SSD for storing data.\n``` \nThis took about 8 hours to train the model (on validation set) and then 5 hours to generate complete predictions.\n\nThis configuration does works flawlessly but does incur a lot of bill to me. \n\nWhat system configurations have you been using and how does it performs?", "votes": null }, { "id": "596423", "postDate": "08/10/2019 16:11:03", "content": "K80 has a large memory size, but low performance as per TFLOPS... I you can select a different GPU with a newer architecture and smaller memory size, go for it and change the batch size depending your new GPU memory.
\n\n", "rawMarkdown": "K80 has a large memory size, but low performance as per TFLOPS... I you can select a different GPU with a newer architecture and smaller memory size, go for it and change the batch size depending your new GPU memory.\n\nhttps://www.pugetsystems.com/labs/hpc/TensorFlow-Performance-with-1-4-GPUs----RTX-Titan-2080Ti-2080-2070-GTX-1660Ti-1070-1080Ti-and-Titan-V-1386/", "votes": null } ], "comments": [ { "id": 596423, "author_name": "marcunzueta", "author_url": "", "post_date": "08/10/2019 16:11:03", "content": "K80 has a large memory size, but low performance as per TFLOPS... I you can select a different GPU with a newer architecture and smaller memory size, go for it and change the batch size depending your new GPU memory.
\n\n", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "590364": "Object detection is a painstaking taking and requires lots and lots of computational resources. I have been trying to train my model using GCLOUD with the following configurations.\n``` \nMemory optimized instance with 30GB RAM and 4 cores processor.\nNvidia K80 GPU with 12GB VRAM \n585GB SSD for storing data.\n``` \nThis took about 8 hours to train the model (on validation set) and then 5 hours to generate complete predictions.\n\nThis configuration does works flawlessly but does incur a lot of bill to me. \n\nWhat system configurations have you been using and how does it performs?", "596423": "K80 has a large memory size, but low performance as per TFLOPS... I you can select a different GPU with a newer architecture and smaller memory size, go for it and change the batch size depending your new GPU memory.\n\nhttps://www.pugetsystems.com/labs/hpc/TensorFlow-Performance-with-1-4-GPUs----RTX-Titan-2080Ti-2080-2070-GTX-1660Ti-1070-1080Ti-and-Titan-V-1386/" }, "source": "meta" }