Hardware recommendation for a box to train object detection models

General Discussion
,
#1

Hi, I am training tflite models for custom Object Detection. Some of my jobs are running for more than 30hours on my 16GB Ram, i7 Mac. I am thinking of setting up a new box to run the jobs. Is there any information regard what type of hardware configuration works best with Tensorflow model training?

This is the basic configuration I current have in mind:

OS: Windows
Processor: i7
RAM: 32GB
GPU: RTX 3080

Need to balance performance and cost. Am I over/under investing in any part? Any other things I should pay attention to?

#3

@Raymond_Wong,

Welcome to the Tensorflow Forum!

Generally GPU will help you faster rather than CPU. I am confident following configurations will help you train faster and make sure that model is training on GPU.

OS: Windows
Processor: i7
RAM: 32GB
GPU: RTX 3080

Thank you!

#4

@chunduriv Thanks. Do you have any resources that will give us more in-depth insight?

we are all trying to minimize cost here. So frankly, we are trying to get the minimum hardware spec that will do the job for us.

Let’s say if I drop CPU down to i5, more much of a performance degrade? After all, it is the GPU that is doing more of the work?

Or if I drop the graphics card down to RTX 3060, how much slower will the job run?

Hoping to find a way to qualify all these