'Resource usage problem with Python/Tensorflow - 2 PC comparison
I am working on a simple task. I am using python, Anaconda, Tensorflow, Cuda, and OpenCV. Reading a video frame by frame, doing 4 predictions on each frame with 4 different light models. I have 2 PC which have different hardware but completely the same environment. The same TensorFlow/Cuda/python versions etc. One of these PC has really high specs, and the other one has lower specs than the other one. The problem is, I am getting 10 FPS with low specs PC and 5 FPS with high specs PC. By 10 FPS, I mean, the PC can read 10 frames and complete the prediction tasks on these frames in each second. I am not sure if is this something normal or not. I am not sure about is there any boundaries related to Windows, devices, etc. So any idea would be helpful.
An important point is that 40% of CPU and %75 of GPU are used during the process in low specs PC and 3% of CPU and %6 of GPU are used during the process in high specs PC. It seems like, high specs PC is not using its resources or has a bottleneck somewhere else. High specs PC has 4 GPU but I am using only 1 of them during to process for the comparison.
Low specs PC:
Graphics Card: NVIDIA GeForce RTX 2070 Super
CPU: Intel Core i7-10750 CPU @2.60Ghz (12 CPUs)
Ram: 32GB
FPS: 10
CPU Usage: 40%
GPU Usage: 75%
High specs PC:
Graphics Card: 4 x NVIDIA GeForce RTX 3080
CPU: Intel Xeon Silver 4210 CPU 2.20GHz (40 CPUs)
Ram: 250GB
FPS: 5
CPU Usage: 3%
GPU Usage: 6%
Sources
This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.
Source: Stack Overflow
| Solution | Source |
|---|