Need guidance in setting up production server #1067
Replies: 2 comments 4 replies
-
Subcenter Arcface is a very compute-consuming mode. One thing to consider - Subcenter Arcface model requires a little bit more than 4Gb of GPU memory. So I would recommend getting GPU with at least 6Gb of memory. |
Beta Was this translation helpful? Give feedback.
-
@pospielov Thanks for the guidance. 2- 3- Maybe Intel is not good for this algorithm and Ryzen is good. (No clear idea) RTX 3060Ti 8GB with 1 uWSGI process. RTX 4070 12GB RTX 3090 -> 5uWSGI process |
Beta Was this translation helpful? Give feedback.
-
I've installed the library on Ubuntu Server. I needed the most accurate algorithm, so I chose Subcenter Arcface, without GPU.
I installed the system on docker.
The machine specs are:
Intel Core-i7 13700k
Corsair 64GB 5600Mhz
2 x Samsung 980 Pro 2TB NVME (Speed upto 7000 MBPS each)
Still I get 4-5 seconds per image recognition.
I really need help to optimize the system.
I'm wondering what GPU should I use for best efficiency
Have I implemented the system in correct way or not.
Should I implement on kubernetes (Don't know how)
What should be the settings in .env file etc.
Any guidance/help would be appritiated.
Thanks
Beta Was this translation helpful? Give feedback.
All reactions