Performance Results#
The NVIDIA Eye Contact NIM supports multiple concurrent inputs per GPU. The following table shows the performance data for various concurrency levels on various GPUs with the provided sample input files and the Python client.
GPU |
Input Video Resolution |
Concurrency |
Average FPS |
|---|---|---|---|
NVIDIA RTX PRO 6000 Blackwell Server Edition |
720p |
1 |
66 |
720p |
2 |
60 |
|
720p |
4 |
42 |
|
720p |
8 |
20 |
|
1080p |
1 |
51 |
|
1080p |
2 |
42 |
|
1080p |
4 |
31 |
|
1080p |
8 |
15 |
|
RTX L40 |
720p |
1 |
61 |
720p |
2 |
54 |
|
720p |
4 |
39 |
|
720p |
8 |
19 |
|
1080p |
1 |
42 |
|
1080p |
2 |
34 |
|
1080p |
4 |
26 |
|
1080p |
8 |
14 |
|
RTX L4 |
720p |
1 |
60 |
720p |
2 |
51 |
|
720p |
4 |
51 |
|
720p |
8 |
26 |
|
1080p |
1 |
41 |
|
1080p |
2 |
36 |
|
1080p |
4 |
28 |
|
1080p |
8 |
13 |
|
RTX A40 |
720p |
1 |
60 |
720p |
2 |
52 |
|
720p |
4 |
48 |
|
720p |
8 |
24 |
|
1080p |
1 |
38 |
|
1080p |
2 |
32 |
|
1080p |
4 |
22 |
|
1080p |
8 |
11 |
|
RTX A10 |
720p |
1 |
58 |
720p |
2 |
53 |
|
720p |
4 |
39 |
|
720p |
8 |
18 |
|
1080p |
1 |
42 |
|
1080p |
2 |
35 |
|
1080p |
4 |
26 |
|
1080p |
8 |
13 |
|
RTX A2 |
720p |
1 |
52 |
720p |
2 |
46 |
|
720p |
4 |
33 |
|
720p |
8 |
21 |
|
1080p |
1 |
35 |
|
1080p |
2 |
26 |
|
1080p |
4 |
20 |
|
1080p |
8 |
10 |
|
RTX T4 |
720p |
1 |
40 |
720p |
2 |
33 |
|
720p |
4 |
30 |
|
720p |
8 |
18 |
|
1080p |
1 |
29 |
|
1080p |
2 |
21 |
|
1080p |
4 |
18 |
|
1080p |
8 |
11 |
The average FPS is calculated by dividing the sum of number of frames in the input video files by the total inference time in seconds (measured from the time the concurrent requests are sent until complete output files are received by the client). This metric helps determine the effective performance scaling when running multiple inputs simultaneously.
For more information, refer to the Maxine NIM clients Github repository: NVIDIA-Maxine/nim-clients.