Our Verticals
are Horizontal,
all SAAS
and On Prem.
IKM® VIDEO GENAI,
REAL-TIME SYNTHESIS,
EXTRACTION,
DYNAMIC GENERATION
AND SHARING,
OF ALL THE ELEMENTS
THAT CHARACTERIE MEANING, PRESENT IN A VIDEO STREAM.
Evolutionary Performance
The results expressed are the result of the passion, perseverance and sharing that the company has been able to produce.
References
Steaming Video:
2M, 5M, 8M Pixel
10, 15, 20, 20 fps
GPU:
Hopper, Ampere, Volta, Jetson.
Measurements
All performances expressed, are measured mainly with our systems, based on Nvidia Hopper, Ampere, Turing, Volta, Pascal, Kepler, Maxwell, Fermi, Tesla, and Jetson GPUs.
IKM® Neural Networks Performance
+30B
IKM® Analysis per Second per Video Stream
+99%
IKM® GPU Load
is 99,1%.
+250
IKM® Neural Networks per Video Stream
99%
IKM® MIG Each Instance is 99%
+1,5B
IKM® Analysis
per Frame
IKM® Neural Networks Uniqueness
∞
Cameras Managed Simultaneously
∞
Events Extracted Simultaneously
∞
Video Streams Managed Simultaneously
∞
Video Streams Analyzed Simultaneously
∞
12
Milliseconds Average
Time for One frame
IKM® Neural Networks Engines
with a DGX H100
+3K
Video Streams
Analyzed
+147B
Pixels Processed
per Second
+300K
Predictions
per Second
+75K
Decoded Frames
per Second
+28K
Coded Frames
per Second
IKM® Neural Networks Video GenAI & LLM
∞
Information Managed by IKM® Video GenAI Solutions
∞
Information Generated by IKM® Video GenAI Solutions
∞
Personalized
Solutions
∞
Information Managed by IKM® LLM Solutions
∞
Information Provided by IKM® LLM Solutions
IKM® Neural Networks Latency
with a DGX H100
< 1
Millisecond Inference on Simple Neural Networks
< 15
Milliseconds Inference on Average Neural Networks
< 50
Milliseconds of Latency to Process a Frame
< 200
Milliseconds of Average Latency per Event
+7M
Metadata Extracted per Second
Our Icons
Activate existing video streams.
IKM® Neural Networks
are dedicated to the task of
analyzing video streams.
Realized through proprietary development and articulation of different algorithmic sequences of neural networks, specialized in performing parallel computations to perform real-time extraction of the meaning present in a video stream.