Products
Atlas Edge AI Server Powered by NVIDIA IGX Thor™
- Powered by NVIDIA IGX Thor™ with 2070 TFLOPS (FP4-Sparse) of built-in compute, expandable up to 5581 TFLOPS (FP4-Sparse) with an optional NVIDIA RTX PRO™ 6000 Blackwell Max-Q Workstation Edition
- Powered by NVIDIA Blackwell GPU architecture, delivering exceptional performance in generative reasoning and multimodal sensor processing
- First rack-mountable, safety-capable infrastructure, designed for the next generation of industrial, robotics, medical, and security edge applications
#Inside-Out Functional Safety & Outside-In Proactive AI Safety
#Scalable Edge AI Infrastructure
#10 Years of NVIDIA Enterprise Support
Description /
Edge AI Atlas Series Application Scenario
An integrated computing framework driving the future of real-world AI

Physical AI Safety
Proactive Safety in Smart Manufacturing : Inside-Out and Outside-In Safety

In a smart factory, Cobots powered by Atlas T7 can safely work alongside people by combining inside-out and outside-in safety. The robotic arm is designed to stop or halt the machine's actions if it comes into contact with a person or object, while an external infrastructure sensors provide an additional outside-in perspective to proactively identify approaching personnel or potential hazards. During operation, the system can detect potential safety risks and trigger an appropriate response, such as slowing down or pausing the robot, to help prevent unsafe interactions.
Specification /
|
|
Atlas Edge AI Server |
|---|---|
|
GPU |
Integrated GPU :
|
|
CPU |
14-core Arm® Neoverse®-V3AE 64-bit CPU 64 KB I-cache, 64 KB dCache 1 MB L2 cache per core 16 MB Shared System L3 Cache |
|
AI PERFORMANCE |
2070 ~ 5581 TFLOPS (FP4-Sparse) |
|
STORAGE |
1TB SSD |
|
RAM |
128GB 256-bit LPDDR5X |
|
Dimension |
2U form factor |
|
I/O |
2x Gen5 PCIe (x8, x16) |
|
DISPLAY INTERFACE |
1x VESA DisplayPort 1.4a |
|
ETHERNET |
2x RJ45 (1GbE each) |
|
Power Supply |
1000 W |
For detailed product information, please fill out the form on our Contact Us page.