AMD: Real-time control and reliability are the key to physical AI
AMD executive Salil Raje proposed that performance evaluation for physical AI should shift from raw compute power to real-time control and system-level reliability. AMD builds heterogeneous architectures with Kria AI SOM, Versal, Zynq, and Spartan FPGAs; in third-party tests, Kria AI SOM achieved inference times under 0.1 seconds and supported 8,000 control decisions per second; in open navigation benchmarks, the Ryzen AI Embedded configuration delivered control-timing reliability up to 3.4 times that of Nvidia Jetson T5000, with 1.6 times more available compute headroom.
Body · AI translation
Go
The body content of the article will be changed to this letter size.
Scroll position bar
Salil Raje, senior vice president and general manager of AMD’s Adaptive and Embedded Computing Group. (Photo: AMD)
AMD has called for a shift in how physical AI performance is evaluated, arguing that real-time control and system-wide reliability should take precedence over raw AI computing power.
“What robot manufacturers demand most is not the highest AI computing performance,” said Salil Raje, senior vice president and general manager of AMD’s Adaptive and Embedded Computing Group (AECG), in an official blog post on Oct. 19.
“What matters is combining advanced AI models with real-time control capabilities that precisely execute computational results within required time limits.”
Physical AI refers to technologies that enable machines such as robots and autonomous vehicles to interact with the physical world and perform physical actions.
Key performance criteria include rapid response times, predictability, power efficiency and functional safety. Delayed or unstable motion control can lead to collisions and other safety incidents.
Real-time control in physical AI requires minimizing data transfer delays and bottlenecks. AMD proposes a heterogeneous semiconductor architecture, comparing the system’s structure to the human body.
Under this approach, the robot’s brain uses a Kria AI system-on-module (SOM) based on Ryzen AI Embedded X100 series processors. Versal Adaptive system-on-chips (SoCs) serve as the spine, Zynq SoCs control the joints, and Spartan field-programmable gate arrays (FPGAs) handle sensor functions.
The architecture forms the basis of AMD’s Kria AI solutions for robotics.
Compared with a single processor integrating multiple functions, the heterogeneous approach offers advantages in memory bandwidth, power efficiency and latency. It also allows the system to maintain a stable state even if the primary AI computing unit stops operating.
In third-party testing cited by AMD, the Kria AI SOM achieved AI inference times of less than 0.1 seconds and supported 8,000 control decisions per second.
In open navigation benchmark tests, a Ryzen AI Embedded configuration delivered up to 3.4 times greater control timing reliability and up to 1.6 times more available computing headroom than Nvidia’s Jetson T5000.
According to Mimic analysis, the configuration also supported up to 2.3 times as many simultaneously operating autonomous agents.
Foundation Robotics and Generative Bionics are designing products using AMD’s Kria AI solutions.
JEONG IL JOO 1week@thelec.kr
My comment