NVIDIA Jetson AGX Thor is the latest generation of platforms for edge AI and advanced robotics. These environments demand deterministic communication, fast cycles and perfect synchronization between sensors, controllers and actuators.
As an official acontis partner, ISIT-Neperis helps manufacturers integrate EtherCAT using the EC-Master stack, known for its efficiency, low CPU footprint and real-time capabilities. This article looks at how acontis’ EC-Master stack reaches different levels of EtherCAT real-time performance on the NVIDIA Jetson AGX Thor, through three increasing optimization steps within a Linux environment.
Context and test environment
The Jetson AGX Thor ships with Linux via JetPack 7.0 and already includes an SMP PREEMPT kernel — the foundation needed for real-time. To evaluate EtherCAT performance, acontis connected the platform to a set of Beckhoff terminals (DO, DI, AO, EK1100, EK1110) and used the EcMasterDemoDc test application, which integrates Distributed Clocks and measurement tools (the -dcmlog option).
In this context, two parameters are the key indicators of real-time quality:
- CPU Load — the time EC-Master needs to run the EtherCAT stack;
- Jitter — the cycle-to-cycle timing variation, decisive for precise motion control.
Stage 1 — EtherCAT with minimal effort
Here the Jetson uses the stock Linux kernel shipped with JetPack, the generic SOCK_RAW driver, and no CPU or system tuning.
Results: excellent CPU Load (~1%), but very high jitter — ±330 µs, i.e. 33% of a 1 ms cycle.
Verdict: communication works, but the timing variation is too large for a demanding robotics application. This level can suit slower cycles (≥5 ms) or systems that are not jitter-sensitive.
Stage 2 — EtherCAT with high efficiency
Optimizations added:
- full RT kernel installed for Jetson Thor;
- CPU 3 dedicated to the EtherCAT stack (
isolcpus); - all CPU cores set to performance mode (scaling governor);
- EC-Master pinned to that CPU via
-a 3.
Results: jitter reduced to ±125 µs; CPU Load unchanged (~1%).
Verdict: the system becomes more stable, but still not enough for strict Distributed-Clocks synchronization or fast cycles (<1 ms).
Stage 3 — EtherCAT with the highest performance
To reach true real-time performance:
- disable the stock Linux Realtek driver;
- load the acontis real-time driver (
atemsys+ RTL driver); - use it via the
-rtl8169option.
Results: extremely low jitter (±6 µs), CPU Load of 1.38% for a 1 ms cycle, and perfect stability (no spikes in the Distributed-Clocks graphs).
Fast-cycle test: at a 250 µs cycle the system stays stable, with a CPU Load of just 5.4% — excellent for such a frequency.
Verdict: this configuration reaches the high cycle rates — beyond 2.5 kHz, down to a 250 µs (4 kHz) cycle — required by modern robots (complex kinematics, cobots, AGVs, fast manipulators).
Results at a glance
| Stage | Key setup | Jitter | CPU Load |
|---|---|---|---|
| 1 · Minimal effort | Stock kernel, SOCK_RAW driver, no tuning | ±330 µs | ~1% |
| 2 · High efficiency | RT kernel, CPU isolation, performance governor | ±125 µs | ~1% |
| 3 · Highest performance | acontis real-time network driver (atemsys) | ±6 µs | 1.38% @ 1 ms / 5.4% @ 250 µs |
Conclusion
With acontis’ solutions and ISIT-Neperis’s expertise, the Jetson AGX Thor can become a high-performance EtherCAT controller able to meet the strict requirements of real-time robotics:
- a quick start with no tuning;
- improved performance via the RT kernel;
- maximum industrial real-time thanks to the acontis driver.
For any EtherCAT integration, real-time optimization or support on NVIDIA Jetson, ISIT-Neperis is your dedicated technical partner. We also provide certifiable Safety-over-EtherCAT (FSoE) and other industrial communication stacks that build on the same EtherCAT foundation.
Talk to our team about your EtherCAT-on-Jetson project.
Adapted and translated into English from the ISIT article “Intégration d’EtherCAT sur NVIDIA Jetson AGX Thor” (isit.fr), based on the original technical write-up by acontis. Benchmark figures are acontis’ measurements.