Platforms · NVIDIA
Jetson AGX Orin
NVIDIA's established embedded AI platform widely adopted in robotics which require edge AI such as detection, segmentation, and reinforcement learning. The Orin is the current workhorse of many production robotics deployments.
Max Power at 65 W
- CPU available
- 6.4%
- 0.8 of 12 cores free
- Missions completed
- 3
- in 900 s
- Control loop misses
- 5.65/s
- against a 30 Hz target
- VLM queries completed
- 0 of 71
- scene-understanding queries
Max Power at 65 W
- CPU available
- 6.4%
- 0.8 of 12 cores free
- Missions completed
- 3
- in 900 s
- Control loop misses
- 5.65/s
- against a 30 Hz target
- VLM queries completed
- 0 of 71
- scene-understanding queries
CPU
Total load, its distribution across cores, and the clock behind it.
#CPU utilization
Total CPU utilization across all cores, sampled once a second for the length of the run.
Data table (4 rows)
| Metric | Mean | p50 | p95 | Max |
|---|---|---|---|---|
| CPU total (%) | 93.62 | 95.80 | 97.20 | 98.00 |
| Peak core (%) | 97.44 | 98.00 | 100.00 | 100.00 |
| Mean core (%) | 93.62 | 95.80 | 97.20 | 97.90 |
| Active cores (%) | 99.89 | 100.00 | 100.00 | 100.00 |
Source: max_power/jetson_orin · CSV
Supporting charts
Per-core utilization over time
#Per-core utilization over time
One row per core. Persistent dark bands are cores that stay pinned for the whole run.
Data table (12 rows)
| Core | Mean (%) |
|---|---|
| 0 | 93.6 |
| 1 | 93.3 |
| 2 | 93.2 |
| 3 | 93.4 |
| 4 | 93.8 |
| 5 | 93.8 |
| 6 | 93.8 |
| 7 | 93.7 |
| 8 | 93.6 |
| 9 | 93.9 |
| 10 | 93.7 |
| 11 | 93.7 |
Source: max_power/jetson_orin · CSV
Per-core utilization distribution
#Per-core utilization distribution
Every per-core sample in the run, bucketed into 5% bands.
Data table (20 rows)
| Utilization band (%) | Samples |
|---|---|
| 0–5 | 12 |
| 5–10 | 0 |
| 10–15 | 0 |
| 15–20 | 0 |
| 20–25 | 0 |
| 25–30 | 0 |
| 30–35 | 0 |
| 35–40 | 0 |
| 40–45 | 2 |
| 45–50 | 3 |
| 50–55 | 4 |
| 55–60 | 20 |
| 60–65 | 49 |
| 65–70 | 110 |
| 70–75 | 221 |
| 75–80 | 303 |
| 80–85 | 283 |
| 85–90 | 295 |
| 90–95 | 2958 |
| 95–100 | 6516 |
Source: max_power/jetson_orin · CSV
CPU clock frequency
#CPU clock frequency
Sustained CPU clock frequency under load.
Data table (1 rows)
| Metric | Mean | p50 | p95 | Max |
|---|---|---|---|---|
| CPU frequency (MHz) | 2,201.29 | 2,201.60 | 2,201.60 | 2,201.60 |
Source: max_power/jetson_orin · CSV
GPU and memory
Accelerator load from the VLM workload, and the memory it leaves behind.
#GPU utilization
GPU utilization while the VLM issues queries continuously.
Data table (3 rows)
| Metric | Mean | p50 | p95 | Max |
|---|---|---|---|---|
| GPU utilization (%) | 14.25 | 0.00 | 100.00 | 100.00 |
| GPU clock (MHz) | 428.69 | 306.00 | 1,300.50 | 1,300.50 |
| Effective throughput (GHz) | 15.98 | 0.00 | 130.05 | 130.05 |
Source: max_power/jetson_orin · CSV
Supporting charts
System memory usage
#System memory usage
System memory in use across the run, as a share of the total fitted.
Data table (3 rows)
| Metric | Mean | p50 | p95 | Max |
|---|---|---|---|---|
| RAM (%) | 78.73 | 79.00 | 80.00 | 81.80 |
| RAM used (MB) | 48,834.77 | 49,011.55 | 49,619.63 | 50,740.50 |
| Swap (%) | 1.04 | 1.00 | 1.10 | 1.10 |
Source: max_power/jetson_orin · CSV
Power and thermal
Board draw and the thermal margin left underneath it.
#Board power
Board power draw across the run, against the configured TDP.
Data table (1 rows)
| Metric | Mean | p50 | p95 | Max |
|---|---|---|---|---|
| Board power (W) | 31.34 | 28.55 | 57.09 | 64.50 |
Source: max_power/jetson_orin · CSV
Supporting charts
CPU and GPU temperature
#CPU and GPU temperature
CPU and GPU package temperature under sustained load, against a 100 °C throttle point.
Data table (2 rows)
| Metric | Mean | p50 | p95 | Max |
|---|---|---|---|---|
| CPU temperature (°C) | 59.71 | 60.50 | 61.80 | 63.10 |
| GPU temperature (°C) | 55.27 | 54.65 | 60.54 | 61.40 |
Source: max_power/jetson_orin · CSV
Clock against temperature
#Clock against temperature
A clock that falls as temperature climbs is thermal throttling; a flat clock means the platform held its performance for the whole run.
Data table (2 rows)
| Metric | Mean | p50 | p95 | Max |
|---|---|---|---|---|
| CPU temperature (°C) | 59.71 | 60.50 | 61.80 | 63.10 |
| CPU clock (MHz) | 2,201.29 | 2,201.60 | 2,201.60 | 2,201.60 |
Source: max_power/jetson_orin · CSV
Compute delivered per watt
#Compute delivered per watt
CPU utilization divided by instantaneous board power, showing how efficiency moves as the workload shifts between planning and perception.
Data table (3 rows)
| Metric | Value |
|---|---|
| Mean available compute per watt (GHz-cores/W) | 0.054 |
| Mean GPU utilization per watt (%/W) | 0.455 |
| Mean board power (W) | 31.3 |
Source: max_power/jetson_orin · CSV
System behaviour
Stability, I/O, and process load, the context around the headline counters.
#CPU against GPU utilization
Each point is one second of the run. A tight cluster means the two resources move together; a wide spread means one is waiting on the other.
Data table (2 rows)
| Metric | Mean | p50 | p95 | Max |
|---|---|---|---|---|
| CPU total (%) | 93.62 | 95.80 | 97.20 | 98.00 |
| GPU utilization (%) | 14.25 | 0.00 | 100.00 | 100.00 |
Source: max_power/jetson_orin · CSV
Supporting charts
Rolling stability
#Rolling stability
Standard deviation over a 30-sample window. Spikes are moments the load became bursty rather than steady.
Data table (2 rows)
| Series | Window (samples) |
|---|---|
| CPU total | 30 |
| GPU utilization | 30 |
Source: max_power/jetson_orin · CSV
Network throughput
#Network throughput
Sensor data arriving over the wired DDS link from the simulation machine. Every platform is offered the same load.
Data table (3 rows)
| Metric | Mean | p50 | p95 | Max |
|---|---|---|---|---|
| Network received (MB/s) | 66.43 | 66.40 | 67.60 | 71.60 |
| Network sent (MB/s) | 0.28 | 0.30 | 0.40 | 0.50 |
| Network errors | 0.00 | 0.00 | 0.00 | 0.00 |
Source: max_power/jetson_orin · CSV
Disk throughput
#Disk throughput
Disk read and write throughput across the run.
Data table (2 rows)
| Metric | Mean | p50 | p95 | Max |
|---|---|---|---|---|
| Disk read (MB/s) | 0.93 | 0.00 | 0.10 | 769.90 |
| Disk write (MB/s) | 0.23 | 0.00 | 0.30 | 68.30 |
Source: max_power/jetson_orin · CSV
Load average
#Load average
Run-queue depth. A load average well above the core count means threads are waiting for a core rather than running on one.
Data table (3 rows)
| Metric | Mean | p50 | p95 | Max |
|---|---|---|---|---|
| Load average (1m) | 23.06 | 23.76 | 26.86 | 27.81 |
| Load average (5m) | 17.44 | 19.31 | 23.04 | 23.73 |
| Load average (15m) | 13.50 | 13.86 | 17.57 | 17.97 |
Source: max_power/jetson_orin · CSV
Process count
#Process count
Total processes on the machine while the benchmark runs.
Data table (1 rows)
| Metric | Mean | p50 | p95 | Max |
|---|---|---|---|---|
| Process count | 401.83 | 402.00 | 403.00 | 404.00 |
Source: max_power/jetson_orin · CSV
All metrics for Max Power
| Metric | NVIDIA Jetson Orin | ||
|---|---|---|---|
| Mean | p95 | Max | |
| CPU Total (%) | 93.62 | 97.20 | 98.00 |
| CPU Frequency (MHz) | 2,201.29 | 2,201.60 | 2,201.60 |
| Peak Core Utilization (%) | 97.44 | 100.00 | 100.00 |
| Mean Core Utilization (%) | 93.62 | 97.20 | 97.90 |
| Active Cores (%) | 99.89 | 100.00 | 100.00 |
| RAM Usage (%) | 78.73 | 80.00 | 81.80 |
| RAM Used (MB) | 48,834.77 | 49,619.63 | 50,740.50 |
| RAM Total (MB) | 62,828.10 | 62,828.10 | 62,828.10 |
| Swap Usage (%) | 1.04 | 1.10 | 1.10 |
| GPU Utilization (%) | 14.25 | 100.00 | 100.00 |
| GPU Clock (MHz) | 428.69 | 1,300.50 | 1,300.50 |
| GPU Effective Throughput (GHz) | 15.98 | 130.05 | 130.05 |
| Board Power (W) | 31.34 | 57.09 | 64.50 |
| CPU Temperature (°C) | 59.71 | 61.80 | 63.10 |
| GPU Temperature (°C) | 55.27 | 60.54 | 61.40 |
| Disk Read (MB/s) | 0.93 | 0.10 | 769.90 |
| Disk Write (MB/s) | 0.23 | 0.30 | 68.30 |
| Disk Usage (%) | 28.40 | 28.40 | 28.40 |
| Net Recv (MB/s) | 66.43 | 67.60 | 71.60 |
| Net Sent (MB/s) | 0.28 | 0.40 | 0.50 |
| Network Errors | 0.00 | 0.00 | 0.00 |
| Load Average (1m) | 23.06 | 26.86 | 27.81 |
| Load Average (5m) | 17.44 | 23.04 | 23.73 |
| Load Average (15m) | 13.50 | 17.57 | 17.97 |
| Process Count | 401.83 | 403.00 | 404.00 |
Raw samples: CSV · max_power/jetson_orin
Verdict
Significantly behind newer platforms like the Thor or Strix Halo. The Orin AGX was unable to process simulatneous navigation and AI workload in real time. It remains the current workhorse of many production robotics deployments, but not for a stack that adds a modern physical AI workloads.
Strong
- The established embedded AI platform, widely adopted and well understood in production robotics.
- The lowest cost and power platform.
- Fine for the detection, segmentation, and reinforcement-learning workloads it was designed around.
Weakness
- Mean CPU utitilization near 100% saturated.
- Unable to complete 70% of missions in the allotted time due to resource saturation.
- Zero VLM queries were successfully processed in the mission's execution time.
- The highest control-loop miss rate.
Specification
The vendor's numbers on the left; what this platform's sensor drivers actually cost it, measured on real hardware, on the right.
| CPU | 12x Arm Cortex-A78AE @ 2.2 GHz |
|---|---|
| GPU architecture | Ampere |
| GPU cores | 2048 CUDA + 64 Tensor |
| Memory bandwidth | 204.8 GB/s |
| NPU / DLA | 2x NVDLA 2.0 |
| RAM | 64 GB LPDDR5 |
| Power (TDP) | 15-60 W |
| Product page | NVIDIA |
| 3D LiDAR | 0.670 |
|---|---|
| 2D safety LiDAR | 0.120 |
| RGBD camera | 1.550 |
| Total for the robot's sensor suite | 6.90 cores |
Setup instructions for this platform are in the platform setup guide.
Results by TDP or Configuration
Each category fixes a TDP power budget or optimized configuration and runs the workload on every platform. Open one to see how this platform behaved under those conditions, and how it compared against the others.