Component Runtime Profile#

This page summarizes platform details and runtime resource usage for components on the IGX Thor reference platform.

Note

The runtime profile data reported in this section is representative of the Control of safety function on a forklift (ATL) use case running with a 3-camera RTSP input stream.

Platform Details#

Field	IGX-Thor
Architecture	aarch64
CPU vendor	ARM
CPU model	Arm CPU on IGX-Thor
CPU op-mode	64-bit
Physical cores	14 cores per cluster
Logical CPUs	14
Threads per core	1
Socket / cluster count	1 socket
CPU max frequency	2601 MHz
CPU min frequency	54 MHz
NUMA nodes	1

GPU Details#

Field	IGX-Thor
GPU model	NVIDIA Thor iGPU (Blackwell)
Driver version	580.00
CUDA version	13.0

Runtime Usage Summary#

Component nvpsd_gateway#

Field	IGX-Thor
Process CPU avg / peak	0.00% / 0.00%
Average memory usage, RSS	3.9 MB
Peak memory usage, VmHWM	3.9 MB
GPU utilization avg	NA
GPU kernel time / wall time	NA
SMs active avg / peak	NA
GPU memory usage avg / peak	NA
NVDEC utilization	NA

Component nvpss_daemon#

Field	IGX-Thor
Process CPU avg / peak	2.36% / 2.40%
Average memory usage, RSS	4.2 MB
Peak memory usage, VmHWM	4.2 MB
GPU utilization avg	NA
GPU kernel time / wall time	NA
SMs active avg / peak	NA
GPU memory usage avg / peak	NA
NVDEC utilization	NA

Component mdx_client#

Field	IGX-Thor
Process CPU avg / peak	2.22% / 2.30%
Average memory usage, RSS	97.8 MB
Peak memory usage, VmHWM	121.1 MB
GPU utilization avg	NA
GPU kernel time / wall time	NA
SMs active avg / peak	NA
GPU memory usage avg / peak	NA
NVDEC utilization	NA

Component atl_sdm#

Field	IGX-Thor
Process CPU avg / peak	0.00% / 0.00%
Average memory usage, RSS	2.1 MB
Peak memory usage, VmHWM	2.1 MB
GPU utilization avg	NA
GPU kernel time / wall time	NA
SMs active avg / peak	NA
GPU memory usage avg / peak	NA
NVDEC utilization	NA

Component safety_monitor#

Field	IGX-Thor
Process CPU avg / peak	2.64% / 3.90%
Average memory usage, RSS	124.8 MB
Peak memory usage, VmHWM	124.8 MB
GPU utilization avg	32.31%
GPU kernel time / wall time	~32.31%
SMs active avg / peak	~32.31% / 91.00%
GPU memory usage avg / peak	224.0 MB / 224.0 MB
NVDEC utilization	15.74%

Usage Summary Field Notes#

Process CPU avg / peak: Average and peak CPU usage of the profiled process during the profiling run.
Normalized CPU avg: Average process CPU usage divided by the number of logical CPUs visible to the operating system.
Average memory usage, RSS: Average physical RAM used by the profiled process during the profiling run.
Peak memory usage, VmHWM: Maximum physical RAM used by the profiled process during the profiling run.
GPU utilization avg: Average whole-device GPU utilization from nvidia-smi live samples. For safety_monitor, this is not a per-process value; it includes any other GPU tenants (e.g., NVIDIA DeepStream / NVIDIA Metropolis) running on the same GPU during the profiling run.
GPU kernel time / wall time: Live estimate derived from the whole-device GPU utilization sample. For safety_monitor, this is not attributed to the safety_monitor PID. Use nsys profile -p <pid> for per-process CUDA kernel timeline values.
SMs active avg / peak: Live estimate derived from the whole-device GPU utilization sample. For safety_monitor, this is not attributed to the safety_monitor PID. Use nsys GPU_METRICS for per-process SMs Active values.
GPU memory usage avg / peak: Average and peak CUDA compute-app memory attributed to the profiled PID via nvidia-smi --query-compute-apps. On Tegra integrated GPUs, this represents system memory used by that process’s GPU context, not dedicated GPU frame-buffer memory.
NVDEC utilization: Average whole-device video decoder engine utilization from nvidia-smi live samples. For safety_monitor, this is not a per-process value; it includes any other process using NVDEC on the same GPU during the profiling run.
NA: Metric is not applicable to that component or was not collected.

Note

A reported value of 0.00% / 0.00% for Process CPU avg / peak (for example nvpsd_gateway and atl_sdm) should be read as below this profiler’s measurement resolution during the profiling window, not as proof that the component was inactive. These components are event-driven and may spend most of the run waiting for safety events, with CPU activity occurring only in short bursts when messages arrive.

Metric Attribution Scope#

The GPU metrics reported for safety_monitor have two attribution scopes. Read each row according to its label.

Per-process (attributed to the safety_monitor PID):

GPU memory usage avg / peak

On Tegra integrated GPUs, this value represents system memory used by the process’s GPU context. It is the only GPU row in this table that is attributed to safety_monitor rather than to the whole device.

Whole-device (entire GPU or engine, summed across all processes):

GPU utilization avg
GPU kernel time / wall time
SMs active avg / peak
NVDEC utilization

When safety_monitor shares the GPU with another CUDA or NVDEC client (for example, NVIDIA DeepStream or NVIDIA Metropolis), the whole-device rows report the combined load on the device. They must not be interpreted as the safety_monitor process’s individual footprint.