# Jetson AGX Orin Unified Memory observation
# Date: 2026-08-14 KST
# Device: NVIDIA Jetson AGX Orin Developer Kit
# L4T: R36.5.0
# JetPack: 6.2.2+b24
# CUDA toolkit/runtime/driver API: 12.6
# Power mode: MODE_30W (unchanged)
# Repository: Bias92/Bias92.github.io main @ b88fc745264bb4c19708028de129739a28fcb672
# Probe source preserved byte-for-byte at static/code/cuda-04/orin_um_probe.cu

$ nvcc -O2 -arch=sm_87 -std=c++17 /tmp/cuda04-orin-probe/orin_um_probe.cu -o /tmp/cuda04-orin-probe/orin_um_probe
$ /tmp/cuda04-orin-probe/orin_um_probe 0
CUDART_VERSION=12060
runtimeVersion=12060 driverVersion=12060 deviceCount=1
device=0 name=Orin cc=8.7 integrated=1
selectedDevice=0 name=Orin cc=8.7 integrated=1
managedMemory=1
concurrentManagedAccess=0
pageableMemoryAccess=0
usesHostPageTables=0
directManagedMemAccessFromHost=0
hostRegisterSupported=1

$ nvcc -O2 -arch=sm_87 -std=c++17 static/code/cuda-04/managed_add.cu -o /tmp/cuda04-orin-probe/managed_add
$ CUDA_VISIBLE_DEVICES=0 /tmp/cuda04-orin-probe/managed_add
before kernel: 41
after kernel:  42

# Interpretation boundary
# - concurrentManagedAccess=0 classifies this environment as limited Unified Memory.
# - pageableMemoryAccess=0 means system allocations are not in the full Unified Memory model.
# - usesHostPageTables=0 is not interpreted after the classifier stops at the limited branch.
# - 41 -> 42 confirms visibility after synchronization only.
# - No page-fault, migration, cache-maintenance, prefetch, or performance measurement was made.