11.5-Hour Continuous Multi-OS Hardware Burn-In Across Ryzen AI Max+ 395, Air-Gapped 20 Gbps USB4/Thunderbolt Peer-to-Peer, and Three Bare-Metal STM32H533RE Nodes: Empirical Throughput, Jitter, and Silicon Endurance Receipts

Short synthetic benchmarks routinely miss sustained thermal droop, flash SLC-cache folding, interactive scheduler stutter, and USB serial-driver lockups that surface only under multi-hour continuous load. This technical brief documents complete, unedited empirical telemetry from two consecutive continuous hardware endurance runs totaling 542 full-stack rounds over 11.5 hours on November 20, 2026 (3.5-hour overnight run of 151 rounds plus 8.0-hour daytime run of 391 rounds) across a heterogeneous dual-operating-system bench pairing a Windows 11 Pro AMD Ryzen AI Max+ 395 workstation (32 logical processors, 128 GB unified memory, 2 TB NVMe) with an air-gapped macOS companion node over link-local 169.254.0.0/16 20 Gbps USB4/Thunderbolt peer-to-peer and three bare-metal STM32H533RE microcontroller boards. Across all 542 rounds, the bench recorded zero faults, flat 22.80 GiB/s median 28-thread SHA-256+CRC32 throughput with zero thermal throttling, 1.77 ms median p95 user-interface scheduler jitter (0 of 391 rounds exceeding 10 ms) via LP 0..3 core isolation, 136.4 GiB of honest unbuffered Win32 FILE_FLAG_NO_BUFFERING NVMe verification (including two candidly reported SLC-cache fold dips with immediate recovery), 2.168 TiB of 4-lane Thunderbolt transfers at 8.05 Gbps median (9.31 Gbps peak) with 2,168 of 2,168 bit-exact SHA-256 lane matches, 139,294 SPI-to-ADC linearity steps (7.56 mV median mean error), 2,710 of 2,710 valid DHT11 checksums tracking the Kalaheo ambient diurnal curve (69.8 °F to 77.0 °F), and 1,084 of 1,084 lockup-free macOS USB-CDC hardware TRNG queries.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24411981
Primary Topic
Parallel Computing and Optimization Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

11.5-Hour Continuous Multi-OS Hardware Burn-In Across Ryzen AI Max+ 395, Air-Gapped 20 Gbps USB4/Thunderbolt Peer-to-Peer, and Three Bare-Metal STM32H533RE Nodes: Empirical Throughput, Jitter, and Silicon Endurance Receipts

Reid Kawane
Zenodo (CERN European Organization for Nuclear Research)
Parallel Computing and Optimization Techniques
article

11.5-Hour Continuous Multi-OS Hardware Burn-In Across Ryzen AI Max+ 395, Air-Gapped 20 Gbps USB4/Thunderbolt Peer-to-Peer, and Three Bare-Metal STM32H533RE Nodes: Empirical Throughput, Jitter, and Silicon Endurance Receipts

Reid Kawane
article en

Abstract

Short synthetic benchmarks routinely miss sustained thermal droop, flash SLC-cache folding, interactive scheduler stutter, and USB serial-driver lockups that surface only under multi-hour continuous load. This technical brief documents complete, unedited empirical telemetry from two consecutive continuous hardware endurance runs totaling 542 full-stack rounds over 11.5 hours on November 20, 2026 (3.5-hour overnight run of 151 rounds plus 8.0-hour daytime run of 391 rounds) across a heterogeneous dual-operating-system bench pairing a Windows 11 Pro AMD Ryzen AI Max+ 395 workstation (32 logical processors, 128 GB unified memory, 2 TB NVMe) with an air-gapped macOS companion node over link-local 169.254.0.0/16 20 Gbps USB4/Thunderbolt peer-to-peer and three bare-metal STM32H533RE microcontroller boards. Across all 542 rounds, the bench recorded zero faults, flat 22.80 GiB/s median 28-thread SHA-256+CRC32 throughput with zero thermal throttling, 1.77 ms median p95 user-interface scheduler jitter (0 of 391 rounds exceeding 10 ms) via LP 0..3 core isolation, 136.4 GiB of honest unbuffered Win32 FILE_FLAG_NO_BUFFERING NVMe verification (including two candidly reported SLC-cache fold dips with immediate recovery), 2.168 TiB of 4-lane Thunderbolt transfers at 8.05 Gbps median (9.31 Gbps peak) with 2,168 of 2,168 bit-exact SHA-256 lane matches, 139,294 SPI-to-ADC linearity steps (7.56 mV median mean error), 2,710 of 2,710 valid DHT11 checksums tracking the Kalaheo ambient diurnal curve (69.8 °F to 77.0 °F), and 1,084 of 1,084 lockup-free macOS USB-CDC hardware TRNG queries.

Zenodo (CERN European Organization for Nuclear Research)
Parallel Computing and Optimization Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon
11.5-Hour Continuous Multi-OS Hardware Burn-In Across Ryzen AI Max+ 395, Air-Gapped 20 Gbps USB4/Thunderbolt Peer-to-Peer, and Three Bare-Metal STM32H533RE Nodes: Empirical Throughput, Jitter, and Silicon Endurance Receipts — Reid Kawane · Zenodo (CERN European Organization for Nuclear Research) (2026) | TGRS Research Map | TGRS