Engineering Journal
The Playbook
First-principles engineering guides.
Acoustic Leq and A-Weighting: Measuring How Loud a GPU Really Is
How a Premium Assurd certificate turns one calibrated recording into an A-weighted equivalent sound level under load: decibels from first principles, the A-weighting curve, the 94 dB calibrator tone, the 50 cm geometry, the 6 dB noise-floor rule and tonal peaks.
Certificate Signatures: How to Verify an Assurd Certificate Yourself
What exactly an Assurd certificate signature covers, why the certificate is signed as canonical JSON bytes, how the ES256 key in Cloud KMS signs it, and how to check one yourself: in the browser, or offline with OpenSSL and no trust in this website.
Fleet Statistics: Cohorts, Percentiles and the n ≥ 30 Rule
How an Assurd certificate compares a card with the other cards Assurd has certified: where the fleet data comes from, which cards count as a cohort, why nothing is ranked below 30 samples, how a percentile is computed from a z-score, and what clock per watt and run-to-run variation mean.
Frame Pacing from First Principles: Consistency Gap, Stutter Magnitude and the Fluidity Score
How an Assurd certificate turns a benchmark's frame times into three numbers — consistency gap, stutter magnitude and the Fluidity Score — with every definition, a worked 120 / 90 / 60 FPS example, and the limits of what the numbers can tell you.
Hotspot Delta and TIM Pump-Out: Reading the Thermal Interface from Two Sensors
Why the gap between a GPU's hotspot and core temperatures is the best telemetry signal for thermal interface health, how an Assurd certificate computes it over the steady-state window of the 1800 s trace, and what it can and cannot tell you.
Lifetime ECC Double-Bit Errors: What nvidia-smi Can Tell You About a GPU's Memory
Cards with ECC memory count the memory errors they correct and the ones they cannot. How error-correcting codes work, the difference between volatile and lifetime counters, how an Assurd lab reads them from an nvidia-smi report, the 100-error Hard Limit, and why most cards report nothing at all.
Rail-to-Ground Resistance: What a Multimeter Tells You About a GPU's Power Stages
Before an Assurd lab applies power to a GPU, it reads each power rail's resistance to ground with a bench multimeter. What that 2-wire reading measures, why the leads are zeroed and the reading is left to settle, how the dead-short Hard Limit and the review floors work, and what a 0.1 Ω meter cannot see.
The Thermal Time Constant and the 1800 s Trace: Why a GPU Thermal Test Runs for 30 Minutes
A first-order model of how a GPU cooler heats up, what fraction of its final temperature a test of a given length actually sees, how an Assurd certificate finds the start of load, and how the peak clock, VRAM and VRM figures are taken from the 1800 s trace.
VBIOS SPI Dumps, Hashing and Silicon Identity: What a Registry Match Proves
In Phase 2 an Assurd lab reads a GPU's firmware straight off its flash chip, and the server hashes it and looks it up in a registry of factory images. What the SPI read gives you, what a SHA-256 match proves, why the lookup is keyed by board, the three possible verdicts, and what a hash cannot detect.
Silicon DNA: The Physics of Clock States, Voltage and the V/F Curve
Why a GPU needs more voltage to run faster, why two cards of the same model boost differently, how heat and current slowly wear out on-chip wiring, and how the voltage regulator feeds the die. The physics behind clock and power figures, and which of them an Assurd certificate actually records.