Observed arrival · 2026-09-10
Nameplate Analytics Measures the GPU Work Hidden Inside the Bill
A focused AWS-and-Kubernetes measurement practice that compares GPU telemetry with amortized cloud costs to identify spend that produced little or no work.
Field notes
The measurement begins with per-second DCGM profiling attached to a Kubernetes workload, then joins those observations to hourly AWS Cost and Usage Report data through provider and resource IDs. The example uses a Tesla T4 on g4dn.xlarge, with 242 samples collected at 1 Hz. The site distinguishes reported kernel residency from multiprocessor activity and says the resulting analysis can be organized by GPU resource, team, namespace, and model before remediation is ordered by dollars and disruption.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue