K/TKYLE TECH TESTINGStart a conversation ↗

FIELD REPORT / 001 · OCTOBER 2026

RTX 4070 12GB
Windows 11 · ComfyUI

Idle baseline and initial mixed-workload telemetry evaluation. The report and source data distinguish what the sensor recorded from what the operator reported.

INITIAL TELEMETRY EVALUATIONOwner-owned hardware · no manufacturer involvementTwo 5-minute captures · 118 total samples

THE RESULT, IN SCOPE

More activity in the
mixed trial. Nothing more assumed.

The mixed capture contains much higher sampled GPU activity, memory use, temperature, and GPU-reported power than the idle-labeled baseline. Its 59 observations span about five minutes and include non-generation intervals.

64°CSampled temperature peak
100%Sampled utilization peak
11,571 MiBSampled VRAM peak
184.81 WGPU-reported power peak

Peak readings are individual samples. GPU-reported power is not wall power; the telemetry does not measure image speed, output quality, or sustained stability.

INTERACTIVE TELEMETRY

Inspect the recorded series.

Data loads from the exact public CSV copies linked below. Select a metric and capture; use a pointer or the arrow keys to inspect samples and timestamps.

Loading captured samples…
Mixed ComfyUI trialIdle baselineTEMPERATURE / °C

Capture elapsed time starts at zero separately for each file. Means in the table are arithmetic sample means across the complete capture, not time-weighted or generation-only.

FULL-CAPTURE STATISTICS

Both windows, side by side.

Each CSV contains 59 valid samples for all four measures. The baseline is owner-labeled and includes brief GPU activity, so it was not completely inactive.

MetricIdle meanIdle peakMixed meanMixed peak
Temperature30.66 °C32 °C42.44 °C64 °C
GPU utilization1.03%35%33.22%100%
VRAM used1,687.46 MiB1,815 MiB7,540.88 MiB11,571 MiB
GPU-reported power9.51 W23.64 W54.84 W184.81 W

WHAT WAS — AND WAS NOT — ESTABLISHED

Keep the boundaries
in view.

A useful technical report does not turn a handful of sensor values into claims they cannot support.

✓

Recorded in telemetry

UTC timestamps, GPU name, driver version, temperature, utilization, used/total VRAM and driver-reported GPU power. Both captures passed the published data checks.

~

Reported by the operator

Two ERNIE Image Turbo image generations were reported complete during the mixed trial. No screenshot, output or app-history record was supplied, so completion is not independently verified.

…

Still unavailable

Generation times, workflow/model hashes, output evidence, resolution, ambient conditions, clock/fan behavior, CPU/RAM load, wall power, repeat-run variation and sustained behavior.

VERDICT

The evidence supports an initial short mixed-session telemetry evaluation. It does not establish throughput, thermal equilibrium, long-term stability, or a completed endurance benchmark.

STATIC SOURCE-DERIVED CHARTS

Readable without interaction.

Dark-theme charts are regenerated from the preserved CSVs; raw files and the integrity manifest remain available above.

Recorded temperature samples from the separate idle baseline and mixed ComfyUI capture, charted against elapsed seconds with a labeled Celsius axis.
Temperature over elapsed capture time
Recorded GPU utilization samples from the separate idle baseline and mixed ComfyUI capture, charted against elapsed seconds with a labeled percent axis.
GPU utilization over elapsed capture time
Recorded used VRAM samples from the separate idle baseline and mixed ComfyUI capture, charted against elapsed seconds with a labeled MiB axis.
VRAM use over elapsed capture time
Recorded GPU-reported power samples from the separate idle baseline and mixed ComfyUI capture, charted against elapsed seconds with a labeled watt axis.
GPU-reported power over elapsed capture time