Read-only inference telemetry

spoon-ai01 / Ollama

Model, generation, GPU and host status — no controls, no inference probes.

Connecting Activity unknown

Overview

Decode

—

Awaiting request metrics

Prefill

—

Awaiting prompt metrics

Active requests

No live task observed

Context

—

Awaiting context metrics

GPU utilisation

—

NVIDIA compute activity

GPU power

—

Awaiting board power

VRAM in use

—

Awaiting allocation

Host memory

—

Awaiting host memory

Activity

Request and GPU activity

Normal workloads only · every completed request is marked

Building local history

Requests running0 active

GPU power—

Running requestsGPU powerRequest completed

Recent events

Newest first · state changes only

Health

—Waiting for state changes…Info

Selected model

—

No loaded model reported

Model ID: —
Parameters: —
Family: —
Format: —
Quantisation: —
Allocated context: —
Model size: —
VRAM allocation: —
Residency: —

Recent completed requests

Ollama journal · metrics only

Decode median—

Decode p95—

Prefill median—

Latency p95—

Context cost / 10k—

Normalized trend—

Completion:oldestnewestexpected at mean completion time

Decode throughput versus final context depth · colour encodes completion time, not speed.

Context-normalized decode over completion time · awaiting enough comparable requests.

Task	Model	Quant	Prompt tokens	Prompt t/s	Output tokens	Generation t/s	Total time	Completed
Waiting for completed request metrics…

GPU telemetry

NVIDIA GPU

GPU utilisation—

VRAM occupancy—

Memory activity—

Board power ceiling—

Current performance cap—Awaiting NVIDIA limit reasons.

Temperature—

Thermal thresholds—

Power—

Power limits—

SM clock—

Memory clock—

PCIe link—

Performance state—

Processes—

Driver—

Observed active-workload peaksRetained window

GPU load—

Board power—

Temperature—

Raw decode—

Observed peaks come only from normal active requests in retained telemetry; they are not a synthetic benchmark or a guaranteed hardware maximum.

Host resources

—

CPU utilisation—

System memory—

Root filesystem—

Model storage · /ai-data—

Load 1m—

Load 5m—

Swap—

Ollama—

Inference path

Current state

Read-only collector · three-second hardware sampling · filtered performance events only · prompts and responses are never collected or displayed.