---
title: Benchmarks & methodology | Garmo Labs
url: https://garmolabs.com/gsp/benchmarks
description: Garmo Labs chips, software and semiconductor IP.
---

GSP / BENCHMARKS & METHODOLOGY

# A result.
With its full context.

Inputs, quality checks, source identity and execution environment travel with the measurements.

[Download benchmark data ](https://garmolabs.com/gsp/benchmarks.json) [Run your own scenario](https://garmolabs.com/gsp/workbench)

## Measured execution.
Not imagined performance.

5 repeated bounded engine executions per synthetic scenario. Wall time includes Python subprocess startup. Independent checks validate the stated outcome. No competitor workloads, hardware PPA or customer ROI are compared.

| Scenario | Verdict | Repeats | Median wall time | Stable artifact identities |

| Feasible baseline Synthetic input / known-answer check | completed | 5 | 65.398 ms | Yes |

| Impossible failure-domain diversity Synthetic input / known-answer check | blocked | 5 | 26.28 ms | Yes |

| Working set exceeds selected memory Synthetic input / known-answer check | blocked | 5 | 54.351 ms | Yes |

| Insufficient power budget Synthetic input / known-answer check | blocked | 5 | 24.923 ms | Yes |

Linux x86_64 / Python 3.13.5. Includes process startup. These are local software timings, not silicon PPA, customer ROI or a competitor comparison.

## Compare the work.
Not just the stopwatch.

A blocked case can be the correct answer. The benchmark retains impossible diversity, insufficient memory and constrained power alongside feasible cases.

01

### Exact scenario and engine hashes

02

### Independent outcome checks

03

### Repeated local execution

04

### Raw artifacts and clear boundaries