Benchmark methodology

One path from hardware to score.

Lydian combines focused CPU, GPU, ray tracing, and local AI workloads in a consistent benchmark journey shared by the desktop app and command line.

Local execution Comparable results

The shared execution layer

Meet Lydian Runner

Runner coordinates the benchmark from start to finish. It keeps workload order, progress, cancellation, and result handling aligned across every Lydian experience without changing what each platform is asked to do.

01

Prepare

Reads the system profile and prepares optional assets or acceleration only when they are needed.

02

Orchestrate

Runs each workload in a defined order with the same settings and boundaries.

03

Measure

Uses warmups and repeated samples while reporting live progress and respecting cancellation.

04

Score

Turns raw timings and throughput into normalized component scores and a balanced suite result.

05

Verify

Packages the result with its system context and authenticates public submissions before ranking.

Same benchmark semantics
Desktop app Lydian Runner CLI

Full benchmark

A deliberate sequence

The full suite moves from focused processor work to parallel compute, rendering, and local inference. Runner carries progress and results between each component.

  1. 01 CPU single-core
  2. 02 CPU multi-core
  3. 03 GPU compute
  4. 04 Ray tracing
  5. 05 Local AI

What is measured

Four views of performance

01

CPU

Single + multi-core

Ten practical workloads exercise computation, cryptography, compression, navigation, image processing, text, web, and asset handling.

Measures
General processor throughput across varied work.
Method
Single-core and all-core passes with warmup and repeated measurement.
Result
Normalized stage scores combined into representative CPU results.
02

GPU Compute

Parallel throughput

Compute workloads cover matrix math, memory bandwidth, cryptography, transforms, and image processing through the best supported graphics API.

Measures
General-purpose GPU compute rather than game frame rate.
Method
Warmup passes followed by repeated throughput samples and stability checks.
Result
A normalized compute score supported by workload-level metrics.
03

Ray Tracing

Progressive rendering

A consistent 3D scene is rendered progressively in tiles, putting traversal, shading, and parallel execution under sustained load.

Measures
Rendering speed across CPU and GPU execution modes.
Method
The same scene, resolution, and quality settings are used for comparable runs.
Result
A performance score derived from completed rendering work and elapsed time.
04

AI Inference

On-device language model

A compact language model runs locally, with required assets prepared on demand and available hardware acceleration selected automatically.

Measures
Prompt processing, token generation, first-token latency, and stability.
Method
A repeatable generation task runs on the device without sending prompts to a hosted model.
Result
A composite AI score that rewards responsive and sustained inference.

Methodology and trust

Designed for useful comparisons

A benchmark is only useful when the journey behind the number remains consistent and understandable.

01

Measured on your device

Workloads execute locally, so the result reflects the hardware and software environment being tested.

02

One runner, consistent behavior

Desktop and CLI use the same orchestration path, keeping suite order and scoring semantics aligned.

03

Progress you can follow

Stages, preparation, and long-running work report progress, and a cancelled run stops cleanly.

04

Authenticated submissions

Public results are signed with a per-install identity and verified before reaching the leaderboard.

Questions

Benchmark FAQ

What is Lydian Runner?

It is the shared layer that coordinates benchmark execution for the desktop app and CLI. It keeps the public benchmark behavior consistent without being a separate benchmark itself.

What does the full suite measure?

The suite combines CPU single-core and multi-core performance, GPU compute, progressive ray tracing, and local AI inference into a broad view of the system.

How are scores made comparable?

Each component uses repeatable workloads, representative samples, and normalized scoring. Higher scores indicate more work completed or faster completion under the same benchmark conditions.

Is Lydian a gaming benchmark?

Not directly. It measures compute and rendering capabilities that matter to many workloads, but it does not report game frame rates.