SiliconScope

SiliconScope is a premium native macOS benchmark and diagnostics app for Apple Silicon, built with Swift, SwiftUI, Metal, Core ML, Charts, and SwiftData.

It focuses on real local workloads instead of synthetic placeholder numbers. The app combines CPU, memory, disk, GPU, and AI inference benchmarking with live monitoring, local history, comparison views, and export tools in a polished macOS utility-style interface.

Highlights

• Native macOS app built with SwiftUI and the Swift app lifecycle
• Apple Silicon-focused benchmark flow with quick, standard, and extended modes
• Real benchmark modules for:
  • CPU single-core
  • CPU multi-core
  • memory throughput
  • disk read
  • disk write
  • GPU compute with Metal
  • AI inference with Core ML
• Live monitoring for CPU, memory pressure, thermal state, disk activity, and benchmark progress
• Local run history with comparison and reporting surfaces
• Transparent scoring model with documented weights instead of hidden magic numbers

Benchmark Modules

CPU

The CPU passes exercise integer, floating point, and SIMD-heavy workloads in both single-threaded and multi-threaded configurations.

Memory

The memory benchmark measures read, write, copy, and latency-style behavior over real allocated buffers.

Disk

Disk read and write modules use temporary on-disk workloads to estimate sequential and random throughput.

GPU

The GPU benchmark uses Metal compute pipelines and bundled shaders to measure:

• vector math throughput
• matrix compute throughput
• memory streaming throughput

The run includes warmup, repeated iterations, timing aggregation, and a normalized GPU score.

AI

The AI benchmark uses Core ML for real local inference timing. SiliconScope builds a lightweight local regression model on first use, then benchmarks:

• model load time
• first inference latency
• warm inference latency
• predictions per second

Where supported, the app compares multiple compute-unit modes including CPU-only, CPU+GPU, and all compute units.

App Sections

• Dashboard
• Benchmark
• Live Monitor
• History
• Compare
• Reports
• Settings

Scoring

SiliconScope keeps scoring intentionally transparent.

• CPU, memory, disk, and GPU contribute to the main weighted score
• AI is benchmarked and stored separately so inference modes remain easy to inspect
• Weights live in one place in the codebase and can be tuned without rewriting the benchmark engine

Current weighted modules:

• CPU Single-Core: 20%
• CPU Multi-Core: 24%
• Memory Throughput: 16%
• Disk Read: 10%
• Disk Write: 10%
• GPU Metal: 20%

Tech Stack

• Swift 5.10+
• SwiftUI
• SwiftData
• Charts
• Metal
• Core ML
• Create ML

Requirements

• macOS 14 or newer
• Apple Silicon Mac recommended
• Xcode 15.4+ for app development workflows

Build And Run

SiliconScope currently ships as a Swift package executable with bundled resources.

swift build --scratch-path .build-scratch
./.build-scratch/arm64-apple-macosx/debug/SiliconScope

To refresh the local .app bundle:

./scripts/package_app.sh
open dist/SiliconScope.app

Project Structure

Sources/SiliconScope/App
Sources/SiliconScope/Benchmarks
Sources/SiliconScope/Components
Sources/SiliconScope/Models
Sources/SiliconScope/Monitoring
Sources/SiliconScope/Persistence
Sources/SiliconScope/Resources
Sources/SiliconScope/Utilities
Sources/SiliconScope/Views

Notes

• The app uses public native frameworks only.
• GPU benchmarking fails gracefully when Metal is unavailable.
• AI benchmarking skips unsupported compute modes and records notes instead of failing the whole run.
• Benchmark results are stored locally.

Status

SiliconScope is already functional and benchmark-capable, but it is still evolving. The next major polish areas are richer report layouts, more advanced comparison visuals, and deeper diagnostics.