How to Benchmark in Linux

Chris Titus — Fri, 15 May 2026 00:00:00 +0000

Proper Linux benchmarking is not about the biggest single run. It is about building a clean baseline, controlling variables, repeating runs for statistical validity, and generating charts that show the data clearly.

The Phoronix Test Suite provides the structure: repeatable workloads, saved results, baseline comparison, and batch automation. MangoHud captures frame-time data. Together with a Python plotting tool, you get a complete workflow from raw test data to publication-ready charts.

Core Principle: Test One Variable at a Time

Before running any benchmarks:

Same hardware, distro, and driver stack each time
Same game patch or application version
Same graphics preset and resolution
Baseline run first, then one controlled change, then re-run
Repeat at least 3 to 5 times per configuration
Log hardware info (CPU, GPU, kernel, Mesa version) before each session

Phoronix Test Suite saves this metadata automatically. If you change multiple variables at once, your data proves nothing.

Install the Tools

Phoronix Test Suite

# Debian/Ubuntu
sudo apt install phoronix-test-suite

# Fedora
sudo dnf install phoronix-test-suite

# Arch
sudo pacman -S phoronix-test-suite

MangoHud (required for 1% lows and frame consistency)

MangoHud captures frame-time data needed to calculate 1% lows and analyze frame pacing. Without it, you only get average FPS.

# Debian/Ubuntu
sudo apt install mangohud

# Fedora
sudo dnf install mangohud

# Arch
sudo pacman -S mangohud

Plot.py for Charts

Clone the Linux benchmarks repository to get the plotting script:

git clone https://github.com/ChrisTitusTech/linux-benchmarks.git
cd linux-benchmarks

Install Python dependencies:

pip install numpy pandas matplotlib

Prep Your System

Before every test session:

Reboot and let the system settle
Close browsers, Discord, RGB software, and unnecessary services
Plug in laptops and use the same power mode for every run
Keep the same CPU governor and GPU mode
Check kernel, Mesa version, and game patch level
Note the date/time and save this metadata somewhere

Phoronix automatically logs hardware info, so you just need to know what you changed between runs.

Quick Workflow: Batch Setup and Benchmark

1. Configure Batch Mode Once

phoronix-test-suite batch-setup

This prompts you to set defaults for automated testing: whether to save results to OpenBenchmarking, how many times to run each test, and whether to use strict mode. Set these once and reuse them for consistency.

2. Run Batch Benchmarks

export MANGOHUD=1
export MANGOHUD_CONFIG="output_folder=$HOME/benchmark-logs,autostart_log=1,benchmark_percentiles=AVG+1+0.1"

TEST_RESULTS_NAME="baseline" \
TEST_RESULTS_IDENTIFIER="stock-system" \
TEST_RESULTS_DESCRIPTION="Baseline on stock kernel" \
TEST_EXEC_PREPEND="mangohud" \
phoronix-test-suite batch-benchmark pts/c-ray pts/compress-7zip pts/openssl

The batch mode processes all tests without prompting you. Logs are saved to ~/.phoronix-test-suite/ with metadata automatically included.

Repeat this command after making one change (kernel, driver, Proton version) with a different identifier:

export MANGOHUD=1
export MANGOHUD_CONFIG="output_folder=$HOME/benchmark-logs,autostart_log=1,benchmark_percentiles=AVG+1+0.1"

TEST_RESULTS_NAME="tested" \
TEST_RESULTS_IDENTIFIER="zen-kernel" \
TEST_RESULTS_DESCRIPTION="Same workload on Zen kernel" \
TEST_EXEC_PREPEND="mangohud" \
phoronix-test-suite batch-benchmark pts/c-ray pts/compress-7zip pts/openssl

Now you have two saved result sets to compare.

3. Export Results to CSV

After running benchmarks, export the results to ~/benchmark-logs:

phoronix-test-suite result-file-to-csv baseline ~/benchmark-logs/baseline.csv
phoronix-test-suite result-file-to-csv tested ~/benchmark-logs/tested.csv

This creates CSV files that plot.py can read. The CSV contains:

Benchmark name
Average score (or FPS)
Run count
Variance, standard deviation, and standard error (if available)
System metadata (CPU, GPU, kernel)

MangoHud logs are already being saved to ~/benchmark-logs as individual CSV files with frame-time data.

4. Understanding the Export-CSV Workflow

The export process transforms Phoronix binary result files into text formats:

result-file-to-csv: Standard export with aggregated stats per benchmark
result-file-raw-to-csv: Individual run data; useful if you need to inspect variance
result-file-stats: Statistical summary

For plot.py, you need CSV files with:

Phoronix CSV exports — Benchmark names, average FPS/scores, run count, variance
MangoHud CSV logs — Frame-time data with columns like fps and frametime

Both get saved in the same directory and plot.py finds them automatically.

Running plot.py to Generate Charts

Once you have CSV files from both Phoronix exports and MangoHud logs in ~/benchmark-logs, generate visualizations:

cd /path/to/linux-benchmarks
python plot.py --log-dir ~/benchmark-logs

The script generates two types of charts:

1. MangoHud FPS Distribution Heatmap

Shows frame-time consistency across all runs. Each row is one run, and the heatmap colors show how many frames hit each FPS bucket. Vertical lines mark:

0.1%ile (darkest): lowest 0.1% of frames
1%ile (dark): lowest 1% of frames
50%ile (blue): median frame rate

If the heatmap is wide and flat, frames are inconsistent. If it is tall and narrow, frame pacing is smooth.

2. Phoronix Average FPS Bar Chart

Shows average performance per benchmark with run count, variance, and standard error labeled. If variance is high or run count is low, that test may not be stable yet.

Output Files

Both charts are saved as PNG files in ~/benchmark-logs:

mangohud_bar_chart.png — FPS distribution heatmap (frame consistency from MangoHud logs)
*_avg_fps_bar_chart.png — One per Phoronix CSV export (average FPS per benchmark)

Enriching Phoronix Data with MangoHud Stats

If your Phoronix export only has one aggregate result per benchmark (common when exporting from the GUI), plot.py automatically matches MangoHud runs by name pattern and enriches the Phoronix data with:

Actual run count from multiple MangoHud passes
Calculated variance and standard error
Frame-time percentiles (0.1%, 1%, 50%)

Complete Example: Kernel Comparison

Here’s a real-world walkthrough:

# 1. Configure batch mode
phoronix-test-suite batch-setup

# 2. Set up logging environment
export MANGOHUD=1
export MANGOHUD_CONFIG="output_folder=$HOME/benchmark-logs,autostart_log=1,benchmark_percentiles=AVG+1+0.1"

# 3. Run baseline on stock kernel (reboot first)
TEST_RESULTS_NAME="kernel-compare" \
TEST_RESULTS_IDENTIFIER="stock-distro" \
TEST_RESULTS_DESCRIPTION="Stock distro kernel" \
TEST_EXEC_PREPEND="mangohud" \
phoronix-test-suite batch-benchmark pts/c-ray pts/compress-7zip pts/openssl

# 4. Reboot into Zen kernel and run same tests
TEST_RESULTS_NAME="kernel-compare" \
TEST_RESULTS_IDENTIFIER="zen-kernel" \
TEST_RESULTS_DESCRIPTION="CachyOS Zen kernel" \
TEST_EXEC_PREPEND="mangohud" \
phoronix-test-suite batch-benchmark pts/c-ray pts/compress-7zip pts/openssl

# 5. Export both results to CSV
phoronix-test-suite result-file-to-csv kernel-compare ~/benchmark-logs/kernel-compare.csv

# 6. Run plot.py to generate charts
cd /path/to/linux-benchmarks
python plot.py --log-dir ~/benchmark-logs

Now you have:

One bar chart showing average FPS for each test under both kernels
Run counts, variance, and error bars on the chart
MangoHud frame-time heatmaps if you also logged with MangoHud

This is publication-ready data: controlled variables, multiple runs, statistical validity, and clear visualization.

What the Charts Tell You

Bar chart reading:

Higher bars = better performance
Variance labeled in red: if it is high, the result is noisy
Small error bars = consistent performance across runs
Large error bars = inconsistent or thermal-throttled performance

Heatmap reading:

Tall narrow heatmap = consistent frame pacing
Wide scattered heatmap = erratic frame delivery, even if average FPS is high
Compare rows: if one kernel produces more consistent colors, it has better frame consistency

Workflow Checklist

Reboot and verify kernel/driver/Mesa version
Close unnecessary services
Run phoronix-test-suite batch-setup once
Run baseline: phoronix-test-suite batch-benchmark pts/...
Make one change only (kernel, driver, etc.)
Run same tests again with new identifier
Export: phoronix-test-suite result-file-to-csv <name>
Generate charts: python plot.py --log-dir <path>
Compare against baseline
Repeat for next variable

Distros on Chris Titus Tech | Tech Content Creator