Testing GuideBest Practices

GPU Benchmark Best Practices

Comprehensive guide to preparing your system, controlling testing environment, and validating results for accurate, reproducible GPU benchmark testing.

System Preparation

Accurate GPU benchmarking requires careful system preparation to eliminate variables that could skew results. Following this preparation protocol ensures reproducible, comparable measurements.

Step 1: Driver and Software Updates

Driver Preparation Checklist
Update GPU drivers: Install latest stable drivers from manufacturer (NVIDIA, AMD, Intel). Avoid beta drivers unless specifically testing them. Driver version significantly impacts performance.
Update browser: For WebGL benchmarks like ours, ensure browser is current. Chrome/Edge typically offer best WebGL performance. Firefox acceptable alternative.
Update OS: Install Windows/macOS/Linux updates. Graphics stack updates often come through OS updates, especially on Windows 10/11.
Reboot after updates: Always reboot system after driver/OS updates before benchmarking. Ensures clean state without update services running.
⚠️
Note driver version: Record exact driver version for comparison. Different driver versions can show 5-10% performance variance for same hardware.

Step 2: Close Background Applications

Background processes steal GPU resources and create performance inconsistency. Minimize running applications for clean benchmarking environment:

Applications to Close Before Testing

Critical (Must Close)

  • • Other browsers or browser tabs (except testing tab)
  • • Video playback (YouTube, Netflix, media players)
  • • Games or 3D applications
  • • Video editing software
  • • Screen recording/streaming software (OBS, etc.)
  • • RGB control software (can cause polling overhead)

Recommended (Should Close)

  • • Discord, Slack, communication apps
  • • Cloud sync services (Dropbox, OneDrive, Google Drive)
  • • Torrent clients
  • • Antivirus real-time scans (temporarily disable if safe)
  • • Browser extensions (use clean profile or incognito)

Acceptable (Can Keep Running)

  • • System monitoring tools (HWiNFO, GPU-Z) for data collection
  • • Lightweight system utilities
  • • Background OS services (don't disable these)

Step 3: Power and Performance Settings

Power Configuration (Desktop & Laptop)
Windows Power Plan:
  • • Desktop: Set to "High Performance" or "Ultimate Performance"
  • • Laptop: Set to "High Performance" AND plug into AC power
  • • Navigate: Settings → System → Power & battery → Power mode
NVIDIA Control Panel (NVIDIA GPUs):
  • • Manage 3D Settings → Power management mode: "Prefer maximum performance"
  • • Disable battery boost features (laptops)
AMD Settings (AMD GPUs):
  • • Gaming → Global Graphics → GPU Workload: "Graphics"
  • • Performance → Tuning: Enable "Tuning Control" if overclocking
Laptop-Specific:
  • • CRITICAL: Plug into AC power (battery severely limits performance)
  • • Disable battery saver mode
  • • Check manufacturer control panel (Dell Power Manager, Lenovo Vantage, etc.)
  • • Set to "Performance" or "Turbo" mode if available

Step 4: System Cleanliness

Physical cleanliness directly impacts thermal performance and therefore benchmark results:

  • Dust cleaning: Clean GPU and case fans, air filters, and heatsink fins before thermal testing. Dust accumulation can raise temperatures 10-20°C.
  • Cable management: Ensure cables don't block airflow to GPU area. Improves cooling efficiency.
  • Laptop elevation: For laptops, use cooling pad or elevate 2-3 inches for bottom intake clearance.

Environmental Control

Environmental factors significantly affect GPU thermals and therefore sustained performance. Controlling testing environment ensures reproducible results.

Ambient Temperature Control

Temperature Impact on Performance

GPU temperatures are directly influenced by room temperature. Every 5°C ambient increase = approximately 5-7°C GPU temperature increase under load.

Ideal testing temperature:20-22°C (68-72°F)
Acceptable range:18-25°C (64-77°F)
Problematic (hot room):>28°C (>82°F)

If comparing results with online leaderboards, note your room temperature. Summer testing in 30°C room cannot fairly compare to winter testing at 18°C.

Airflow Optimization

Testing Environment Airflow Checklist
Case panels: Desktop case side panels should be installed (open case often worsens airflow by disrupting designed air paths)
Room air circulation: Ensure room has air movement. Stagnant hot air around computer raises intake temperatures. AC or fan helpful.
Direct sunlight: Avoid running thermal tests with direct sunlight on case. Can add 5-10°C to ambient.
Surface placement: Laptops: hard, flat surface with ventilation clearance. Never test on bed, couch, or lap (blocks vents).

Display and Browser Configuration

For WebGL benchmarks specifically:

  • Fullscreen or maximized: Run benchmark in fullscreen mode (F11) or maximized window for consistent resolution and reduced overhead
  • Monitor refresh rate: Note your monitor's refresh rate. VSync can cap FPS to display refresh (60Hz, 120Hz, 144Hz, etc.)
  • Browser tab focus: Keep benchmark tab active and visible. Backgrounded tabs may be throttled by browser
  • Browser hardware acceleration: Ensure enabled (Settings → System → Use hardware acceleration when available)

Testing Procedure

Following consistent testing procedure ensures reproducible results and meaningful comparisons across test runs or different systems.

Pre-Test Warmup

GPU Warmup Protocol

GPUs perform differently when cold vs warmed up. For consistent results, bring GPU to thermal equilibrium before starting measured test.

Warmup procedure:
  1. 1. Run short test (5 minutes) or light game to bring GPU to operating temperature (60-70°C)
  2. 2. Wait 2-3 minutes for temperatures to stabilize
  3. 3. Begin actual benchmark test
Our benchmark includes built-in 10-second warmup before measurement begins. For thermal testing, optional manual warmup ensures GPU starts from consistent thermal state.

Test Duration Selection

Choosing Appropriate Test Duration

10-Minute Test

Use for: Quick performance checks, comparing before/after driver updates, initial system validation
Limitation: May not capture full thermal throttling on some systems. Desktop GPUs usually reach equilibrium by 10 minutes; laptops may need longer.

15-Minute Test (Recommended)

Use for: Standard thermal characterization, most desktop GPUs, general benchmarking
Sweet spot: Balances test duration with comprehensive thermal exposure. Adequate for most use cases.

30-Minute Test

Use for: Laptop GPU testing, compact desktop GPUs, comprehensive thermal validation, sustained workload simulation
Advantage: Ensures complete thermal saturation. Critical for laptops where throttling may develop gradually over 20+ minutes.

During Test: Do's and Don'ts

Test Execution Guidelines

✓ DO

  • • Keep benchmark tab active and visible
  • • Leave computer alone (don't move mouse excessively)
  • • Monitor temperatures if possible (HWiNFO, GPU-Z)
  • • Note any unusual system behavior
  • • Let test complete fully to end
  • • Allow fans to spin up as needed (noise OK)

✗ DON'T

  • • Switch to other browser tabs
  • • Launch applications during test
  • • Move/resize browser window
  • • Manually override fan speeds mid-test
  • • Stop test early (invalidates results)
  • • Disconnect laptop from AC power

Multiple Test Runs

For maximum accuracy and reproducibility, run benchmark 2-3 times and average results:

Multi-Run Testing Protocol
Procedure:
  1. 1. Complete first benchmark run (note all metrics)
  2. 2. Wait 15-20 minutes for GPU to cool to idle temperature (<50°C)
  3. 3. Run second benchmark (note metrics)
  4. 4. Optional third run if first two differ by >5%
  5. 5. Average FPS, SLS, P99 frame times across runs
When to investigate:
  • • If results vary by >10% between runs (indicates instability)
  • • If SLS differs significantly (thermal inconsistency)
  • • If one run shows abnormal behavior (driver issue, background process)

Result Validation

After completing benchmark, validate results to ensure test accuracy and detect potential issues.

Result Sanity Checks

Validation Checklist
Check for "degraded" flag: Our benchmark marks results as degraded if WebGL support is limited or tab was backgrounded. Degraded results shouldn't be compared with clean tests.
Verify FPS is reasonable: Typical ranges: Low-end GPU 20-40 FPS, mid-range 45-75 FPS, high-end 80-150+ FPS. Extremely low (<15) or high (>200) FPS warrants investigation.
SLS should be 80-100%: SLS below 80% indicates severe throttling. Above 100% is mathematically impossible (indicates measurement error).
P99 vs P95 spread: P99 should be 10-40% higher than P95. Extreme spread (>50%) suggests instability. Very small spread (<5%) unusual but possible with excellent consistency.
SFD range: Stutter Frequency Density typically 0.5-5%. Above 8% indicates serious performance issues. Below 0.1% unusual but excellent.

Comparing with Expected Performance

Research typical performance for your GPU model:

  • Search online for your GPU model + "volume shader benchmark" or similar WebGL benchmarks
  • Check hardware review sites for general GPU performance tier
  • Note: Laptop GPUs perform 20-40% below desktop counterparts despite same model number (power limitations)
  • Driver version differences can account for 5-10% variance

When Results Don't Make Sense

Troubleshooting Unexpected Results
Much lower performance than expected:
  • • Check power plan settings (might be on battery saver)
  • • Verify GPU drivers are up to date
  • • Check if GPU is thermal throttling severely (temps >85°C)
  • • Ensure browser is using discrete GPU (not integrated graphics)
  • • Try different browser (Chrome recommended for WebGL)
SLS below 80% (severe throttling):
  • • Clean dust from GPU and case
  • • Check GPU temperatures during test (should be <85°C)
  • • Verify case fans are working
  • • For laptops: ensure on AC power, cooling pad recommended
  • • See our thermal throttling guide for detailed solutions
Inconsistent results between runs (>15% variance):
  • • Close all background applications
  • • Disable real-time antivirus temporarily
  • • Check for Windows updates running in background
  • • Update GPU drivers (old drivers can cause instability)
  • • Ensure adequate GPU cooling (unstable temps = unstable clocks)

Common Mistakes to Avoid

Top 10 Benchmarking Mistakes
1.

Testing laptop on battery power

Laptops severely limit GPU performance on battery. ALWAYS test plugged into AC power or results will be 30-60% lower than actual capability.

2.

Running benchmark in background tab

Browsers throttle background tabs. Benchmark must be visible, active tab for valid results. Our benchmark detects backgrounding and marks results as degraded.

3.

Not closing other applications

Background apps steal GPU resources. Close video players, Discord, other browsers, and unnecessary software before testing.

4.

Testing immediately after driver update without reboot

Driver updates may not fully apply until reboot. Always restart system after GPU driver changes before benchmarking.

5.

Using too-short test duration

Short tests (<10 minutes) don't capture thermal throttling. Use 15-30 minute tests for thermal characterization, especially for laptops.

6.

Testing laptop on soft surface (bed, couch)

Blocks bottom air intake, causing severe thermal throttling. Always test laptops on hard, flat surface with ventilation clearance.

7.

Comparing results across different driver versions

Driver version significantly affects performance (5-10% variance possible). When comparing results, ensure same driver version or note version difference.

8.

Ignoring ambient temperature

Testing in 30°C summer heat vs 18°C air-conditioned room produces different results. Note ambient temp for meaningful comparisons.

9.

Single test run without validation

One-off results can be anomalous. Run 2-3 tests and average for reliable data. Investigate if variance exceeds 10%.

10.

Not noting system details

Record GPU model, driver version, test date, and configuration. Essential for tracking performance over time or comparing with others.

Troubleshooting Guide

Quick Reference Problem Solver

Common Issues and Solutions
❓ Benchmark shows "Degraded Mode" warning
  • • Browser lacks WebGL2 support: Update browser to latest version (Chrome/Edge/Firefox recommended)
  • • Tab was backgrounded: Keep benchmark tab active and visible throughout test. Don't switch tabs or minimize window.
  • • Integrated graphics fallback: Ensure browser is using discrete GPU (check GPU settings in Windows Graphics Settings)
❓ FPS extremely low (under 15)
  • • Likely using integrated graphics instead of discrete GPU
  • • Windows: Settings → System → Display → Graphics → Add browser → High performance
  • • NVIDIA: Control Panel → Manage 3D Settings → Program Settings → Add browser → High-performance NVIDIA processor
  • • Check Task Manager during test to verify correct GPU is active
❓ Benchmark won't start or crashes browser
  • • Update GPU drivers to latest stable version
  • • Try different browser (Chrome generally most compatible)
  • • Disable browser extensions (use incognito/private mode)
  • • Clear browser cache and cookies
  • • Ensure hardware acceleration enabled in browser settings
❓ Results vary wildly between test runs (>20%)
  • • Close ALL background applications (especially video, chat apps)
  • • Disable real-time antivirus temporarily (if safe to do so)
  • • Check for Windows updates running in background (Task Manager → check disk/CPU usage)
  • • Verify GPU temperatures stable (severe throttling = inconsistent results)

For detailed thermal troubleshooting, see our complete Thermal Throttling Guide.