Skip to content

Performance Testing

Introduction

This document showcases the performance testing results for Exponential Moving Average (EMA) computations, comparing single-threaded and multi-threaded approaches. Tests were conducted on two platforms: a Windows laptop featuring an Intel Core i7-13700 processor and a Mac Mini powered by an Apple M4 chip. The analysis highlights how thread count impacts computational efficiency across diverse dataset sizes, with a special focus on the superior scalability of kand over talib.

Info

kand vs. talib: While talib is constrained to single-threaded execution, kand leverages multi-threading to unlock significant performance gains, especially on modern multi-core hardware.

Test Environment

Hardware Specifications

  • Device: Laptop
  • CPU: Intel Core i7-13700
    • 6 Performance cores
  • OS: Windows 11 Pro
  • Device: Mac Mini
  • CPU: Apple M4
    • 4 Performance cores
  • OS: macOS Sonoma

Software Stack

Component Version Description
🐍 Python 3.11 Core runtime environment
📈 TA-Lib 0.6.3 Technical analysis baseline
⚡ Kand 0.0.11 High-performance implementation

Version Compatibility

All tests were conducted using the latest stable releases of each component. Performance characteristics may vary with different versions.

Test Methodology

Test Code Location

All benchmark code is available in the python/benches directory: - bench_ema.py: Single-thread performance testing - bench_ema_mt.py: Multi-thread performance testing

System Variability

Performance results may vary significantly across different systems due to: - CPU architecture and clock speeds - Memory configuration and speed - Operating system scheduling - System load and background processes - Thermal conditions

We strongly recommend running the benchmarks on your specific system for the most accurate performance assessment.

  • Single-threaded tests: Executed using 1 thread with talib.EMA.
  • Multi-threaded tests: Conducted with kand.ema:
  • Windows i7-13700: 2, 4, and 6 threads.
  • Mac Mini M4: 2 and 4 threads.
  • Dataset sizes: 50K, 100K, 250K, 500K, 1M, 2.5M, 5M, 10M data points.
  • EMA period: 30.
  • Runs per test: 1000 iterations to calculate average execution time.
  • Tools: Python scripts leveraging talib.EMA (single-threaded) and kand.ema (multi-threaded).

Results

Windows i7-13700 Mobile

Windows Single-threaded Figure 1: EMA computation performance on Windows with 1 thread using talib.

Windows 2 Threads Figure 2: Performance with 2 threads using kand.

Windows 4 Threads Figure 3: Performance with 4 threads using kand.

Windows 6 Threads Figure 4: Performance with 6 threads using kand.

Mac Mini M4

Mac Single-threaded Figure 5: EMA computation performance on Mac Mini with 1 thread using talib.

Mac 2 Threads Figure 6: Performance with 2 threads using kand.

Mac 4 Threads Figure 7: Performance with 4 threads using kand.

Analysis

The results reveal compelling insights into performance trends:

  • Multi-threading Advantage with kand: Unlike talib, which is bottlenecked by single-threaded execution, kand harnesses multiple threads to drastically reduce EMA computation times. This advantage becomes increasingly evident with larger datasets (e.g., 5M and 10M points).

  • Scalability:

  • On the Windows i7-13700, performance peaks at 6 threads, fully utilizing its 6 P-cores.

  • On the Mac Mini M4, efficiency maxes out at 4 threads, aligned with its 4 P-cores.

Why kand Outshines talib

The single-threaded nature of talib limits its ability to exploit modern multi-core CPUs. In contrast, kand’s multi-threaded design scales seamlessly with core count, delivering superior performance on datasets of any size.

Key Takeaway

By embracing multi-threading, kand unlocks the full potential of modern processors, leaving talib’s single-threaded approach in the dust—especially for high-performance financial computations.

Key Takeaway

📌 Thread counts are capped by the number of performance cores (P-cores). The i7-13700 supports up to 6 threads with its 6 P-cores, while the M4 is limited to 4 threads due to its 4 P-cores.

Conclusion

Multi-threading capabilities in kand deliver exceptional performance gains across various technical analysis computations, consistently outperforming the single-threaded limitations of traditional libraries like talib. Our benchmarks on the Windows i7-13700 (6 P-cores) and Mac Mini M4 (4 P-cores) demonstrate impressive scalability, with the i7-13700 achieving superior throughput at higher thread counts. For applications dealing with large-scale financial data processing, kand's sophisticated multi-core utilization architecture provides a significant competitive advantage. This foundation sets the stage for future optimizations across different hardware configurations and computational scenarios, making kand an ideal choice for high-performance financial analysis systems.