🧠 Deep Learning System Benchmark

📌 Overview

Deep Learning System Benchmark is a Python-based benchmarking suite designed to assess your system's performance for deep learning workflows. It supports:

• ✅ CPU & GPU performance (PyTorch & TensorFlow)
• ✅ Matrix multiplication, convolution, SVD ops
• ✅ Model training and inference speed
• ✅ Memory allocation, copy bandwidth
• ✅ Disk read/write throughput
• ✅ CUDA-specific metrics (Tensor Cores, transfers, precision)

It generates a full report + performance plots — perfect for comparing environments, checking bottlenecks, or validating hardware upgrades.

---

🚀 Installation

git clone https://github.com/mertcanozdemir/deep-learning-system-benchmark.git
cd deep-learning-system-benchmark
pip install -r requirements.txt

Optional: Install GPU libraries like torch, tensorflow, GPUtil, py-cpuinfo for full capabilities.

---

🧪 Running the Benchmark

python benchmark.py

To skip plot generation:

python benchmark.py --no-plots

Specify a custom output directory:

python benchmark.py --output-dir my_results/

---

📊 Output

• ✅ JSON log of all results (benchmark_results/)
• ✅ Text report summary
• ✅ Plot visualizations: matrix ops, I/O, CUDA precision

Example:

benchmark_results/
├── benchmark_results_20250408_143021.json
├── benchmark_report_20250408_143021.txt
└── plots/
    ├── matrix_multiplication.png
    ├── inference_performance.png
    ├── cuda_precision.png
    └── disk_performance.png

Example text output:

System: Windows 10 CPU: AMD64 (8 physical / 16 logical cores) Memory: 15.37 GB GPU: NVIDIA GeForce GTX 1650 Ti (4.0 GB CUDA)

Overall Rating: 10/10 Category: Excellent for deep learning

🧠 Deep Learning Performance (PyTorch):

• ResNet50 inference: 214.4 images/sec
• CNN inference: 236,845 images/sec
• CNN training: 400.3 iterations/sec
• Matrix multiplication (1024x1024): 0.0010 sec

🧠 Deep Learning Performance (TensorFlow):

• ResNet50 inference: 23.0 images/sec
• CNN inference: 1,961.7 images/sec
• CNN training: 0.025 sec/epoch
• Matrix multiplication (1024x1024): 0.0059 sec

🔢 NumPy:

• Matrix multiplication (1024x1024): 0.0050 sec
• SVD (512x512): 0.307 sec

💾 Disk I/O:

• Read speed: 2433 MB/s
• Write speed: 92.6 MB/s

🧬 Memory:

• Estimated bandwidth: 2511 MB/s
• Copy speed (1GB): 2631 MB/s

⚠️ Weakness:

• Slow CPU-GPU transfers (4.9 GB/s)

💡 Recommendation:

• Consider upgrading RAM to at least 16GB for better performance with larger models.

---

📎 Requirements

• Python ≥ 3.7
• NumPy, pandas, matplotlib
• Optional: PyTorch, TensorFlow, GPUtil, psutil, cpuinfo

Install extras with:

pip install torch tensorflow gputil py-cpuinfo psutil

---

🛠 Features

• 🔬 Compare CPU vs GPU performance
• 🔁 Measure TensorCore acceleration & mixed precision
• 📉 Profile I/O bottlenecks and memory limits
• 📥 Export results for tracking across machines

---

📄 License

MIT License © Mertcan Özdemir

---

🤝 Contributing

Pull requests are welcome! If you find a bug or have a feature idea, please open an issue.