Explore how the CPU and GPU collaborate during gaming workloads, focusing on identifying and analyzing bottlenecks. This project serves as a comprehensive toolkit for analyzing CPU-GPU interaction patterns in gaming workloads through simulation, real-time monitoring, and performance analysis.
This project provides tools to:
- Simulate realistic gaming workloads on both CPU and GPU
- Monitor real-time performance metrics during gameplay
- Automate performance profiling
- Analyze CPU-GPU parallelism and bottlenecks
- Generate detailed performance reports and visualizations
- CPU-based simulation with multi-threading support
- GPU-accelerated simulation using CUDA
- Configurable parameters:
- Resolution (720p to 1440p)
- CPU thread count (1-8)
- Workload intensity (light/medium/heavy)
- Simulates:
- Graphics rendering (matrix operations)
- Physics calculations
- AI pathfinding
- Live CPU and GPU metrics tracking
- FrameView integration for FPS monitoring
- Metrics logged:
- FPS and frame times
- CPU/GPU utilization
- Memory usage
- Temperature and power draw
- Thread utilization
- Tmux-based monitoring interface
- Automated benchmark execution
- CSV logging of performance metrics
- Support for multiple game configurations
- Bottleneck identification
- CPU-GPU parallelism efficiency analysis
- Resource utilization patterns
- Performance visualization
- Optimization recommendations
- Python 3.8+
- NVIDIA GPU with CUDA support
- Required Python packages:
numpy cupy pandas matplotlib psutil pynvml - NVIDIA drivers and CUDA toolkit
- Tmux (for monitoring interface)
- Clone the repository:
git clone https://github.com/samisrana/Gaming-Workload-Profiling-and-CPU-GPU-Parallelism-Analysis.git
cd Gaming-Workload-Profiling-and-CPU-GPU-Parallelism-Analysis- Install dependencies:
pip install numpy cupy pandas matplotlib psutil pynvml- Ensure NVIDIA drivers and CUDA toolkit are installed:
nvidia-smi # Verify NVIDIA drivers
nvcc --version # Verify CUDA toolkit- CPU-only simulation:
python Workloadsim/test_gaming_workload.py- GPU-accelerated simulation:
python Workloadsim/test_gaming_workload_gpu.pyThe real-time monitoring interface provides comprehensive system metrics visualization:
- Monitor a specific game:
python realtime/main.py [game_name]- Launch the monitoring interface:
./setup_profiling.sh
The monitoring tool starts with a simple interface for selecting the game to monitor. It supports various games and provides case-insensitive search.
When no game is running, the interface shows baseline system metrics:
- Real-time CPU and GPU usage graphs
- Memory utilization
- Temperature and power consumption
- System resource baselines
During active gameplay (Terraria shown):
- Split-screen view of metrics and game
- Live performance tracking
- CSV logging for post-session analysis
- Complete system resource utilization
The workload simulation provides comparative performance analysis between CPU-only and GPU-accelerated implementations.
Key findings from GPU implementation:
- Higher FPS at lower resolutions (60+ FPS at 720p)
- Increased GPU utilization with resolution (30% at 720p to 75% at 1440p)
- Linear scaling of GPU memory usage with resolution
- CPU usage remains moderate even at higher resolutions
Key findings from CPU implementation:
- Lower overall performance (max 22 FPS at 720p)
- More consistent CPU utilization across resolutions
- Significant performance drop at higher resolutions
- Better power efficiency but lower absolute performance
These results demonstrate the efficiency of GPU acceleration for graphics-intensive workloads, while highlighting the importance of CPU-GPU balance in gaming applications.
A real-world performance analysis of Rocket League demonstrates the practical application of our monitoring tools:
The graph above shows resource utilization across different game states, demonstrating key patterns in CPU-GPU interaction:
- Launch Phase: Initial low resource usage as the game starts up
- Menu Phase:
- Increased CPU activity for game setup and UI rendering
- Gradual increase in GPU power consumption (11.45W to 43.46W)
- Initial memory loading occurs
- Gameplay Phase:
- Peak GPU usage (~31%) with stable temperatures
- Decreased CPU usage as GPU takes over rendering workload
- Memory usage stabilizes around 43%, indicating efficient resource management
- Consistent FPS throughout match duration
- End Phase: Gradual resource reduction
The analysis reveals efficient CPU-GPU load balancing with no significant bottlenecks, as evidenced by stable frame times and optimized resource distribution.
- Session Duration: 1 hour 6 minutes
- Game Active Time: 98.1% of session
- Average FPS: 143.5
- Average Frame Time: 6.94ms
- 1% Low FPS: 140.6
- 0.1% Low FPS: 138.6
- Average Utilization: 30.2%
- Peak Utilization: 92.0%
- Average Memory Used: 2176MB
- Peak Memory Used: 2398MB
- Maximum Temperature: 47.0°C
- Average Power Draw: 30.3W
- Power Scaling: 11.45W (idle) to 43.46W (peak gameplay)
- Average CPU Usage: 9.6%
- Peak CPU Usage: 100.0%
- Average Memory Usage: 44.2%
Our simulation results showed:
- CPU Version:
- Light workload (720p, 2 threads): ~22 FPS
- Medium workload (1080p, 4 threads): ~10 FPS
- Heavy workload (1440p, 8 threads): ~5 FPS
- CPU usage scales with workload: 10% → 11% → 14%
- GPU Version:
- Light workload (720p, 2 threads): ~62 FPS
- Medium workload (1080p, 4 threads): ~30 FPS
- Heavy workload (1440p, 8 threads): ~20 FPS
- GPU utilization scales with resolution: 30% → 50% → 75%
- GPU memory scales linearly: 1500MB → 2000MB → 3000MB
- CPU usage increases with resolution: 25% → 35% → 50%
These simulation results contrast with the real game performance, demonstrating the optimization potential in professional game engines compared to our simulation models.
Our research and development have yielded several significant achievements:
- Successfully developed a foundational gaming workload simulation that accurately models CPU-GPU interaction patterns seen in production games
- Validated simulation insights against real-world data from Rocket League, confirming our model's ability to predict resource utilization patterns
- Established a comprehensive profiling methodology combining simulation baselines with real-time monitoring
- Demonstrated the effectiveness of GPU acceleration, achieving up to 7x performance improvement over CPU-only implementation
- Identified efficient resource management patterns in commercial games, particularly in areas of power scaling and memory utilization
- Developed tools for analyzing CPU-GPU load balancing and bottleneck detection
- Expand testing to cover more games across different genres
- Implement additional workload types beyond gaming
- Develop more sophisticated physics and AI simulation modules
- Add memory access pattern analysis to better understand data flow between CPU and GPU
- Develop automated bottleneck detection using machine learning to provide real-time optimization suggestions
- Implement predictive performance modeling based on collected metrics
- Create automated reporting and analysis features
- Develop real-time visualization improvements for better monitoring
- Implement cross-platform support for various gaming engines
├── setup_profiling.sh # Monitoring setup script
├── monitors/
│ ├── plotcpu.py # CPU monitoring visualization
│ └── plotgpu.py # GPU monitoring visualization
├── realtime/
│ ├── main.py # Main monitoring program
│ ├── metrics_collector.py # Performance metrics collection
│ ├── data_logger.py # Data logging functionality
│ ├── display_manager.py # Real-time display management
│ ├── game_performance_logs # csv file logs
│ └── game_performance_analysis # Graphs, logs, and reports of average data per session
└── Workloadsim/
├── gaming_workload_simulation.py # CPU simulation workload
├── gaming_workload_simulation_gpu.py # GPU simulation workload
├── test_gaming_workload.py # CPU simulation tests
├── test_gaming_workload_gpu.py # GPU simulation tests
└── benchmark_results # test results
The toolkit collects and analyzes the following metrics:
- Frame rate (FPS) and frame times
- CPU utilization (overall and per-thread)
- GPU utilization and memory usage
- Temperature and power consumption
- Memory bandwidth and usage patterns
- Thread scaling efficiency
Contributions are welcome! Please feel free to submit a Pull Request.
- NVIDIA for CUDA and NVML toolkits
- Python scientific computing community
- Gaming performance analysis community