Tenstorrent Cookbook: Learn by Building
Welcome to the Tenstorrent Cookbook! This series of hands-on projects teaches you TT-Metal fundamentals through creative, visual applications.
What You'll Build
Each recipe is a complete, working project that demonstrates core TT-Metal techniques:
š® Recipe 1: Conway's Game of Life
Learn: Cellular automata, parallel tile computing, convolution operations
Build the classic cellular automaton where simple rules create complex emergent behavior. Perfect introduction to parallel computation on TT hardware.
Time: 30 minutes | Difficulty: Beginner
šµ Recipe 2: Audio Signal Processing
Learn: FFT, mel-spectrograms, real-time audio effects
Process audio files on TT hardware - compute spectrograms, detect beats, extract pitch. Foundation for speech recognition models like Whisper.
Time: 45 minutes | Difficulty: Intermediate
š Recipe 3: Mandelbrot Fractal Explorer
Learn: GPU-style parallel computation, complex number operations
Render beautiful fractals with an interactive zoom explorer. Demonstrates embarrassingly parallel computation - ideal for understanding performance.
Time: 30 minutes | Difficulty: Beginner-Intermediate
š¼ļø Recipe 4: Custom Image Filters
Learn: 2D convolution, computer vision kernels
Build a library of image filters from edge detection to artistic effects. Same techniques used in ResNet50, MobileNetV2, and ViT models.
Time: 30 minutes | Difficulty: Beginner-Intermediate
š Recipe 5: Particle Life - Emergent Complexity
Learn: NĀ² algorithms, physics simulation, multi-device parallelization
Simulate emergent patterns from simple particle interactions. Includes multi-chip acceleration bonus for QuietBox systems!
Time: 30 minutes | Difficulty: Intermediate
How the Cookbook Works
1. Deploy All Projects
š¦ Deploy All Cookbook Projects
This creates all 5 projects in ~/tt-scratchpad/cookbook/ with one command.
2. Pick a Recipe
Choose any recipe - they're independent and can be completed in any order.
3. Learn by Doing
Each recipe includes:
- ā Complete source code with detailed comments
- ā Visual output (animations, plots, fractals)
- ā Extension ideas for further exploration
- ā One-click execution buttons
4. Experiment and Extend
Modify the code, try different parameters, combine projects. The cookbook is your playground!
Building Blocks for Real Models
These recipes teach fundamental techniques used in production models:
| Recipe | Technique | Used In |
|---|---|---|
| Game of Life | Convolution | YOLO v10-v12, SegFormer |
| Audio Processor | Spectrograms | Whisper speech recognition |
| Mandelbrot | Parallel pixel processing | Stable Diffusion 3.5 |
| Image Filters | 2D convolutions | ResNet50, MobileNetV2, ViT |
| Particle Life | NĀ² algorithms | Physics simulation, pairwise interactions |
What You'll Learn
After completing the cookbook, you'll have:
- ā 5 complete, working projects you built yourself
- ā Deep understanding of TTNN operations (convolution, FFT, parallel compute)
- ā Experience with parallel tile computing optimized for TT hardware
- ā Foundation for production applications and model bring-up
Next Steps After Cookbook
Combine Projects
- Use audio processor to trigger mandelbrot zoom
- Apply image filters to Game of Life visualization
- Real-time audio-reactive fractal visualizations
Explore Production Models
- Convolution ā
models/demos/yolov12x/(object detection) - Transformers ā
models/demos/gemma3/(multimodal AI) - Vision ā
models/demos/mobilenetv2/(mobile inference) - Audio ā
models/demos/whisper/(speech recognition)
Contribute
- Submit your projects as examples
- Participate in bounty program
- Help other developers on Discord
Resources
- Discord: discord.gg/tvhGzHQwaj
- GitHub: github.com/tenstorrent/tt-metal
- Documentation: docs.tenstorrent.com
- METALIUM_GUIDE.md:
~/tt-metal/METALIUM_GUIDE.md- Architecture deep-dive - Tech Reports:
~/tt-metal/tech_reports/- Research papers and optimizations
Ready to start? Pick a recipe above or deploy all projects and dive in!
Happy coding! š