# Extropic's TSU is the next big thing (Do NOT miss this!) Source: https://www.youtube.com/watch?v=mNw7KLN7raU Recap page: https://rapidrecap.app/video/mNw7KLN7raU Generated: 2025-11-04T13:35:09.139+00:00 --- ## Quick Overview The video demonstrates that Extropic's Thermal Sampling Unit (TSU) can solve complex optimization problems like Sudoku and the 8-Queens problem almost instantly by leveraging the inherent physical properties of the device, showcasing a fundamentally different approach to computation compared to classical or current quantum computing methods. **Key Points:** - Extropic's TSU solved a generated Sudoku puzzle in one shot (0.38 penalty) using THRM sampling, which is significantly faster than traditional methods for complex problems. - The TSU approach, leveraging natural noise and thermal dynamics, is fundamentally different from both classical computation (which relies on sequential logic/guess-and-check) and current quantum computing paradigms. - The 8-Queens problem, which has 92 unique solutions, was also solved by the TSU, demonstrating its capability to find the lowest energy state (zero energy level) solution. - The presenter notes that the TSU's method is analogous to simulating physical systems (like protein folding or fusion containment) by mapping constraints onto the device's magnetic field configuration. - The speaker highlights that the first-generation TSU prototype is already commercially viable for certain optimization tasks, contrasting with the theoretical nature of many quantum computing claims. - The TSU simulates the energy landscape of a problem, where the lowest energy state corresponds to the optimal solution, contrasting with classical AI's reliance on iterative gradient descent. - The presentation emphasizes that this technology is the default path for future large-scale AI, suggesting it will be utilized in cloud infrastructure rather than relying solely on traditional GPUs/TPUs. ![Screenshot at 0:05: The title screen prominently displays the Extropic logo and the central theme: "THERMODYNAMIC COMPUTING FROM ZERO TO ONE," setting the stage for the technology demonstration.](https://ss.rapidrecap.app/screens/mNw7KLN7raU/00-00-05.png) **Context:** The video features David Shapiro from Extropic explaining the capabilities and underlying principles of their Thermodynamic Sampling Unit (TSU), a novel computing hardware designed for solving complex optimization problems. The demonstration centers around solving a Sudoku puzzle and referencing the 8-Queens problem to illustrate how the TSU uses physical annealing and thermal noise to efficiently find global minima in complex energy landscapes, contrasting this with standard classical and quantum computing methods. ## Detailed Analysis David Shapiro introduces Extropic's first Thermodynamic Sampling Unit (TSU), which he states is designed to solve optimization problems by mimicking physical annealing processes. He demonstrates this by running a Sudoku solver using the TSU hardware (though he clarifies he is using a simulator for the demo), which finds a valid solution in a single iteration with a low penalty (0.38). This is contrasted with previous, less efficient Python tutorials for the same task. Shapiro further illustrates the TSU's power by mentioning its ability to solve the 8-Queens problem, finding all 92 solutions, all of which correspond to the lowest possible energy state (energy level 0). He explains that the TSU approach differs from classical computing (which uses sequential logic or gradient descent) and quantum computing because it uses the physical noise inherent in the system to explore the energy landscape, allowing it to jump out of local minima. He posits that this technology, which simulates the physical constraints of problems like protein folding or magnetic confinement in fusion reactors, is the future of large-scale AI, potentially surpassing current GPU/TPU reliance by being able to calculate the optimal structure directly in real-time rather than relying on iterative matrix calculations. He concludes by noting that the first-generation TSU is already commercially viable, unlike purely theoretical quantum solutions. ### TSU Sudoku Demonstration - TSU successfully generates and solves a random Sudoku puzzle in one iteration (Penalty/diag: 0.38) - Validation reports (Clue prepared, Rows/Cols Boxes OK, Overall Grid valid) all pass - The process involves pressing Enter to start THRM sampling. ### Thermodynamic vs. Classical/Quantum Computing - Classical optimization relies on sequential logic or gradient descent to find the lowest point - Quantum computing is still largely theoretical - Thermodynamic computing uses natural noise and thermal fluctuations to explore the energy landscape and settle into the global minimum. ### 8-Queens Problem Demonstration - The TSU approach allows for finding all 92 unique solutions for the 8-Queens problem, all corresponding to the lowest energy level (0) - This demonstrates the ability to solve complex, constrained problems efficiently. ### Real-World Applications & Future - The method is applicable to complex physical problems like protein folding and magnetic confinement for fusion - The technology is already commercially viable, unlike many quantum efforts - The speaker suggests this approach will be the default for future large-scale AI, potentially replacing GPU/TPU reliance in cloud centers. ![Screenshot at 0:05: The title screen prominently displays the Extropic logo and the central theme: "THERMODYNAMIC COMPUTING FROM ZERO TO ONE," setting the stage for the technology demonstration.](https://ss.rapidrecap.app/screens/mNw7KLN7raU/00-00-05.png) ![Screenshot at 0:11: Output from the Sudoku solver running on the TSU \(or simulator\), showing the initial puzzle state with blanks represented by dots.](https://ss.rapidrecap.app/screens/mNw7KLN7raU/00-00-11.png) ![Screenshot at 0:50: The console output showing the successful validation report after the TSU \(simulator\) solved the Sudoku puzzle, noting 503 iterations.](https://ss.rapidrecap.app/screens/mNw7KLN7raU/00-00-50.png) ![Screenshot at 1:17: The command line interface showing the execution of 'thermoputer-sudoku.py' running on the virtual TSU.](https://ss.rapidrecap.app/screens/mNw7KLN7raU/00-01-17.png) ![Screenshot at 2:01: The TSU simulation output displaying the first few iterations, showing the penalty decreasing and solutions being found.](https://ss.rapidrecap.app/screens/mNw7KLN7raU/00-02-01.png) ![Screenshot at 3:04: The final validation report after the Sudoku solution was found, showing all checks passed \(e.g., Overall grid valid: PASS\).](https://ss.rapidrecap.app/screens/mNw7KLN7raU/00-03-04.png) ![Screenshot at 3:09: The equation displayed graphically: AI/PERSON = AI/ENERGY \* ENERGY/PERSON, illustrating the relationship between computational efficiency and energy consumption.](https://ss.rapidrecap.app/screens/mNw7KLN7raU/00-03-09.png) ![Screenshot at 3:33: The 8-Queens simulation running, showing iterations finding unique solutions \(Q represents a Queen position\).](https://ss.rapidrecap.app/screens/mNw7KLN7raU/00-03-33.png)