Question:
hi Gemini, I just read some phrases about Extropic. Do you have some news about that company?
Answer:
That's an interesting company! Extropic is an AI hardware startup that has recently been in the news for its work on "thermodynamic computing."
Here are the key takeaways and the latest updates:
? Latest News (October 2025):
Chip Unveil: Extropic has recently unveiled its new thermodynamic computing architecture and its first working chips, which they call Thermodynamic Sampling Units (TSUs).
XTR-0 Prototype: They have produced a prototype development platform, the XTR-0, and are currently testing it with early partners, including frontier AI labs and a weather modeling startup.
Energy Efficiency Claims: The startup is making bold claims, suggesting their thermodynamic chips could be thousands of times more energy-efficient—potentially up to 10,000 times—than current GPUs for certain generative AI tasks.
Software Release: They have also open-sourced a Python library called
thrml, which allows developers to simulate the TSU behavior on existing GPUs, lowering the barrier to entry for their new ecosystem.
? What is Extropic and Thermodynamic Computing?
Mission: Extropic, founded in 2022 by former Google quantum computing researchers, aims to build the world's most energy-efficient computers by pioneering a physics-based computing paradigm.
Harnessing Noise: Traditional computers fight against "noise" and entropy (randomness) in their circuits. Extropic's core idea is to harness this natural thermal noise and out-of-equilibrium thermodynamics to perform probabilistic calculations.
Probabilistic AI: They argue this approach is a perfect fit for probabilistic AI workloads like generative models, which rely heavily on statistical sampling. Their TSUs use "p-bits" (probabilistic bits), which represent a state of uncertainty, rather than the definite 1s or 0s of conventional digital bits.
Funding: They secured a $14.1 million seed funding round to pursue this ambitious vision.
In short, Extropic is challenging the current GPU-centric, energy-intensive model of AI by proposing a fundamentally different type of hardware. The tech is highly ambitious, and the next steps will involve seeing if their efficiency claims hold up at a larger scale.
Would you like to know more about the technical details of how their thermodynamic chips work, or perhaps what the broader implications of this technology could be for the AI industry?
Question:
at this moment, no. What do you think about their new approach?
Answer:
That's an excellent question. The new approach by Extropic is highly polarizing within the tech and physics communities.
It represents a classic "high risk, high reward" scenario in deep tech.
Here is a breakdown of the expert opinion and the core challenges:
⭐ The High-Reward Potential (The Upside)
The primary reason Extropic's approach is exciting is its potential to solve the "AI Energy Wall."
Massive Efficiency: Their main claim is a 100x to 10,000x improvement in energy efficiency for specific AI tasks like probabilistic sampling (which is crucial for generative models like Diffusion Models and certain areas of Bayesian inference).
Physics-First Architecture: Instead of forcing algorithms that naturally involve randomness (like AI) onto rigid, energy-hungry digital hardware (CPUs/GPUs), Extropic is creating hardware whose physical dynamics are the algorithm. This eliminates the massive energy cost of trying to simulate randomness digitally.
Breaking Moore's Law Limits: Thermodynamic computing is seen by its proponents as a way to create a new S-curve of computational growth that bypasses the looming physical limitations (heat, energy) of traditional transistor scaling (Moore's Law).
? The High-Risk Challenges (The Skepticism)
The skepticism surrounding Extropic is rooted in the significant technical and commercial hurdles they face:
1. Scalability and Janky-ness
Analog vs. Digital: Thermodynamic computing is essentially a form of analog or stochastic computing. Historically, analog systems have struggled with scalability, precision, and reliability compared to pristine digital systems.
Proof of Concept vs. Data Center: They have a working prototype (XTR-0), but scaling a small chip from a physics lab to a massive, reliable, and manufacturable data-center-sized accelerator is an engineering challenge that could take decades.
2. Limited Application Scope
Niche Workloads: Extropic's chips are explicitly designed for probabilistic workloads that benefit from sampling. They are not general-purpose computers. While generative AI is a huge and growing market, their architecture is not an immediate replacement for a GPU's core function (like matrix multiplication for training large language models).
3. The Hype vs. Science Balance
Bold Claims: The claims of a 10,000x advantage are extremely bold and currently lack independent, rigorous benchmarks on real-world, large-scale AI models.
Ecosystem Challenge: NVIDIA's CUDA has an iron grip on the AI software ecosystem. Even if Extropic's hardware is technically superior, they still need to convince developers to adopt their new algorithms and their open-source library (
thrml), which is a massive commercial undertaking.
⚖️ Conclusion
The general consensus is that the science of thermodynamic computing is legitimate and represents a crucial area of research to tackle the AI energy problem. Extropic is seen as the most high-profile and ambitious player pushing this from academia to a commercial product.
The question isn't whether the physics works, but whether Extropic can:
Scale it up reliably and cost-effectively.
Prove the 10,000x claim on industry-relevant benchmarks.
Build a software ecosystem that overcomes the inertia of the current GPU standard.
It's a company to watch closely, as success would be truly revolutionary, but failure is a very real possibility given the difficulty of building a completely new computing architecture.
Would you be interested in learning about the founder's background and his connection to the "Effective Accelerationism" (e/acc) movement, which is often discussed alongside the company?
no and thanks for now. Its readable stuff for my blog to let readers know that research is staying on the fast moving line.
That's fantastic! I'm glad I could provide you with some valuable and highly relevant information for your blog. Extropic's journey is certainly one that demonstrates the cutting edge of research pushing against current technological limits.
I wish you the best of luck with your writing and your readers will surely appreciate being informed about this fast-moving field!
Nessun commento:
Posta un commento