Introducing Z1T:
Our first family of transformer-like models
made for sparse probabilistic hardware like Z1
Achieving up to 140x energy efficiency gains over GPUs
and revealing a new scaling law for sparse transformers
Read the blog:
extropic.ai/writing/z1t
Extropic retweeted
Disaggregated inference is the future.
Great panel with @general_compute discussing their heterogenous compute cloud, the speciation of compute for disaggregated inference, and how thermo compute will fit in the picture
More coming soon. Stay tuned 😉
Extropic retweeted
Will be on a panel discussing the future of inference with friends @general_compute today for SF tech week.
You won't want to miss it!
What evolves together fits together.
We’re using RL to post-train models for Thermo AI research, accelerating algorithmic discovery for our new species of computer.
The first sparks of Thermo RSI,
in collaboration with @PrimeIntellect.
extropic.ai/writing/baby-the…
Extropic founder @beffjezos explains how Z1 chips sitting alongside GPUs could make AI data centers 100X more energy efficient:
"The reason this hasn't been really explored is that there's not a huge upside for sparsity on traditional accelerators. Because GPUs aren't really good if you try to do matmuls that are sparse."
"It's like shipping a big freight ship across the ocean with a bunch of empty containers. You're wasting a lot of gas. Whereas for us it's compute and memory, it's naturally sparse. We get a huge energy efficiency gain there"
"Nowadays everybody's obsessed with tokens per watt at a data center level. If you incorporate TSUs like Z1 into your data center and park them next to your GPUs, you can get 100X energy efficiency gain at a data center level."
"People are struggling to power these data centers. It's become the bottleneck. There's a bunch of GPUs sitting on the sidelines right now that can't be powered on, and people are thinking of creating a Dyson swarm to get more power. Why don't we just compute more efficiently first?"
@extropic
Introducing Z1T:
Our first family of transformer-like models
made for sparse probabilistic hardware like Z1
Achieving up to 140x energy efficiency gains over GPUs
and revealing a new scaling law for sparse transformers
Read the blog:
extropic.ai/writing/z1t
Extropic founder @beffjezos explains why his new Z1 chip uses 10,000x fewer transistors than a GPU to run the same AI algorithms:
"pbits are not traditional bits that are definitely zero, definitely one. They're oscillating between the two. It's like a light switch dimmer. You ever have it flicker when you're at the edge? It's kind of like that."
"It's a fractional bit. I could be like, hey, 20% of the time I'm zero, 80% of the time I'm one. That would be a fractional flip."
"The most interesting thing is that you can do all sorts of AI algorithms with about 10,000 times less transistors than if you were to do them digitally."
"If you go back to the ancestry of neural networks, they came from something called energy-based models from Hinton."
"We can do that specifically with Z1. We have literal Boltzmann machines in the hardware. They're very dense, and they use very few transistors, because each transistor is operating in the regime that would be usually considered janky. It's so low power, it's flipping all the time."
@extropic
Major update from us @extropic
Announcing Torx and Thermalizers, basically our version of PyTorch + kernels, but for Thermo hardware
Also: Z1 is now taped out and in fabrication! Will be available on 0.5M pbit sticks, and 4M pbit PCIe cards
+1 more thing at the end... enjoy!
Scaling Thermodynamic Computing:
From One to One Billion
Announcing:
- Our newest chip: Z1 + hardware systems
- Torx: our software framework for stochastic programming
- Thermalizers: our thermodynamic compiler
- Our new live simulator API
Read more :
extropic.ai/writing/from-one…
Our paper on Denoising Thermodynamic Models is now published in Nature's npj Unconventional Computing.
We demonstrate thermodynamic hardware built on the primitives validated in X0 could unlock generative modelling at 10,000x lower energy than GPUs.
nature.com/articles/s44335-0…
Thermodynamic computing isn't just for AI.
It will be an unprecedented tool for simulating the biological world.
New paper: TSUs can run mRNA codon optimization at ~1M× lower energy than GPUs. Estimates grounded in real hardware.
Paper + THRML code :
docs.thrml.ai/en/latest/pape…











