US EditionPublished • loading... • Updated
Extropic Unveils Thermodynamic 10,000x Energy Saving AI Chips
Extropic's thermodynamic sampling units use probabilistic bits to reduce AI energy use by up to 10,000 times, addressing data center power challenges, the startup said.
- Three years ago Extropic bet energy—not chips or data—would limit AI scaling, and the startup has produced its first working Thermodynamic Sampling Unit hardware as proof.
- Because data movement, GPUs evolved for matrix multiplication that drives heavy energy use; Extropic designs TSUs to sample probabilities directly, reducing power strain on data centers.
- Extropic's XTR-0 development platform pairs an FPGA with two X-0 probabilist chips, using p-bits that fluctuate between 0 and 1, and released TRHML for open-source simulation.
- The first working chip has been shared with early partners including Atmo, startup customer working on weather forecasting, whose CEO Johan Mathe tests it for Department of Defense; Extropic is now hiring engineers to scale production.
- Simulations show DTMs running on TSUs offer about 10,000x energy savings, and Extropic says scaling could remove AI power constraints and challenge Nvidia's dominance.
Insights by Ground AI
15 Articles
15 Articles
Coverage Details
Total News Sources15
Leaning Left2Leaning Right0Center4Last UpdatedBias Distribution67% Center
Bias Distribution
- 67% of the sources are Center
67% Center
L 33%
C 67%
Factuality
To view factuality data please Upgrade to Premium
