1M P-Bit Probabilistic Computer Achieves 1 Trillion Gibbs Flips Per Second via Networked FPGAs
June 28, 2026
Networking FPGAs into a distributed Ising machine breaks the single-chip capacity ceiling, realizing a 1,000,000 p-bit probabilistic computer that performs Gibbs sampling at over one trillion flips per second with all coupling weights stored in local on-chip memory. Inter-device communication is reduced to 1-bit boundary exchanges, enabling scaling beyond any single device's memory bandwidth. This architecture targets combinatorial optimization and sampling workloads that are intractable on conventional hardware.
HOW THIS AFFECTS YOU
●
researcherWorth watching as a scalable FPGA-based approach to Ising machines that sidesteps memory bandwidth bottlenecks — relevant if you work on probabilistic inference, combinatorial optimization, or hardware-accelerated sampling.