Accelerating Distributed Digital Ising Machines
Organizations: Carnegie Mellon University · University of Pittsburgh · University of California, Santa Barbara
Abstract
Ising machines, dedicated accelerators built around the dynamics of Ising-like spin models, have emerged as a leading platform for combinatorial optimization and probabilistic sampling. The problems that motivate them require millions of spins, far beyond the on-chip capacity of any single device, so a practical Ising machine must distribute its computation across many devices. Prior work reports throughput at a single operating point, leaving the binding resource unidentified and platform crossovers measurable only by experiment; no calibrated cross-platform cost model of distributed Ising machines has been reported, nor have systematic techniques to accelerate their Markov chain Monte Carlo (MCMC) workloads. This work presents a calibrated cost model that decomposes digital Ising sweep time into compute, memory, synchronization, and inter-device communication terms, validated with 4.8% mean error across three GPU generations, FPGA and ASIC up to 1 million spins. Guided by the model, we address each term in turn and propose: (i) a data placement strategy, worth up to 4.2 speedup; (ii) multi-rate and bounded-staleness schedules that relax the barrier that Gibbs sampling imposes, cutting time-to-solution (TTS) by up to 3.16 on three benchmarks; and (iii) a phase-offset streaming dataflow that hides boundary exchange behind interior computation whenever a link satisfies a latency-transparency criterion, yielding up to 1.25/1.74/2.01 on an H200 pair, a ZCU106 pair, and a four-board XEM8320 mesh. The streaming dataflow is also implemented on a fabricated four-chip 28 nm ASIC sustaining 28.1 Gflips/s at 4.81 pJ/flip.
Figures & tables
| Symbol | Meaning and units |
|---|---|
| , , | Total spins, device count, and spins per device |
| , | Spin degree and average graph degree |
| , | Color counts of the coupling graph and interconnect |
| , | Parallel compute units (SMs or PEs) and threads per block |
| , | Waves per color step and order of (color, wave) slots |
| , | CG color steps and scheduled IT phases between synchronizations, respectively |
| Technique | Control | Metric | Range |
|---|---|---|---|
| Multi-rate | refresh every sweep | TTS 99 | – |
| Bounded stale | TTS 99 | – | |
| Streaming | same schedule in lock-step | – |
| Work | Platform | Gflips/s | nJ/flip | |
|---|---|---|---|---|
| Ours † | 2 H200 | 4.2M | 14.3 | 60.8 |
| 4 28 nm ASIC [ 34 ] | 27.6k | 28.1 | 0.00481 | |
| 4 ZCU106 | 32.8k | 0.697 | 15.5 | |
| 4 XEM8320 | 4k | 1.61 | 61.3 | |
| Yang ’19 [ 93 ] | V100 | N/R | 11.37 | 21.99 nJ ∗ |
| Chowdhury et al. ’25 [ 4 ] | 1 U250 | 4k | 185 | 50 pJ |
Distributed Optimization
Distributed optimization coordinates computation across multiple agents or machines while balancing convergence, communication, privacy, and reliability.
See all 101 papers