As-Is Approximate Computing

Although approximate computing promises better performance for applications allowing marginal errors, dearth of hardware support and lack run-time accuracy guarantees makes it difficult to adopt. We present As-Is, an Anytime Speculative Interruptible System that takes program executes with time-proportional approximations. That is, version the output is generated early gradually refined over time, thus providing guarantee eventually reaching 100% accuracy. The novelty our As-Is architecture in its ability conceptually marry speculative computing. show how existing innovations architectures can be repurposed anytime, best-effort approximation, facilitating design efforts overheads needed support. provides a platform real-time constraints interactive users interrupt programs accept their current results as is. always guaranteed if more time spared. Our evaluations demonstrate favorable performance-accuracy tradeoffs range applications.