Is dark silicon useful? Harnessing the four horsemen of the coming dark silicon apocalypse

By Michael Bedford Taylor. DAC Design Automation Conference 2012, 2012

364 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Limits on fundamental limits to computation (2014, background): “Modern CPUs have not been able to use all their circuits at once, but this asymptotic eﬀect — termed the utilization wall [50] — will soon black out 99% of the chip, hence the term dark silicon and a reasoned reference to the apocalypse [50].”
Rethinking floating point for deep learning (2018, background): “Energy for the majority of work is thus lower than MAC/FMA (Section 8); increased area for increased energy efﬁciency is generally useful in the era of “dark silicon” [32], or conversion module instances can be rationed (limiting throughput) and/or clock gated.”
ASIC Clouds: Specializing the Datacenter (2016, background): “The second change is the rise of dark silicon [1, 2, 3, 4] and dark silicon aware design techniques [5, 6, 7, 8, 9, 10] such as specialization and near-threshold computation, each of which help overcome threshold scaling limitations that prevent the full utilization of transistors on a silicon die.”
Accelerometer: Understanding Acceleration Opportunities for Data Center Overheads at Hyperscale (2020, background): “With the end of Dennard scaling [45, 120], successive server generations running these microservices exhibit diminishing performance returns.”