Insights from a Google researcher

Cliff Young, a member of Google Brain Team, is the featured speaker for UWL’s Distinguished Lecture Series in Computer Science.

‘Google Brain Team’ member to give Computer Science lecture at UWL

Cliff Young, a member of the Google Brain team, will give a public lecture “Hints for Special-Purpose Computer Design” at 4 p.m. Friday, Nov. 10, in the Strzelczyk Great Hall, at UWL’s Cleary Alumni & Friends Center.

Young, who visits UWL as part of a Distinguished Lecture Series in Computer Science, will also give a technical symposium “In-Data Center Performance Analysis of a Tensor Processing Unit” at 11 a.m. Friday, Nov. 10.

Check in and refreshments will be 30 minutes prior to each lecture. Both events are free. More on parking below.

Public lecture

A new age of special-purpose computers is beginning. Both supply and demand are focusing computer architects and system designers to explore domain-specific architectures. The exponential supply of transistors, Moore’s Law, is at long last slowing down. At the same time, new applications (some driven by deep learning) are generating new demands for effective computation. Machines that focus on a single application offer opportunities to tailor hardware and software structures to algorithms, and perhaps surprisingly, also offer opportunities for the converse, adapting algorithms to build a better system overall. This talk takes its inspiration from Butler Lampson’s 1983 “Hints of Computer System Design” and focuses on guidelines, rules of thumb, and experiences relevant to these new kinds of machines.


With the ending of Moore’s Law, many computer architects believe that major improvements in cost-energy performance must now come from domain-specific hardware. The Tensor Processing Unit (TPU), deployed in Google datacenters in 2015, is a custom chip that accelerates deep neural networks (DNNs). We compare the TPU to contemporary server-class CPUs and GPUs deployed in the same datacenters. Our benchmark workload, written using the high-level TensorFlow framework, uses production DNN applications that represent 95 percent of our datacenters’ DNN demand. The TPU is an order of magnitude faster than contemporary CPUs and GPUs and its relative performance per Watt is even larger. The TPU’s deterministic execution model turns out to be a better match to the response-time requirement of the DNN applications than are the time-varying optimizations of CPUs and GPUs (caches, out-of-order execution, multithreading, multiprocessing, prefetching, …) that help average throughput more than guaranteed latency. The lack of such features also helps explain why despite having myriad arithmetic units and a big memory, the TPU is relatively small and low power.

About Cliff Young

Young is a member of the Google Brain team, which aims to develop deep learning technologies and deploy them throughout Google. He is one of the designers of Google’s Tensor Processing unit (TPU), which is used in production applications including Search, Maps, Photos and Translate. Before joining Google, Young worked at D.E. Shaw Research, building special-purpose super computers for molecular dynamics and Bell Labs. Research areas include Hardware and Architecture, Machine Intelligence and Software Systems.


Visitors can purchase a half ($3) or full-day ($5) permit from Parking Services, located at 605 17th St. North, in the parking ramp. Visitors can also use the new pay stations, any commuter lot, as well as the first level of the parking ramp, (pay-by-phone app also available with these stations.)

Parking information, parking map and link to purchase permits can be found online at