Amdahl's Law in the Multicore Era

Google Tech Talks
February 6, 2009

ABSTRACT

Over the last several decades computer architects have been phenomenally successful turning the transistor bounty provided by Moore's Law into chips with ever increasing single-threaded performance. During many of these successful years, however, many researchers paid scant attention to multiprocessor work. Now as vendors turn to multicore chips, researchers are reacting with more papers on multi-threaded systems. While this is good, we are concerned that further work on single-thread performance will be squashed.

To help understand future high-level trade-offs, we develop a corollary to Amdahl's Law for multicore chips [Hill & Marty, IEEE Computer 2008]. It models fixed chip resources for alternative designs that use symmetric cores, asymmetric cores, or dynamic techniques that allow cores to work together on sequential execution. Our results encourage multicore designers to view performance of the entire chip rather than focus on core efficiencies. Moreover, we observe that obtaining optimal multicore performance requires further research BOTH in extracting more parallelism and making sequential cores faster.

This talk is based on an HPCA 2008 keynote address.

Speaker: Mark D. Hill
Mark D. Hill (http://www.cs.wisc.edu/~markhill) is professor in both the computer sciences department and the electrical and computer engineering department at the University of Wisconsin--Madison, where he also co-leads the Wisconsin Multifacet (http://www.cs.wisc.edu/multifacet/) project with David Wood. His research interests include parallel computer system design, memory system design, computer simulation, and recently transactional memory. He earned a PhD from University of California, Berkeley. He is an ACM Fellow and a Fellow of the IEEE.

Category:

Science & Technology

Tags:

• google
• techtalks
• techtalk
• engedu
• talk
• talks
• googletechtalks
• education

[fmsf303]

If Amdahls low would be appliable to increase in parallelism it would take arround 74 years to get the number of cores being equal to the number of neurons in a human brain. 2^37 = 137438953472 number of cores

[hugojosestalin]

Politics is the poison for the masses! Take your conversation elsewhere comrade.

[randomscene09385]

hahah this is awesme my family creadted this law!!

[darkflib]

a core isn't like a neuron though... it is closer to a cortex subsystem in the brain and as such, the 74 year estimate by fmsf303 is probably a few decades out due to this many orders of magnitude error...

[OtagoMark]

@soylentgreenb He talked about numbers VS numbers... nothing "moore" lol...he wasn't argueing that a core equalled a neuron.

[violafinishda]

REAL Free energy technology exists!But the Oil companies want these technologies unknown to the masses,Go to LT-MAGNET-MOTORdotCOM and get the blueprints for a genuine magnet motor ,Be a part of the energy revolution!

[soylentgreenb]

@marcusklaas A neuron does much more than a transitor. It averages over a series of inputs and fires if it is above some treshhold. You'd need about a million transistors to replace one neuron. Digital transistors is massive overkill, it should be possible to build much simpler analog electronics that does the job well enough with far fewer components. Synapses form on an adhoc basis, this ought to be very hard to simulate compared to some static configuration of neurons and their synapses.

[soylentgreenb]

@fmsf303 You're way overestimating the capability of the brain if you think 2^37 cores are necessary.

A neuron in the human brain is capable of firing once every ~5 ms; a "clock frequency" if you will of about 200 Hz. A dedicated ASIC simulating a neuron would require only about a million transistors; it wouldn't be pipelined as a modern CPU and so operate at 100 MHz instead.

That's just a factor ~1000 more transistors than a modern CPU core for the same computational power as the brain.

[kalm77]

(PS. to my below message on the challenges of architecture evolution and the current OS monopoly)

The future of the OS market may crack open with luck. The world's future OS belongs to the best parallelization code writers on the planet. YOU guys can kick MS tail. Monopoly my a**. Cheers to the future.