Intel Labs – Future of Energy Efficiency Processors – Self-Tuning Performance

I learned more than I thought I would at Intel Developer Forum. There was a lot of excitement about the latest processors and news.com has thorough coverage of IDF.

IDF 2009: Intel plays to its strengths

by CNET News staff

At the annual developer forum, Intel shows off what it can do with silicon and what to look forward to from systems built around its chips.

Intel unveils system-on-a-chip for TVs

The CE4100 is designed to bring Internet content and services to digital TVs, DVD players, and advanced set-top boxes.
(Posted in Nanotech: The Circuits Blog by Brooke Crothers)
September 24, 2009 1:30 PM PDT

Intel's Maloney: Our business is do or die

Sean Maloney, a favorite to eventually become Intel's CEO, says there are good reasons the chipmaker is pushing back against Europe's antitrust charges.
(Posted in Nanotech: The Circuits Blog by Brooke Crothers)
September 24, 2009 10:26 AM PDT

With all the hype, I was filtering, looking for something really game changing.  Something that will change things to be more efficient.  I found it in a booth staffed by Intel Labs stuck in the back of the exhibit area where Shih-Lien Lu was demonstrating Self-Tuning Processors.

By modifying Vcc Voltage  and clock frequency, the processor can be set up for energy efficiency or performance.

How big?  21% more throughput or 37% less power!!!

And there is a middle ground of 5% better performance and 28% less power.

Here is the prototype board.

There must be a catch to why isn’t Intel shipping this concept already.

Because it requires a different mindset for the market and users.  The below diagram shows the Vcc Voltage and Temperature Fclk guardband typically existing for processors.  There is a margin of safety to insure Intel Processors reliability over a 7 year period.  Huh?  But, what if I don’t want seven years?  Welcome to the problem with enterprise computing.  Lowest common denominator type of thinking to reach the market masses means you get burdened with conservative designs.

What happens if you only wanted a guardband designed for a 3 year period?  You could in theory do what Intel Lab shows and have lower Vcc voltages with higher clock frequencies, but  this would require Intel marketing and finance to rethink how they price processors.  What is the value of a 7 to 3 year change in product reliability?

Why go through all this effort? 

1. Do you want a 21% performance improvement for the same power?


2. Do you want to save 37% processor power for the same performance?


3. Do you want 5% more performance for 28% less processor power?

Sound confusing.  Yes it will make customer procurement process complaints increase as they are handed a performance energy design envelope.

This is another example of the Flaw of Averages where people want a single number when in reality there is a distribution of performance.