Powers Unfiltered

 The New York Times reminded everyone this week what many of us already know — that pretty much all new improvements in processing speed will come from more cores.  In this article by John Markoff, Intel prepares a demonstration of a single chip with 80 cores.  (We’ve been hearing about this for many months, but apparently now it’s ready to show off.) 

The article repeats Intel’s claim that this is the first “Teraflop chip,” says that it will make custom video creation a snap, and generally paints a bright picture.

Only much later in the article do we reach some words of caution.

The shift toward systems with hundreds or even thousands of computing cores is both an opportunity and a potential crisis, computer scientists said, because no one has proved how to program such chips for many applications.

“If we can figure out how to program thousands of cores on a chip, the future looks rosy,” said David A. Patterson, a University of California, Berkeley computer scientist who is a co-author of one of the standard textbooks on microprocessor design. “If we can’t figure it out, then things look dark.”

Mr. Patterson is one of a group of Berkeley computer scientists who recently issued a challenge to the chip industry, demanding that companies like Intel begin designing processors with thousands of cores per chip.

In a white paper published last December, the scientists said that without a software breakthrough to take advantage of hundreds of cores, the industry, which is now pursuing a more incremental approach of increasing the number of cores on a computer chip, is likely to hit a wall of diminishing returns — where adding more cores does not offer a significant increase in performance.

This analysis is correct in my view, and is a major reason why we started Digipede almost four years ago.  Intel and AMD will no longer bail us out with higher clock speeds — if we want software to run faster and scale better, we just have to learn to deal with more processors (more cores on more chips in more boxes).

And let’s be clear — I’m not claiming Digipede is “the answer” to this software “crisis,” because lots of answers are needed; so far, no one multi-processing approach will address all application loads, and it is doubtful any single approach ever will.  But for a fairly broad class of CPU-intensive processes, we’ve taken our shot and put it out there for developers (it’s free — go now — I’ll wait). 

Right now, our software is best suited for combining single- and dual-core chips in single- or dual-CPU boxes (i.e., making grids out of lots of boxes), because that’s where the sweet spot is for commodity hardware today.  But as the market evolves, you’ll see us bring out more features to help you make use of more cores on more CPUs in each box, too.

Digipede Evangelista Kim Greenlee writes and speaks to developer organizations about concurrent development.  See this recent post for example.  The age of concurrency is here — let’s embrace it together!