Intel's business strategy may be under threat from a technological problem which was described as 'perhaps unsolvable' at a technical conference in San Francisco this week.
The biggest problem in computing today is programming multi-core chips for general purpose computing applications, according to Chris Rowen, president and CEO of Tensilica, on the opening day of this week’s Globalpress Summit Conference in San Francisco.
“The challenge of writing software for programming general purpose computing applications is generally recognised in the scientific computing community as the biggest single unsolved, and perhaps unsolvable, computing problem,” said Rowen.
Asked by EW what that meant for the business strategies of Intel and AMD trying to get more performance by adding more cores to their x86 architecture chips, Rowen replied: “It means they will find the utilisation will be poor, until we find this hypothetical breakthrough. Until then, the value of the addition of a tenth, or eleventh or fifteenth processor core will be problematical.”
Recognising this bottleneck to their strategy which depends on adding performance by adding more x86 cores to Pentiums, Intel and Microsoft recently announced a partnership with academia to create two universal parallel computing research centres (UPCRC).
The centres will be set up at the University of California, Berkeley (UC Berkeley), and the University of Illinois at Urbana-Champaign (UIUC), with the companies committing a combined $20m to the centres over the next five years.
An additional $8m will come from UIUC, and UC Berkeley is said to have applied for $7m in funds from a state-supported programme to match industry grants.
Where multiple cores are used on one chip which is used for a specifically targeted application, then you can actually get a linear improvement in performance by adding cores.
Rowen pointed out that, by using Tensilica’s processor generators you can put on one chip cores which run fast alongside cores which run with low power, in order to maximise performance/power efficiencies.
That works for specific applications, but when it comes to programming multiple cores for general purpose computing applications, the addition of more and more cores becomes less and less effective in increasing performance.
Unless, of course, that breakthrough, which may be ‘unsolvable’, occurs.